What this company sells is a measurement of the internet. If you want to know how much traffic a competitor’s website pulls, where that traffic comes from, what people are searching for, or which apps are winning a category, this is the subscription you buy to find out. It takes the exhaust of billions of digital interactions and turns it into a product that marketing teams, strategy teams, and investors check before they make a decision. Several thousand companies pay for it, and the base skews toward enterprises that spend well into six figures a year. The asset underneath is a data panel and modeling pipeline built over nearly two decades, the kind of thing a competitor cannot stand up in a quarter, because the value compounds with history and coverage.

SMWB (Similarweb) | $4.38 | $383M Market Cap

The company is Similarweb (NYSE: SMWB), and that data asset is the whole point. The moat does not show up as a factory or a patent. It shows up as the dataset that marketing and strategy teams across the category quietly treat as the reference, built over roughly 18 years of panel and modeling work, and increasingly as a dataset that other companies want to license rather than rebuild, including for training AI models. About 6,128 customers subscribe. The base skews enterprise: 461 of them pay more than $100,000 a year and account for 64 percent of recurring revenue, with another large tier paying more than $25,000.

Founder Or Offer has run the company since 2007 and still owns about 6.8 percent of it, a stake worth far more than the modest salary he draws. The chairman, Harel Beit-On, is a co-founder of Viola Growth who once took Tecnomatix through a sale to a strategic buyer. The board around them is not filler: it includes the former chief executive of NICE, the chief marketing officer of HubSpot, and the former finance chiefs of Zillow and Gett. In May, the board announced an orderly CEO succession, with Offer transitioning out by mid-2027 while staying on through the handover. He has continued buying stock since.

The numbers, and the part the headline misses

Revenue was $282.6 million in 2025, up 13 percent, and $73.9 million in the first quarter of 2026, up 10 percent. The company guided full-year 2026 to between $307 and $315 million, roughly 10 percent growth. Gross margin sits around 80 percent, which is the economics of selling the same data to many customers at once. This is no longer a hypergrowth company. It is a steady grower with very good unit economics, and the market has priced the deceleration as if it were terminal.

Here is the part that takes reading the filing to see. On a GAAP basis the operating loss actually got wider last year, to $23.6 million from $9.7 million the year before. That looks like deterioration until you ask why. The company spent into its sales force, lifting sales and marketing expense 17 percent, and it absorbed a swing of roughly $5 million in non-operating currency effects tied to the Israeli shekel. Strip the non-cash and non-operating items and the picture inverts. First-quarter 2026 produced a $2.4 million adjusted operating profit, against an adjusted loss a year earlier, and management guided to $17 to $19 million of adjusted operating profit for the full year. Stock-based compensation, the usual place software losses hide, runs around 7 percent of revenue here and is falling, so the adjusted figure is not doing heavy lifting it has not earned.

The cash flow confirms it. The first quarter was the tenth consecutive quarter of positive normalized free cash flow. A company can argue about adjusted margins, but ten straight quarters of cash generation while reporting GAAP losses is the signature of a business carrying non-cash compensation and acquired-intangible amortization, not one burning money. The balance sheet has about $65 million of cash and no financial debt, and capital spending is roughly $1.5 million a year. There is no leverage to worry about and nothing forcing management’s hand.

Now the valuation. Similarweb trades at about 1.2 times trailing sales and a touch above one times forward sales. Its closest public peer, Semrush, was acquired by Adobe at roughly 3.7 times revenue, and traded near 3.3 times before the deal was struck. A strategic acquirer just told the market what the category is worth, and it was triple what Similarweb fetches. Apply a multiple that respects Similarweb’s slower growth and the gap is still wide. At two times forward sales the equity is worth about $7.80 a share, at two and a half times about $9.60, and at three times about $11.40, against roughly $4.38 today. Those are modeled estimates based on a multiple assumption, not price targets, and they exclude any acquisition premium.

Enterprise value to trailing sales. Similarweb against its closest peer before and at the Adobe takeout price.

What we found in the filings

The most important number is not in the income statement. It is in the disclosure on remaining performance obligations, the revenue that customers have already contracted but the company has not yet recognized. That figure reached $297.7 million at the end of the first quarter, up 18 percent from a year earlier. Revenue grew 10 percent. When the backlog grows almost twice as fast as reported revenue, the income statement is lagging the business, not leading it. Customers are committing ahead of what the P&L shows.

Year-over-year growth in the first quarter of 2026. The backlog is the leading indicator.

That contracted revenue is also getting more durable. The share of recurring revenue under multi-year contracts climbed to 64 percent from 52 percent a year earlier. Longer contracts mean more visibility and less churn risk, and they are the kind of structural change that does not reverse quickly. Management also noted that sales productivity improved for the third consecutive quarter and that the quarter produced the strongest first-quarter recurring-revenue addition since 2022. These are the fingerprints of a sales investment that is starting to pay back, which is exactly what you would want to see after a year in which that investment widened the reported loss.

Then there is the line that did not exist in the old Similarweb story. During the first quarter the company signed a seven-figure contract to license its data for training an existing big tech customer’s large language model, one of two such deals that slipped out of the fourth quarter, and management said it expects a second large data-training contract over the coming quarters. The “data asset in the age of AI” language that the chairman uses is no longer a slogan. It is a disclosed, dollar-denominated revenue line. Similarweb is selling a raw material the AI buildout needs, which is a very different and higher-value use of the same panel it already runs.

Page 2, Form 6-K

Finally, the people who read these filings most closely have been buying. Across roughly two weeks from mid-May into early June, four insiders bought about $1 million of stock in the open market: the founder-CEO three times, the chairman, and two directors, one of whom is the former chief executive of NICE. They were not buying from a standing start. Insiders already hold close to 30 percent of the company. The Form 4 filings show open-market purchases at prices from roughly $3.20 to $4.15, into a rising tape rather than a falling one.

What has to happen, and when

The simplest catalyst is the arithmetic continuing to surface. If the company delivers the $17 to $19 million of adjusted operating profit it guided to, and the 18 percent backlog growth keeps converting into reported revenue, the gap between a profitable, cash-generative business and a 1.2-times-sales multiple becomes harder for the market to hold. The second large data-training contract management flagged would be a clean, near-term confirmation that the AI-data line is a trend and not a single deal.

The other source of upside is structural, and it sits in the background rather than the base case. Adobe’s purchase of Semrush established a strategic clearing price for digital-intelligence data at close to four times sales. Similarweb has a chairman who has sold a company before, a strategic holder in Prosus and Naspers that owns nearly 13 percent, and a depressed multiple. None of that predicts a transaction, and the thesis does not need one. It does mean the same data asset is worth far more to a strategic owner than the public market currently assigns, and that is a useful floor to have under a stock. On the macro side, the AI wave that the market often treats as a threat to legacy software is, for this company, showing up first as a new customer for its data.

The risks

• Net revenue retention has fallen below 100 percent. Overall retention was 98 percent in the first quarter, down from 101 percent, and even the large-customer cohort slipped to 103 percent from 111 percent. Below 100 percent overall means the existing base is contracting slightly before new sales, so growth increasingly depends on winning new logos rather than expanding within current ones.

• The company is unprofitable on a GAAP basis and the loss widened in 2025. Operating loss was $23.6 million and net loss was $32.9 million last year, both wider than 2024. The profitability case rests on adjusted figures and cash flow, and GAAP breakeven has not arrived.

• Growth is moderate and still decelerating. Full-year guidance implies about 10 percent, and second-quarter guidance implies roughly 6 percent. The investment case depends on durability, margin expansion, and the data-licensing optionality, not on a high top-line rate.

• A founder CEO is on his way out. The transition is orderly and planned for mid-2027, but the successor is unknown, and leadership changes carry execution and culture risk at a company built around its founder.

• Ownership is concentrated. Prosus and Naspers hold about 12.8 percent, Viola about 12.4 percent, and a single individual director about 10.2 percent. Any of these holders deciding to sell into a thinly traded stock would pressure the price regardless of the fundamentals.

The bottom line

At about 1.2 times sales, the market is paying for a no-growth orphan. The filings describe something else: 80 percent gross margins, a contracted backlog growing 18 percent, an inflection to cash profitability, a new AI-data revenue line, and a founder and board buying the stock. We think the distance between those two descriptions is the opportunity, with the Adobe purchase of Semrush as a reminder of what the asset is worth to someone who wants the whole thing.

What we are watching: the second-quarter report this summer, where the two questions that matter are whether the second large data-training contract lands and whether net revenue retention stops falling. Those two data points will tell us whether the turn in the bookings is durable or a single good quarter.

Disclaimer: This is not investment advice. The author does not hold positions in securities discussed. All information is based on publicly available SEC filings, insider transaction records, and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.

The chair is not a passive board appointee. He is the former chief executive of a public building-products company that built itself on a single idea, converting customers from an old material to a better one. The chief executive he recently installed spent more than two decades running the same kind of conversion at another well-known building-products name. Two materials-conversion veterans, quietly buying a small-cap the market has written off, while a private-equity seller overhang caps the price and the operating numbers quietly inflect underneath.

The filings explain why they may be early rather than wrong. The name, the financials, the insider detail, and the valuation case are below.

Read more

The stock is up 197% over the last year. It is also 35% off the March 2026 peak. The market is processing this as a transition story that may have run its course. The filings describe a roll-up that just signed two more deals this quarter.

Top Shelf is one of two regular series at Thrift Shop Stocks. Here explains what Top Shelf covers and how it works, or visit the about page for a full breakdown of the shop.

Read more

What Caught My Eye

Two activist investors sit on the board of a 167-year-old industrial holding company. One runs a value-oriented activist fund and has been involved with the company since around 2013. The other runs a $10 billion investment firm and has been buying open-market stock for the past 18 months. On March 6, 2026, they both stepped into the open market on the same day and bought more shares within ten cents of the 52-week low.

The stock was sitting at a four-year low because the company’s largest customer base, heavy-duty truck OEMs, was finishing year four of what trade analysts had taken to calling a “generationally low” downcycle. The trailing financials show an 8.7% revenue decline, an operating margin that compressed from 7.4% to 4.3%, and a 54.5% drop in net income. Sell-side coverage is two analysts, both with hold ratings.

Inside the same filings, the company refinanced its debt out to 2030, paid its 342nd consecutive quarterly dividend, and absorbed a $7.7 million increase in China tariff costs that it largely passed through to customers. The screener says deteriorating microcap. The filings say something else.

The Business

The company was incorporated in 1912, succeeding a co-partnership established in October 1858. The original business was locks and security hardware. Through 167 years of cycles, two world wars, and one Great Depression, the holding company has paid an uninterrupted quarterly dividend for more than 85 years.

The current business is a single reportable segment selling custom and standard vehicular and industrial hardware to commercial transportation customers. The product line includes turnkey returnable transport packaging (the engineered crates and racks that ship engines, transmissions, and high-value parts between OEM plants), industrial latches, hinges, and access hardware for trucks, buses, and off-highway equipment, and truck mirror assemblies including a growing line of mirror-cameras. The customer base reads like a directory of the heavy-duty truck industry. Foreign sales are not material. This is a US-domiciled supplier to US-built commercial vehicles, which has become an unintentional hedge against the very tariff regime that has dominated the trade press for the last twelve months.

The CEO joined in November 2024 from a private engineered thermoplastics manufacturer. His background is Lean, Six Sigma, and continuous improvement, and his fingerprints are visible in everything the company has done since: divestitures, footprint optimization, restructuring charges, two new operating-head appointments, and the first year of a serious cost takedown. The board has been quietly reshaped over a decade by the activist chairman, who won two board seats through a proxy contest in 2015 and has been adding allies and operators ever since.

The trailing financials look like a deteriorating microcap. The filings show three things the screener doesn’t: a tariff hit roughly equal to full-year operating income that the company largely passed through to customers, a multi-year insider buying pattern still active at the 52-week low, and a new-product growth rate running 6.2% inside an 8.7% headline revenue decline.

The Company

The Eastern Company (EML) | $22.07 | $133M Market Cap | NASDAQ

The company is The Eastern Company, ticker EML. The activist chairman is James Mitarotonda, founder and CEO of Barington Capital Group. The activist director with the 18-month buying pattern is Frederick DiSanto, chairman and CEO of Ancora Holdings. The CEO is Ryan Schroeder, who joined in November 2024 from Plaskolite. The three operating subsidiaries inside the Engineered Solutions segment are Big 3 Precision (returnable transport packaging), Eberhard Manufacturing (industrial latches and access hardware), and Velvac (truck mirror assemblies and mirror-cameras).

The Numbers

Eastern’s reported FY2025 financials look bad on first read. Net sales of $249.0 million, down 8.7%. Net income from continuing operations of $6.0 million, down 54.5%. Operating margin compression from 7.4% to 4.3%. A reader skimming a screener would put this in the “deteriorating microcap” bucket and move on.

The 10-K shows something that changes the picture. Tariff costs on China-sourced products were $10.2 million in 2025 versus $2.5 million in 2024, an incremental $7.7 million headwind. Management notes in the MD&A that “most tariffs were recovered through price increases.” The company absorbed a tariff hit roughly equal to its full-year operating income, mostly passed it through to customers, and still printed positive net income, paid its dividend, and reduced debt by $4.7 million.

The balance sheet has improved over the year. Cash sits at $7.4 million, working capital at $71.7 million, and the current ratio at 3.7. Net debt is around $26 million against trailing EBITDA of $19.8 million, putting net leverage at 1.3x. On October 28, 2025, the company refinanced into a new $100 million five-year unsecured revolving credit facility with Citizens Bank, replacing the prior TD Bank facility. Maturity moved to October 2030, term loan amortization was eliminated, and the company added a $75 million accordion. As of fiscal year end, $66 million was available on the revolver.

At $22.07, EML trades at EV/Sales of 0.73x, EV/EBITDA of 13.4x trailing, and price-to-book of 1.07x. Broader industrial hardware peers under the same SIC code (HLMN, SSD, ACU) trade at EV/Sales of 1.0x to 3.3x. Closer operating peers in commercial vehicle hardware (SRI, MOD, LCII, SMP) sit in the 0.8x to 1.2x EV/Sales band. EML is at a discount to both groups. The dividend yield of roughly 2.0% is close to its 5-year average despite the cyclical low.

The most useful number in the data is not on the income statement. New product sales grew 6.2% in 2025. Sales of existing products declined 14.9%. The 8.7% headline decline is the net of those two. A company with no real product pipeline does not show that mix. The new mirror assemblies, rotary latches, D-rings, and mirror-cameras the MD&A names are the part of the business that grows when the truck cycle turns, and they are already growing now.

What We Found in the Filings

Three things in the filings carry the thesis.

The first is the insider buying pattern, which is materially larger and longer than a casual look at Form 4 alerts would suggest. Frederick DiSanto, the Ancora chairman who joined the EML board in April 2016, began buying open-market stock in November 2024 and has not stopped. Between May and August 2025 alone, he added roughly $200,000 worth of shares. As of his most recent Form 4 he holds 99,689 shares directly, plus 43,797 through Ancora Catalyst and 11,970 through Ancora Merlin. James Mitarotonda joined him on March 6, 2026, buying 5,067 shares at $18.29 through one of the Barington vehicles. Mitarotonda already owned roughly 11.4% of the company. Adding to a 700,000-share position is a different calculus than a token buy from someone who owns nothing.

The second is the operational cleanup that has just been completed. The company classified Big 3 Mold as held-for-sale in the third quarter of 2024, took a $23.1 million write-down to fair value, and reported the entire division as discontinued operations. On April 30, 2025, it sold the Centralia, Illinois injection-stretch-blow-molding division for cash and a $2.0 million gain. As of January 3, 2026, the remaining Big 3 Mold divisions were reclassified back to continuing operations. The accounting was unusually messy because the company changed its mind about which pieces to keep, but the underlying story is that management has surgically removed the loss-making piece, kept what the new operating heads thought they could fix, and refinanced the balance sheet around the new shape of the business.

The third is the cost takedown. Management completed roughly $4 million in annualized cost reductions during 2025, including a 49% headcount reduction and the consolidation of Big 3 Precision’s production into the Centralia facility. The 10-K records $2.5 million of restructuring charges within selling and administrative expenses. The Q4 2025 earnings release reports adjusted EBITDA improved sequentially in Q4 even as revenue declined, and management characterized November and December as “early stabilization” in truck OEM volumes.

What ties the three together is the alignment story buried in the proxy. Schroeder’s 2025 short-term incentive was structured around earnings per share and a sales increase target. The company missed both. The compensation committee paid him zero short-term incentive for the year. CFO Vlahos was structured the same way and also earned zero. The board did not override its own structure to pay management for showing up during a cyclical bottom. That is rare and it is the right read on what kind of board this is.

The Catalyst

The catalyst is not company-specific. It is the heavy-duty truck cycle.

Eastern’s largest products by revenue are returnable transport packaging and truck mirror assemblies, both of which track the OEM build rate for Class 8 trucks and trailers. The cycle that drove EML’s revenue from $272.8 million in 2024 to $249.0 million in 2025 was not idiosyncratic. ACT Research described the period as “Year 4 of generationally low carrier profits.” FTR Transportation Intelligence reported November 2025 Class 8 net orders of 20,200 units, down 44% year-over-year and well below the 10-year November average of 28,910 units.

That was the bottom. December 2025 Class 8 net orders surged to 42,200 units, up 108% month-over-month and 21% year-over-year, the highest single month since October 2022. February 2026 orders were 46,440 units, up 157% year-over-year, one of the strongest order months in the current cycle. ACT Research’s March 2026 outlook describes the trucking industry as “moving beyond the bottoming phase and now in the early stages of a supply-driven recovery.” Two factors drove the inflection: the EPA confirmed it would retain the technology portion of the 2027 NOx emissions rule, and Section 232 tariffs on heavy-vehicle imports were finalized at 25% on November 1, 2025. Both events resolved regulatory overhangs that had kept fleets on the sidelines.

The structural setup underneath supports a multi-year recovery rather than a one-quarter bounce. Driver supply is contracting at the fastest pace in several years. The on-highway tractor population continues to contract as sub-replacement build rates persist. McKinsey is forecasting H1 2026 sales declines of 10% to 25% year-over-year, with replacement-driven recovery beginning in H2 2026 as aging equipment is forced to turn over.

The vocational truck segment, which serves construction, utility, and data center buildout, has its own tailwind. The four largest US technology companies are projected to deploy roughly $650 billion in capital expenditure on data centers and AI infrastructure during 2026, with vocational vehicle demand as one operational consequence. EML’s Eberhard division supplies hardware into precisely this category.

The conviction question is whether Eastern’s customers start placing larger orders in early 2026. The Q4 2025 backlog of $81.1 million was down 9% year-over-year, but management called out “early stabilization” in November and December on the Q4 earnings call, and reported Asia revenue grew 25% in 2025 against the broader cyclical downturn. That commentary lines up exactly with the December order surge in the macro data.

Some background on the assumptions.

The discount rate of 8.6% is built specifically for Eastern, not borrowed from a generic small-cap industrial average. The cost of equity uses the current 10-year Treasury (4.42%), Damodaran's January 2026 implied US equity risk premium (4.23%), Eastern's own 0.87 beta, and a 1.5% size premium appropriate for a sub-$200M market cap. The cost of debt is 5.95% pre-tax, derived from the new Citizens Bank facility's stated SOFR-plus-spread terms at Eastern's current leverage tier, taxed at the 20.6% effective rate disclosed in the 10-K. Weighted at the actual 80/20 equity-to-debt market value capital structure, the result is 8.6%.

The Risks

• Revenue is still declining and the call for a cyclical bottom could be wrong. ACT Research and FTR have called several false bottoms in this cycle. McKinsey’s H1 2026 forecast remains 10% to 25% YoY declines. If the truck cycle does not inflect in 2026, EML’s revenue base remains compressed and the operating leverage from the cost takedown does not materialize.

• Tariff exposure is significant and dynamic. China-sourced product tariffs hit $10.2 million in 2025, up from $2.5 million in 2024. Management says most was recovered through price increases. If trade policy escalates further, the pass-through becomes harder to maintain, particularly with truck OEMs whose own costs are already under pressure from Section 232.

• Customer concentration is real. The 10-K discloses one customer exceeded 10% of accounts receivable in both fiscal 2025 and 2024. The customer is not named. A loss of that customer or a significant slowdown there would have an outsized impact.

• The activist board has not signaled a sale process. Mitarotonda has been involved with this company since 2013, has held two board seats since 2015, and is now independent chairman. He has not publicly called for a sale. Investors expecting a near-term strategic-alternatives announcement may be disappointed.

• Liquidity is thin. Average daily volume is roughly 16,000 shares. Position-sizing matters and the bid-ask can be wide on bad news days.

• The 2025 say-on-pay vote came in at 77% versus 97.5% in 2024. The drop was driven by criticism of the prior CEO’s $1.1 million severance package, not the current compensation structure, but governance ratings agencies have raised it as a concern. The detail is in the DEF 14A compensation discussion.

The Bottom Line

Eastern is a 167-year-old commercial vehicle hardware company with two activist investors on the board buying stock at a multi-year cyclical low after a clean operational restructuring. The business has cut $4 million of annual cost, divested its loss-making division, refinanced into a five-year credit facility, absorbed a $7.7 million tariff hit through pricing, and grown new-product sales 6.2% while existing-product sales fell 14.9%. The largest end market is bottoming. The valuation discount versus both broad industrial peers and closer commercial vehicle peers is real. Management missed every short-term incentive target in 2025 and earned zero bonus, which is what alignment looks like when it works.

What we’re watching: the Q1 2026 earnings release on May 12, 2026, after market close. The data points that matter are quarter-end backlog (currently $81.1 million, looking for stabilization or growth), commentary on order trends in February through April, and whether the new-product growth rate of 6.2% accelerates or holds.

Clearance Rack posts are paid. This one's free. Subscribe below for access to next week’s analysis.

Disclaimer: This is not investment advice. The author may hold positions in securities discussed. All information is based on publicly available SEC filings, insider transaction records, and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.

That trade is mostly dead. Academics have documented its disappearance over the last decade. The arbitrage capital that figured out how to absorb the June reconstitution has been doing it efficiently enough that the post-deletion price impact has shrunk to statistical noise. The S&P 500 deletion effect, sister to the Russell version, went from -16.6% in the 1990s to -0.6% from 2010 to 2020. The 2022 NBER paper documenting this was titled “The Disappearing Index Effect.” The obituary has been written.

This year, something genuinely new is happening, and it’s worth thinking about carefully.

What’s changing

On April 30, FTSE Russell takes its annual rank day snapshot. That’s the day market caps are measured for index eligibility. Same as every year. The preliminary list of additions and deletions publishes May 22. The June 26 effective date marks when index funds rebalance and the new constituents take effect. None of that is new.

What is new is that on October 30, FTSE Russell will take a second rank day snapshot. The second reconstitution becomes effective on December 11, with new index composition trading from December 14. Two reconstitutions per year, the first since the late 1980s when the Russell indexes were briefly semi-annual before settling into annual cadence in 1989.

The reason for the change is mechanical. The June reconstitution had grown to the point where over $200 billion in dollar volume traded in the closing minutes of the effective date as index funds rebalanced. That single trade was getting hard to absorb cleanly. Splitting it across two events smaller than the original one is the response.

Source: Russell US Indexes - Moving to a semi-annual index reconstitution frequency

There’s an asymmetry between the two events worth knowing. The June reconstitution will continue to do the full work, including complete style reconstitution across the Russell 1000 Growth, Russell 1000 Value, Russell 2000 Growth, and Russell 2000 Value indexes. The December event will only update style classifications for new additions and for stocks moving between Russell 1000 and Russell 2000. Existing constituents won’t get re-scored on Growth-Value characteristics until the following June. June stays the heavy event. December does targeted work.

What FTSE Russell is saying

FTSE Russell CEO Fiona Bassett described its analysis of the change in a Bloomberg interview. They ran simulations of turnover and transaction costs under the new structure. The results showed a modest decrease in June reconstitution turnover and a modest increase at the December event. Net total turnover across both events is broadly comparable to the historical single annual event, just split across two windows.

Source: Russell US Indexes - Moving to a semi-annual index reconstitution frequency

There’s a difference between a simulation and a live environment. A simulated environment can perform across decades of historical data and behave differently the first time it runs in a live market. Simulations capture what was modeled. They can’t capture how arbitrage capital actually behaves in an event that’s never been live before.

Why a smaller event might restore an old edge

The natural read on a smaller event is that it will be easier to absorb, not harder. Less volume per event, more capacity from existing arbitrage infrastructure, less price dislocation. If the deletion bounce was already mostly arbitraged away under one big event, a smaller event should be even more efficient. FTSE Russell’s simulations support that read.

That intuition is incomplete. Arbitrage capital prices uncertainty, not just volume. The June reconstitution has 37 years of price history. Every desk that runs a deletion bounce or an inclusion squeeze knows the playbook. The trade is well-rehearsed and the spreads are tight because the math is well-understood.

The December reconstitution has zero years of price history. The microstructure is unknown in two specific ways, neither of which appears in a turnover simulation.

The new December Russell reconstitution happens during the same period as year-end tax-loss selling, institutional window dressing, and lower holiday liquidity. The same event, but very different conditions.

The first is timing. Tax-loss selling intensifies in November and December as investors offset gains with losses. The names being deleted from the Russell 2000 are by definition stocks that have shrunk over the prior six months, which are the same candidates eligible for tax-loss harvesting. Index-driven forced selling stacks on top of tax-loss-driven discretionary selling on the same names. Then there’s window dressing pressure as institutions clean up books for year-end reporting, lower year-end liquidity through the mid-December trading window, and historical small-cap January effect dynamics that have been attributed in part to tax-loss selling pressure releasing in early January. None of that applies to June.

The second is arbitrage capital allocation. Funds that previously sized their exposure for one annual event now have to decide how to split capital across two. The straightforward path is to double capacity. The realistic path is something between doubling and splitting, with the actual allocation pattern only becoming clear after the first December event prints. Whatever desks decide in 2026 may not be what they decide in 2027 or 2028. The early years of the new regime are when arbitrage coverage is most likely to be incomplete.

Stack them together and the result is genuine uncertainty about how the December event will price. Standard asset pricing predicts that uncertainty commands a risk premium. Liquidity providers absorbing flows around an unknown event require more compensation than those absorbing flows around a known one. That risk premium is the mechanism by which structural uncertainty becomes mispricing. Not because the market has failed to notice, but because the market has noticed and is demanding more compensation to take the position.

This list is not exhaustive. There are likely many more factors that could compound or even mitigate the effect. If you have any thoughts on what they could be, let me know in the comments. I am interested to hear others’ thoughts on this.

The June 2026 event is also worth watching

The natural framing is that June is the boring control variable and December is the interesting one. That framing is not entirely true, or at least incomplete.

June 2026 is the last June reconstitution operating under the old mechanical structure. It has the same 12-month accumulation window every prior June has had. There’s no mechanical reason for June 2026 turnover to differ materially from June 2025 or from the historical June baseline. FTSE Russell’s simulations project a modest decrease, but that projection captures mechanical effects, not behavioral ones.

If June 2026 turnover, transaction costs, or price impact differs meaningfully beyond what mechanical projections predict, the variance is behavioral. Market participants positioning around June 2026 with knowledge that another reconstitution is five months away. Funds shortening deletion-bounce holding periods because they need capital available for December. Holders splitting their dump across both events. Recovery strategies delaying re-entry into June deletions if some names face structural reasons to be re-deleted in December. None of this happened in 2025 or earlier. All of it could happen in 2026.

Which means June 2026 is the cleanest test of the anticipation effect available. Same mechanical setup as every prior June. Only difference is the calendar ahead. Variance from the mechanical baseline is evidence that the regime change is already rippling through investor behavior in advance of the structural event.

The honest framing has three layers. The historical baseline (every June through 2025) is the longest data series. The June 2026 event is the cleanest anticipation test. The December 2026 event is the genuinely novel structural event with both known mechanical differences and unknown behavioral dynamics.

Where to pay attention

Three filters separate a deletion candidate worth examining from one that should be ignored. First, filing evidence supporting value above where the forced selling will leave the price: insider buying, cash on the balance sheet relative to market cap, recent footnote changes suggesting improving fundamentals. Second, the absence of a permanent impairment story: going concern qualifications, covenant violations, customer losses that have already materialized. The deletion isn’t creating an opportunity if the company is genuinely dying. Third, structural overlay: deletions that stack on top of another forced-selling event such as a recent spin-off, a post-bankruptcy emergence, or an activist situation where the prior shareholder base no longer fits.

The names that meet all three filters are rare. Most Russell 2000 deletion candidates fail one or more. But the ones that pass are exactly the kind of structural mispricing this publication exists to find.

As already mentioned, this edge has been eroded in years gone by. But now, it might be worth keeping ready in the toolkit for what might unfold.

What we’re watching

The May 22 preliminary list is the next observable event. Between then and the June 26 effective date, the deletion candidates will be visible and the forced selling will play out. That five-week window is where the work happens for the June event. October’s rank day and the December 11 effective date are the genuinely novel piece. The December event is the first opportunity to observe whether the structural uncertainty produces the kind of dislocation that’s been arbitraged out of the June trade.

One way or another, twelve months from now there will be measurable results. Academic working papers will analyze whether the December event produced detectable changes in deletion price impact, in arbitrage capital allocation, in the relationship between the two reconstitutions. The new structure could prove more efficient than the old. It could also prove less. Until it plays out, it’s an unknown. Which means it’s both an execution risk and a potential opportunity.

This publication will be tracking the June event as an anticipation test against the historical baseline, the December event as the first observation of the new structure, and the comparison between them as the most informative natural experiment on index microstructure available in 2026. A watchlist post applying the framework above to specific deletion candidates will follow once the preliminary list is available enough to work with.

The deletion bounce might be dead. It might also be the kind of thing that gets reported as dead right before the conditions that killed it change. The 2026 cycle is when we find out.

Disclaimer: This is not investment advice. All information is based on publicly available sources. Do your own research before making any investment decisions.

Wall Street has decided AI is going to kill business process outsourcing, the industry built on running other companies’ customer service, tech support, and back-office operations. The entire sector has been re-rated downward for 24 months on that assumption. And if you want proof of exactly how cold the market has turned on these names, the best valuation work on a small-cap BPO right now isn’t in that company’s own filings. It’s sitting in a peer’s merger proxy, paid for by somebody else.

We aren’t here to talk about TaskUs today. We’re here because of what’s buried in their merger proxy: a peer valuation framework that names a $370M BPO the market has left in the dust. On May 8, 2025, Blackstone and the two founders of TaskUs offered to take TaskUs private at $16.50 per share. Because Blackstone and the founders sat on both sides of the deal, they had a conflict, so the independent directors on the TaskUs board formed a special committee and hired Evercore, an independent investment bank, to run a fairness opinion. A fairness opinion is a formal analysis that asks a single question on behalf of the minority shareholders: is this offer price fair? To answer it, Evercore had to establish what comparable public BPO companies were worth. In the definitive merger proxy, Evercore named four public BPO companies as its peer comparables, three large-cap names you’d recognize, and one small cap company. Evercore determined the peer group was trading at roughly 5.0 to 6.0 times forward adjusted EBITDA. That’s the market’s current price for a profitable BPO. On October 8, 2025, unaffiliated TaskUs shareholders voted down the $16.50 deal as too low, and the merger was terminated.

So we now have two benchmarks filed with the SEC: a fairness opinion built around 5.0 to 6.0x, and a shareholder vote saying that range was still too low. The $370M company on Evercore’s peer list trades at 4.9x trailing adjusted EBITDA and 4.4x forward. Below the floor of the range the TaskUs special committee’s own advisor used as the BPO peer benchmark. Below the multiple implied by the TaskUs deal price that unaffiliated shareholders rejected as inadequate. And it’s doing that while growing revenue 16.7% with expanding margins. The filings are telling a different story than the stock price.

The valuation came on the house. The mispricing is still there.

This is a Top Shelf post. Top Shelf stock reports are for paid subscribers. Subscribe now to be grandfathered in at our inaugural price and read the full breakdown: the ticker, how Evercore's own peer analysis values this company, the management compensation triggers about to trip, the new market entry and build-out, the capital allocation story, and the full valuation ladder.

Read more

All prices below are split-adjusted for the 3-for-1 stock split effective April 1, 2026.

The Broken Company

Powell Industries makes custom-engineered electrical equipment. Switchgear, power control rooms, electrical houses. The stuff that goes inside industrial facilities to distribute and manage high-voltage power. Founded in Houston in 1947, the family name still on the building, the founder’s son still owns over 10% of the company.

By 2020, nobody was paying attention. Powell had been broken for years. Fiscal 2017 revenue collapsed 30% as the oil price crash gutted the company’s core market. EBITDA went negative. Then five years of muddle: from FY2018 through FY2022, revenue oscillated between $449M and $533M, EBITDA margins ground out in the low single digits, the share price drifted between $7 and $15. Anyone who bought Powell in mid-2018 was sitting on a loss four years later.

In mid-2020, the enterprise value was $104 million. COVID had shut down project sites and customers were requesting schedule delays. Revenue was running around $500 million annually and declining. The CEO was talking about “navigating through this current cycle.” Three sell-side analysts covered the stock, total. Average daily volume was thin. Powell competed against ABB, Eaton, Schneider, and Siemens, multinationals with resources that dwarfed the entire company.

The stock had touched a COVID low of $5.21 in March 2020. On the surface, this was a small, broken cyclical industrial in the worst part of the cycle, run by competent but quiet operators who didn’t court Wall Street attention. The market was right about everything visible on the surface.

What the Filings Showed

Powell’s second quarter fiscal 2020 ended March 31, 2020, right as COVID was shutting down the global economy. The Q2 10-Q, filed May 6, 2020, two weeks after WTI crude had closed at negative $37 per barrel for the first time in history. Backlog had hit a record $566.4 million. Up 33% from three months earlier. Up 43% year-over-year. New orders in the quarter were $300.7 million, a 53% increase. Record backlog. During a pandemic. In the worst oil market ever recorded.

Source: Powell Industries 10-Q, filed May 6, 2020.

A small-cap industrial in the worst quarter of the worst macro environment in a generation just printed its highest-ever order book. Companies about to die don't set order records during a pandemic.

The same filing disclosed why. The company had been awarded a “substantial contract in January 2020”, “to support the integrated electrical distribution requirements for a large industrial complex being built in the United States. The project would take over three years to design, manufacture, integrate, test and ship.”

One contract, awarded weeks before the world shut down, committing three years of work at a company the market valued at $104 million. The fiscal 2022 10-K would later confirm this single project accounted for approximately 20% of consolidated revenues in fiscal 2021 and 15% in fiscal 2022. One contract, quietly anchoring two years of revenue at a company the market had written off as a cyclical loser.

Do the math the market wasn’t doing. A backlog of $566 million was 5.4 times the enterprise value. Powell's order book was worth more than Powell was.

Three structural facts sat alongside the backlog number, visible to anyone reading the same filing.

First, the segment data. Traction market revenue had increased 131% year-over-year. Utility markets were up 19%. Oil and gas was the headline, but new end markets were already growing fast enough to show up in the segment lines. Powell’s diversification into the markets that would later define its run, electric utility, light rail, commercial industrial, was already in the numbers. Most readers wouldn’t think to ask which segments were growing while the headline business was shrinking.

Second, capacity. Powell had built out its Houston manufacturing footprint across the prior decade and was operating well below the envelope those facilities could support. Whatever revenue run rate the backlog eventually produced, the physical capacity to manufacture it already existed. No new capex cycle, no execution risk from fresh construction, no dilution to fund expansion.

Third, the cost structure. Powell’s operating model carried high fixed costs and a wide gross-to-EBITDA spread. In the low-volume years this showed up as margin compression. In a higher-volume environment, the same cost base would produce operating leverage on every incremental dollar of revenue.

A clean balance sheet and a stable share count meant shareholders would keep whatever upside followed. $120 million cash against essentially zero debt, no year-end debt balance above $7 million across the prior decade. Shares outstanding moved from 34.5 million in 2018 to 36.3 million in 2025. Roughly 5% dilution across seven years. No equity issuance during the entire run. Most companies that emerge from a downturn issue equity at the bottom or pay down debt with stock. Powell did neither.

Three signals, but not three simultaneous predictions. The backlog was the live signal: demand already in hand, growing ahead of revenue, anchored by a multi-year contract. Capacity and segment mix were latent. Capacity had been built in earlier cycles and sat waiting. The segment shift was visible in the early form, traction and utility growing off small bases, but wouldn’t become the dominant mix driver for another three years. Three different proofs of readiness, assembled across different periods. The mechanism underneath all of it was the fixed cost structure that would eventually convert volume into returns.

The Conviction Test

Powell stayed quiet through fiscal 2021 and 2022. Revenue actually declined 9.2% in FY2021, one more disappointment for anyone holding. EBITDA margins compressed to 2.4%. Backlog dropped from the $566M peak down to $416M by the end of calendar 2021. The stock spent that stretch between $7 and $15.

This is what testing conviction looks like. The thesis was confirmed in every quarterly filing: the large project was converting on schedule, segment diversification was continuing, balance sheet held. By the end of fiscal 2021, traction revenue had grown another 40%, the substantial contract was on schedule, and cash had built to $134M with debt still at zero. None of it showed up in the price. Most investors who bought the original thesis would have been out long before the catalyst arrived.

Then orders started piling up again. By September 30, 2022, two months before ChatGPT launched, backlog had quietly rebuilt to a new all-time record of $592M. Calendar Q3 had been the strongest order quarter in two years and there was no external catalyst to point to. The next print told the same story: in the quarter ending December 2022, deferred revenue surged 171% year-over-year. The order book was inflecting. Six months later, revenue caught up. From the May 2020 entry through that point, 30 months of nothing in the price while the structural readiness compounded.

The Catalyst

On November 30, 2022, OpenAI launched ChatGPT. Powell closed that day at $8.77. Six months later, NVIDIA’s earnings call ignited the broader AI trade and repriced every company that touched data center infrastructure.

Powell was already inflecting in 2020. The backlog was growing faster than revenue. The factories had room to scale without new capex. The balance sheet was clean. When the AI buildout arrived, Powell's capabilities matched the new demand and the inflection that had been visible in the filings for years finally reached the income statement. Revenue grew from $533 million in FY2022 to $1.1 billion in FY2025. By Q1 fiscal 2026, new orders hit $439 million in a single quarter with a 1.7x book-to-bill ratio. Backlog reached $1.6 billion. The company secured its first individual data center megaproject worth more than $75 million.

Operating margins expanded from 1.3% in FY2022 to 19.7% in FY2025.

Cash and short-term investments grew from $117 million in September 2022 to $476 million in September 2025. Four times the cash balance in three years. Powell paid a small quarterly dividend, never bought back significant stock, and spent $13 million annually on capex. The cash piled up because the business was a machine.

The Honest Caveat

Insider selling exploded as the stock ran. From two transactions worth $0.3 million in 2022, insider sales jumped to $42.9 million across 124 transactions and eleven different sellers in 2024, then another $47.6 million in the first quarter of 2026 alone. Founder's son Thomas W. Powell sold $25 million of stock in a single day on March 19, 2026, at $502.23 per share pre-split. CEO Brett Cope followed with roughly $1 million on April 9. Both sales landed while the stock was trading at or near its all-time high.

Heavy insider selling near a peak is meaningful. Two things keep this on the right side of the line that separates durable winners from round-trippers: there has been no equity issuance, and the most recent earnings print showed revenue still growing, not decelerating. The pattern looks closer to disciplined diversification by long-term holders than to insiders heading for the door before deterioration. Worth watching, not yet alarming.

The Lesson

When a company has demand already in hand, physical capacity waiting to absorb it, and a segment mix drifting toward higher margins, with a fixed cost structure underneath all of it, the arithmetic is loaded before the catalyst arrives. Each force expands margins on its own. Together they compound.

From $7.62 the day before the May 2020 filing to a peak of $197.54 in February 2026, Powell ran 26x, not because the AI catalyst was uniquely large, but because every piece of the structure was already in place. The backlog signal fired in 2020. The capacity had been built in the decade before. The segment shift was visible in early form in 2020 and took another three years to become the dominant driver. None of them predicted AI. Each one, independently, was a proof that the business was ready for whatever demand eventually arrived. Revenue doubled. EBITDA grew fourteen times. That ratio is the signature of a company whose fixed cost structure, capacity runway, and segment mix were all primed to amplify whatever demand showed up. The filings told you what the business could do if volume showed up. Volume showed up.

The backlog disclosure sat in the back of a 10-Q that one or two analysts read. The number was public. It was up 43% year-over-year during a pandemic. It was 5.4 times the company’s enterprise value. The contract that anchored it was disclosed in plain English. The segment data showing the mix shift was on the next page. The balance sheet showing zero debt was right behind it.

Nobody connected the dots because Powell was a small switchgear company during an oil crash, and that’s not the kind of story that attracts attention. The filings didn’t care about the story. They just reported the orders.

The market took 2.5 years to catch up. That gap is where the work pays.

This case study is the first of several. The companies in the series share a handful of traits that kept showing up in the filings years before the market noticed. Different catalysts, different industries, same arithmetic.

The forward-looking version of this work is in the paid tier, covering companies that are exhibiting these traits now. If you want to watch the next one play out in real time, subscribe below.

Disclaimer: This is not investment advice. The author may hold positions in securities discussed. All information is based on publicly available SEC filings and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.

That’s a mistake, because backlog tells you what revenue looks like before it shows up on the income statement. It’s committed demand. Signed contracts. Money a customer has agreed to pay that hasn’t been recognized yet. If you’re trying to figure out where a company’s revenue is heading over the next 12 to 24 months, the backlog disclosure is the closest thing to a cheat sheet you’ll find in a public filing.

Where to find it

Not every company discloses backlog. It’s most common in industrials, defense, engineering and construction, specialty manufacturing, and enterprise software. Companies with long-cycle contracts, where the customer commits to a purchase months or years before delivery, tend to disclose it. Companies with transactional revenue, where customers buy on demand, usually don’t.

When a company does disclose backlog, it typically shows up in one of two places in the 10-K. The first is Item 1, the business description, usually under a subsection called “Order Backlog” or “Contract Backlog.” The second is the MD&A, where management discusses changes in backlog as part of the revenue outlook. Some companies report it in both places. You can search any company’s 10-K on EDGAR here.

The typical disclosure is a single paragraph. It states the dollar value of backlog at the end of the fiscal year, the dollar value at the end of the prior year, and sometimes a breakdown by segment or expected conversion timeline. It looks dry. It is dry. But the numbers in that paragraph can tell you more about the company’s trajectory than anything on the earnings call.

Why it matters

Revenue is a trailing indicator. It tells you what already happened. Backlog is a leading indicator. It tells you what’s contracted to happen.

When backlog is growing roughly in line with revenue, the company is replacing orders at the same rate it’s fulfilling them. Steady state. Nothing to see. When backlog is growing significantly faster than revenue, something has changed. The company is booking new orders faster than it can deliver on existing ones. Demand is outpacing capacity. That gap between committed future revenue and current recognized revenue is where the opportunity lives.

Think about what that gap implies. If a company did $200M in revenue last year and its backlog grew from $300M to $500M, the order book expanded by 67% while revenue only grew by whatever it grew. Customers are committing money at a rate the company can’t yet convert to revenue. That backlog will convert. The question is when, and whether the market has priced it in.

The framework

When a company’s customers are committing money faster than it can collect revenue, the filings tell you before the stock price does.

Here’s a simple way to use backlog disclosures in your own research. Pull two numbers from the 10-K: the year-over-year growth in backlog and the year-over-year growth in revenue. Divide backlog growth by revenue growth.

If that ratio is around 1x, the company is in steady state. Backlog and revenue are moving together. No signal.

If the ratio is between roughly 1.5x and 3x, the company is accumulating committed demand faster than it’s recognizing revenue. This is the range where it’s worth digging deeper. Something is driving accelerating orders. Maybe it’s a new product line. Maybe it’s a macro tailwind hitting the company’s end markets. Maybe it’s a large customer committing to multi-year purchases. Whatever the cause, the filing is telling you that revenue growth is likely to accelerate in future quarters as that backlog converts.

Above 3x, you need to ask why. Is demand genuinely surging, or did the company take a large one-time order that inflates the ratio? Is there a capacity constraint preventing revenue conversion? Is the backlog quality deteriorating, meaning orders are being booked but at risk of cancellation? Very high ratios deserve scrutiny, not automatic excitement.

Negative revenue growth with positive backlog growth is a particularly interesting signal. It means the company’s current revenue is declining, which is why the stock is probably cheap, but new orders are growing, which means the decline may be about to reverse. The market prices what it sees on the income statement. The backlog tells you what’s coming next.

Here’s what this looks like with real data. Sterling Infrastructure (STRL) is a $2.5B revenue engineering and construction company. Over four fiscal years, the relationship between their revenue growth and backlog growth tells a clear story.

In FY 2022, revenue was outpacing backlog. The company was consuming its order book faster than replenishing it. No signal. By FY 2023, the backlog flipped to 47% growth against 11.5% revenue growth, a 4x ratio. Customers were committing money faster than Sterling could convert it. FY 2024 shows a backlog decline, but read carefully: Sterling deconsolidated a joint venture that year, which removed its backlog from the consolidated figure. The underlying demand was still growing. Then FY 2025: 78% backlog growth against 18% revenue growth. The signal went from interesting to undeniable.

That FY 2024 dip is itself a lesson. Always check what changed in how the company defines or reports backlog before assuming a decline is real. One paragraph in the 10-K explained the entire discrepancy.

What the market misses

Screeners don’t capture backlog. Bloomberg terminals can pull it for companies that report it in structured data, but most retail investors never see it. It doesn’t show up in Yahoo Finance. It doesn’t show up in most stock screener apps. It lives in the prose section of a 10-K that most people skip.

This is exactly the kind of information asymmetry that creates opportunities in small-cap investing. A $200M industrial company with three analyst coverage and a backlog growing at 2x revenue growth is not on anyone’s radar. The analysts covering it might mention the backlog in passing. The institutions that own it are probably passive index funds that don’t read the 10-K at all. The company itself isn’t going to issue a press release saying “our backlog implies our revenue is about to accelerate.” They just file the 10-K, disclose the number in a paragraph on page 14, and move on.

What backlog doesn’t tell you

A few important caveats. Backlog is not guaranteed revenue. Contracts can be canceled. Government contracts in particular often have termination-for-convenience clauses. Always check whether the company discloses cancellation rates or at-risk backlog.

Backlog conversion timelines matter. A $500M backlog that converts over five years is a different animal than one that converts in 12 months. Some companies break this out. Many don’t. When they don’t, compare the total backlog to annual revenue to get a rough sense of how many years of work are in the pipeline.

Backlog composition matters too. If 60% of the backlog is from a single customer, that’s concentration risk even if the total number looks strong. The 10-K’s customer concentration disclosures, usually in the risk factors section, should be cross-referenced against the backlog figure.

And not every industry uses backlog the same way. Defense contractors report funded versus unfunded backlog, which is a distinction that changes the interpretation significantly. Construction and engineering companies sometimes include awarded-but-not-yet-contracted work, which is softer than signed contracts. Know what the company is including in its backlog definition before you compare it to peers or to its own history.

Always make sure to check the backlogs.

Especially for industrials, defense, engineering and construction, specialty manufacturing, and enterprise software.

A practical exercise

Pick a company you already follow that operates in industrials, construction, defense, or enterprise software. Pull the last two 10-K filings from EDGAR. Search for “backlog” in both documents. Write down the dollar value for each year. Calculate the growth rate. Then compare it to the revenue growth rate over the same period.

If backlog is growing meaningfully faster than revenue, read the surrounding paragraphs carefully. Management usually explains what’s driving the growth, even if they bury the explanation in bland language. A sentence like “the increase in backlog was primarily attributable to increased demand in our power infrastructure segment” might not sound exciting. But if that segment is 15% of revenue and its backlog just doubled, you’re looking at a potential growth inflection that the market hasn’t caught up to yet.

That’s the work. It’s not glamorous. Nobody is going to make a TikTok about reading backlog disclosures in a 10-K. But it’s the kind of edge that compounds over time, because the information is public and nobody else is reading it.

Here’s what it looks like in practice. This is from page 6 of a real 10-K, filed earlier this year.

Source: Sterling Infrastructure (STRL) 10-K, filed February 2026.

Backlog grew 78% year-over-year, from $1.69B to $3.01B. Contracts convert in 6 to 36 months. That’s committed future revenue nearly doubling while the market was focused on trailing earnings.

And the $3.01B figure excludes another $300M in projects where Sterling is the apparent low bidder but the contract hasn’t been formally signed yet. That’s the kind of detail you only find if you read the full paragraph, not just the headline number.

Thrift Shop Stocks digs through the filings that nobody else wants to. If you found this useful, subscribe to get our stock analysis reports where we put this kind of work into practice on specific companies.

Disclaimer: This is educational content, not investment advice. Do your own research before making any investment decisions.

Top Shelf is a series that covers quality compounding businesses where SEC filing evidence reveals accelerating fundamentals the market hasn’t priced. No forced-sale requirement. No special situation needed. Just companies that are getting better faster than their stock price reflects. The full breakdown of the shop lives on the about page.

A $42 Million Question

Strip out the $114.6 million in net cash on CS Disco’s balance sheet and the market is valuing the entire operating business at $42 million. The software platform, 1,549 customers, the AI products, the 98% net retention rate, $157 million in annual revenue. All of it, forty-two million dollars. The company’s own adjusted EBITDA improved by $8.5 million last year. At this valuation, the market is pricing two years of improvement as the total value of the business.

Meanwhile, Relativity, the dominant legal document review platform, just filed its S-1 targeting a $4 billion valuation. Disco is the only other pure-play public ediscovery company. One of these valuations is very wrong.

What CS Disco Does

Every major lawsuit in America generates a mountain of electronic data that both sides need to search, review, and produce. Emails, Slack messages, documents, databases. This process, called ediscovery, used to require armies of contract lawyers billing $50 an hour to read documents one by one. Disco built a cloud-native platform that automates most of this with AI. Their newest product, Auto Review, processes 32,000 documents per hour at 90%+ precision. A human reviewer handles about 50. The economics of that gap explain why the industry is moving to software.

Customers pay based on data volume processed and stored. Ninety-one percent of revenue is usage-based, which means no long-term contracts propping up the numbers. When customers use Disco more, revenue grows. When they use it less, it shrinks. The 8.3% revenue growth in FY2025, with software revenue up 11.5%, is demand-driven and verifiable. Disco was named a Leader in the IDC MarketScape: Worldwide End-to-End eDiscovery Software 2025 assessment.

The new CEO, Eric Friedrichsen, joined in April 2024 after running Emburse (expense management SaaS) for four years and holding executive roles at Adobe, SAP, and Concur. He replaced founder Kiwi Camara, who stepped down in September 2023. A new CFO, Aaron Barfoot, joined in January 2026. The entire C-suite has turned over in under two years. This is a management reset, and the early signs suggest the new team is focused on the right things: larger enterprise customers, simplified pricing, and AI product adoption.

The Numbers

Revenue came in at $156.9 million for FY2025, up 8.3% from $144.8 million the prior year. Software revenue, the higher-margin component, grew 11.5%. Dollar-based net retention improved from 96% to 98%. The balance sheet holds $114.6 million in net cash with zero debt.

A note on the margin improvement. The raw GAAP numbers show a +1,193 basis point swing in operating margin. That overstates the real improvement because FY2024 included a $15.2 million Fastcase intangible asset impairment (a write-off from a prior acquisition) and FY2025 included a $9.2 million stockholder litigation charge. Strip both one-time items and the adjusted operating loss improved from $45.7 million to $38.9 million, a +675 basis point margin expansion. Still significant, and it represents genuine operating leverage: the company added $12.1 million in revenue while cutting its adjusted operating loss by $6.8 million.

The margin improvement is coming from the right place. Non-GAAP sales and marketing expense fell from 39% of revenue to 35%, while R&D held steady at 31% and non-GAAP gross margin ticked up from 75% to 76%. The new management team is selling more efficiently, not starving the product. Q4 adjusted EBITDA came in at -$2.2 million, a -5% margin, roughly half the -12% a year ago. The Q4 2026 breakeven target requires closing a 5 percentage point gap, which at the current rate of improvement is credible.

The valuation gap is stark. Disco trades at 0.78x EV/Sales. The peer median for comparable public legal tech and SaaS companies (ONTF, INSE, ZENA) is 1.91x. Relativity, in its S-1 filing, is targeting roughly 4x revenue. Everlaw, the other major private competitor, was valued at $2 billion in its 2023 Series D, implying 6-8x revenue. Disco sits at a 59% discount to its closest public peers and an 80%+ discount to the private market comparables.

The balance sheet removes the most common bear case for unprofitable SaaS companies. $114.6 million in net cash, no debt. Even at the elevated FY2025 cash burn rate (which includes the litigation charge), that is nearly three years of runway on a GAAP basis and closer to four years adjusted. There is no imminent dilution risk here. Q4 operating cash flow turned positive, suggesting the burn rate is already inflecting.

What We Found in the Filings

The 10-K reveals several signals the market has not priced.

Deferred revenue grew 25.3% year-over-year, more than three times the rate of recognized revenue. To understand why that matters, you need to understand how Disco gets paid. Ninety-one percent of revenue is usage-based: customers process data, get billed, revenue is recognized. No long-term contracts, no guaranteed minimums. Month-to-month. In that model, deferred revenue only appears when customers prepay for committed capacity or purchase processing credits upfront. The total is still small ($5.4 million on $157 million in revenue), but the growth rate is the signal. More customers are committing cash before they consume. In a business where most revenue can be canceled next month, any increase in prepayment is a forward indicator of customer confidence and revenue durability.

The company disclosed a shift to usage-based per-gigabyte pricing for 2026, with Cecilia AI and Case Builder bundled for all customers at no additional charge. This is a significant commercial decision. By removing the AI pricing friction (no separate upsell, no feature gating), Disco is betting that broad AI adoption will drive usage volume growth that more than offsets the per-unit price concession. Relativity is making the same bet, bundling its GenAI tools into standard RelativityOne pricing effective April 2026. The industry is moving toward AI-as-standard, and Disco is positioned for that shift.

Auto Review, the company’s first generative AI product, was introduced during FY2025. Cecilia AI usage tripled year-over-year. These are not press release claims. The adoption metrics are disclosed in SEC filings and referenced in the Q3 2025 earnings call, where Friedrichsen stated that the strategy of bringing large clients and large matters to Disco was gaining traction with “meaningful acceleration in both software and total revenue.”

Management guided FY2026 revenue of $167-177 million (7-13% growth) with a target of reaching adjusted EBITDA breakeven by Q4 2026. Throughout FY2025, the company consistently beat its own guidance, suggesting the targets are conservative.

What Has to Go Right

Relativity, the category leader, filed its S-1 on March 19 targeting a $4 billion valuation. When it prices, it establishes the first public-market valuation benchmark for ediscovery since DISCO’s own 2021 IPO. Disco is the only comparable public company. Investors looking at Relativity will find Disco trading at one-twenty-fifth of the price. Even if Relativity struggles, the attention the IPO brings to the legal tech category could drive incremental coverage and investor awareness to the only other pure-play in the space.

Beyond the IPO, the catalyst is the cumulative weight of the next two to three quarters confirming the trajectory. If Q1 and Q2 2026 show revenue acceleration above 8%, continued margin expansion, and progress toward the EBITDA breakeven target, the stock will rerate. Two analysts cover this company. Needham has a $10 price target (175% upside from current levels). Jefferies has a $8 target. The stock is so far below both targets that even modest execution would close a meaningful portion of the gap.

The macro environment is mixed for legal tech. Litigation volumes tend to be acyclical, but the company’s 10-K added new language about “decreased regulatory enforcement and government shutdowns” as potential headwinds. The broader AI investment cycle is a tailwind: every law firm evaluating AI tools for document review will encounter Disco in the consideration set.

The Risks

• Cash burn remains elevated. FY2025 operating cash outflow was $42.1 million (GAAP), though this includes the $9.2 million litigation charge. Adjusted cash burn is closer to $33 million. Q4 turned positive but a single quarter does not confirm a trend. If revenue growth stalls, the cash cushion provides time but not indefinite patience.

• Competitive intensity is increasing. Relativity bundling AI into standard pricing, Everlaw ranked #1 on G2 for four consecutive quarters, and AI-native entrants entering the market. DISCO’s 98% net retention rate is improving but remains below the 110%+ benchmark of best-in-class SaaS compounders. Customer expansion is not yet strong enough to outpace potential churn pressure from better-funded competitors.

• Stock-based compensation is elevated. SBC runs at $24.5 million annually (16% of revenue), typical for an unprofitable SaaS company at this stage but dilutive until earnings growth offsets it. The equity plan includes automatic 5% annual share reserve increases through 2031. Management comp is tied to revenue and EBITDA targets with real accountability (73% of target achieved in 2024, 36.5% of PSUs vested). The dilution becomes a non-issue if the company reaches breakeven. If it doesn’t, shareholders are giving away equity into a business that isn’t growing fast enough to compensate.

• Legacy securities litigation is unresolved. A class action filed September 2023 covering July 2021 to August 2022 alleges former management failed to disclose revenue deceleration. The company accrued $9.2 million with $8 million in insurance recovery, implying net exposure of $1.2 million. The case is in the pleading stage under new management that was not involved in the alleged conduct.

• Revenue growth, while accelerating, remains single-digit. The jump from 5% to 8% is encouraging but 8% growth does not command premium multiples. The thesis requires acceleration to 12-15%+ to justify a meaningful rerate. If the new pricing model or Auto Review adoption disappoints, the growth trajectory could stall in the high single digits.

The Bottom Line

CS Disco is a $157 million revenue legal tech platform trading at an enterprise value of $42 million after backing out net cash. The adjusted margin inflection of +675 basis points is genuine operating leverage, driven by sales efficiency gains, confirmed by Q4 positive cash flow, and clearly visible once you strip the one-time items from both years. Two analysts cover the stock, Relativity’s IPO will soon establish a category valuation benchmark, and the company has nearly four years of adjusted cash runway with no debt. The risk is execution: the new management team needs to sustain revenue acceleration and reach EBITDA breakeven by Q4 2026. If they do, the current valuation will look like a pricing error.

We’re watching the Q1 2026 10-Q (expected May 2026) for revenue growth above 8%, continued margin expansion, and management’s updated commentary on Auto Review adoption and the new pricing model’s early impact.

This is a Top Shelf analysis. Future stock reports from both Top Shelf and Clearance Rack will be available to paid subscribers only. Subscribe now to be grandfathered in at our launch price.

Disclaimer: This is not investment advice. The author may hold positions in securities discussed. All information is based on publicly available SEC filings, insider transaction records, and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.

And almost every time, the new company gets sold by the people who receive it.

The Mechanics of a Mispricing

Here’s how a spin-off works. A parent company decides that one of its divisions would be better off on its own. Maybe the division is growing faster than the parent and the market isn’t giving it credit. Maybe the parent wants to focus. Maybe an activist investor pushed for it. Whatever the reason, the parent files a Form 10-12B with the SEC, registers the new entity, and distributes shares of the subsidiary to its existing shareholders on a pro-rata basis.

The shareholders of the parent company wake up one morning with shares of a company they never chose to own. They didn’t research it. They didn’t decide it fit their portfolio. It just showed up.

And that’s where the mispricing begins.

Who Sells and Why

Index funds are the first forced sellers. If the parent is in the S&P 500 and the new spin-off doesn’t meet the index criteria (most don’t, because they’re usually much smaller), the fund has to sell. Not because the stock is bad. Because the mandate says they can only hold what’s in the index. The selling is mechanical. It has nothing to do with the value of the business.

Institutional funds are next. A large-cap growth fund that owned the parent for its size and sector exposure suddenly holds shares in a small-cap company that doesn’t fit their strategy. Their compliance department flags it. Their portfolio manager sells it. Again, not because of fundamentals. Because of fit.

Then come the retail shareholders who simply don't know what they've received. They see a small, unfamiliar position in their brokerage account. They don't understand the business. They don't even remember buying it, because they didn’t. They sell and move on.

All of this selling pressure hits a stock with almost no analyst coverage, limited institutional awareness, and a thin float. The result is predictable. The price drops below what the business is actually worth.

The Evidence

This pattern has been studied for decades, and the results are remarkably consistent.

A study at Penn State measured spin-off performance against the S&P 500 over a 25-year period ending in 1988. Spin-offs returned roughly 20% per year. The S&P 500 returned 10%. That’s a 10% annual outperformance, sustained over a quarter century.

Researchers at Purdue University extended the analysis through 2013 and found the same result. Spin-off subsidiaries averaged a 31.7% buy-and-hold return over the first 22 months of trading, significantly outperforming their size and value benchmarks.

Cusatis, Miles, and Woolridge published the first major academic study on spin-off returns in 1993. They found spin-off subsidiaries generated excess returns of 4.5% after one year and 33.6% after three years.

More recently, Trivariate Research compiled data showing that newly spun-off entities outperform the S&P 500 by an average of 10% over the subsequent 18-24 months. The parent companies, meanwhile, performed roughly in line with the index.

Source: Trivariate Research, “Spin-Offs: We Will See More in 2025”

The data spans different decades, different market conditions, and different researchers. The conclusion is always the same. Spin-offs outperform.

Subscribe now

Why It Persists

Most market anomalies disappear once they become widely known. This one hasn't. The excess returns have compressed over time (the earliest studies showed 10% annual outperformance, more recent data shows 10% cumulative over 18-24 months), but the pattern persists. The reason is structural.

Index funds can’t choose to hold a stock that falls outside their benchmark. They will always sell. Institutional mandates don’t flex because a spin-off looks undervalued. Compliance rules don’t care about your DCF model. The forced selling isn’t driven by human judgment. It’s driven by rules, and rules don’t adapt.

There’s also the coverage gap. When a company splits in two, the parent usually keeps its analyst coverage. The spin-off starts life with zero coverage or maybe one analyst if it’s lucky. No coverage means no institutional awareness. No institutional awareness means no buying pressure to offset the forced selling. The stock sits in no-man’s-land until someone notices it.

Joel Greenblatt wrote about this extensively in “You Can Be a Stock Market Genius.” Forced selling, institutional neglect, and information asymmetry create a repeatable opportunity for anyone willing to do the work. The insiders at these companies know what they have. The market just hasn't caught up yet.

What to Look For

Not every spin-off is a buy. Some parent companies use spin-offs to dump their worst assets and saddle the new entity with debt. The key is separating the genuine opportunities from the garbage.

Insider behavior after the spin-off tells you more than any analyst report. If the CEO and directors of the new entity are buying stock on the open market with their own money in the weeks after separation, they’re telling you something. They know the business better than anyone, and they think it’s too cheap.

Management compensation structures matter. If the new company’s executives have performance-based stock options that vest at prices well above where the stock is trading, their financial incentives are aligned with yours. Check the proxy statement.

The balance sheet of the spin-off is critical. Did the parent load it with debt before separating? A clean balance sheet gives the new management room to operate. A leveraged one can turn a good business into a value trap.

And timing matters. The heaviest forced selling typically happens in the first few weeks after the spin-off goes effective, as index funds and institutions work through their mandated dispositions. The academic research consistently shows that performance improves the longer you hold past that initial selling wave, with the strongest excess returns appearing 12-24 months after separation.

Why This Matters

Spin-offs sit at the center of what Thrift Shop Stocks is built around. These are companies that end up on the clearance rack not because anything is wrong with them, but because the holders were forced to sell. The selling is predictable. The recovery is supported by decades of academic evidence. And the opportunity exists because the structural dynamics that create the mispricing haven’t changed.

Index funds aren’t going away. Institutional mandates aren’t getting more flexible. Analyst coverage for sub-$500M companies isn’t expanding. As long as those conditions persist, spin-offs will keep getting thrown away by sellers who never wanted them in the first place.

Someone just has to be willing to pick through what’s left.

Disclaimer: This post is for educational and informational purposes only. Nothing here constitutes investment advice, a recommendation, or a solicitation to buy or sell any security. Always do your own research before making any investment decisions.

TransAct Technologies just acquired the source code to its own software platform, turned EBITDA positive for the first time in two years, and grew revenue 19%. Its CEO bought $351,000 in stock across two open-market purchases in three days. The market still prices it like a printer company on its way to zero. At an enterprise value of ~$15 million on $51.5 million in revenue, this might be the cheapest software transition story on a public exchange.

The Software Business the Market Doesn’t See

BOHA! is a cloud-based software and hardware platform for restaurant and food service back-of-house operations. It handles food safety labeling, temperature monitoring, inventory management, checklists, and delivery management from a single terminal. Convenience stores and restaurant chains use it to stay FDA-compliant and keep kitchens running without spreadsheets and clipboards. One platform, one terminal, one vendor relationship, which matters to operators managing hundreds of locations.

FST revenue was $19.3 million in 2025, up 20% year-over-year, with terminal unit sales up 36% to 7,317 units. Recurring FST revenue hit $3.4 million in Q4 alone, with label sales reaching an all-time high. Annualized recurring FST revenue is already approaching a $14 million run-rate based on Q4 results.

On August 6, 2025, the company acquired a perpetual, royalty-free license to the BOHA! software source code from Avery Dennison (per the 10-K filed March 12, 2026, referencing the Source Code Purchase and Perpetual License Agreement dated August 5, 2025, disclosed as Exhibit 10.24). Before this deal, TransAct was licensing its core software from a third party. Every feature improvement, every pricing change, every hosting decision required coordination with an outside vendor. That dependency is now gone, and TransAct owns the code.

The company plans to complete the transition to internal hosting by mid-2026 and launch a revamped cloud platform with what CEO John Dillon described on the Q4 earnings call as an “app-store concept” targeting roughly $200 per machine per month in recurring revenue. He also noted the company is already using AI tools to analyze the BOHA! codebase for inefficiencies and plans to add AI-driven features for customers around inventory optimization and operational decisions. Current ARPU is $756 per year, or about $63 per month. The long-term target represents roughly 3x the current run-rate.

Dillon was direct about the strategic bet: “With all rights to the BOHA! software now in hand and full ownership of all associated modifications, innovations and licensing models now secured, we believe that software is unequivocally our growth engine going forward.”

Dillon took over as CEO in April 2023 after serving on the board for 12 years. His resume includes CEO roles at Salesforce.com (early stage), Hyperion Solutions, and Navis (sold to Zebra Technologies). He’s a Naval Academy graduate who spent five years in the nuclear submarine service before his corporate career. He replaced founder Bart Shuldman, who had run TransAct for 27 years. This is not a legacy CEO managing decline. This is someone who has run software transitions before and chose to take this job. Since Dillon arrived, revenue has grown from $43.4 million to $51.5 million, losses have narrowed 87%, and the company crossed into positive adjusted EBITDA.

The Cash Engine Funding All of It

TransAct’s other business makes thermal printers for casino slot machines. The ticket-in/ticket-out (TITO) printers that spit out a paper voucher when you cash out. Nearly every major casino floor runs these. TransAct has sold over 3.9 million printers and terminals globally over the company’s lifetime, and the installed base generates steady consumables and service revenue. Casino and gaming brought in $26.9 million in 2025, up 32% year-over-year, with strong international demand. This segment carries high gross margins and provides the free cash flow that funds the software transition. It’s a thrift shop stock in its own right, a profitable, cash-generative business buried inside a company the market has written off.

Both businesses produce mission-critical revenue. The TITO printers are embedded in casino floor infrastructure. Every slot machine payout runs through them. The BOHA! terminals sit in restaurant kitchens handling FDA compliance. These aren't products a customer evaluates against alternatives every year. They're operational infrastructure that gets installed and stays.

A Micro-Cap Priced Like It’s Going Bankrupt

Yahoo Finance mentions BOHA! in its profile for TransAct, but still leads with 'develops and sells software-driven technology and printing solutions.' For a company that just acquired full ownership of its software platform and whose CEO called it 'unequivocally our growth engine going forward,' the printer is still the headline. For a $34 million market cap stock covered by a only one analyst and 43,000 shares trading hands on an average day, that profile is likely the first and last thing most investors ever read. Nobody is looking closely, and that is opportunity.

Short interest is negligible at 0.07% of float. That’s not a bullish signal so much as a reflection of how far off the radar this stock sits. There’s nobody on the other side of this trade because there’s nobody looking at it.

Revenue grew 19% to $51.5 million, gross margin came in at 48.6%. Adjusted EBITDA flipped from negative $1.5 million to positive $1.2 million, net loss narrowed 87% from $9.9 million to $1.2 million. Cash grew 42% to $20.4 million against just $3 million in borrowings.

Gross margin dipped 90 basis points year-over-year, driven by a higher mix of lower-margin FST hardware sales as terminal unit volumes grew 36%. That's the right kind of margin compression. You're trading a fraction of a point on gross margin for 20% revenue growth and rapid expansion of the installed base that becomes the software platform.

The balance sheet is clean. TransAct holds $20.4 million in cash against just $3 million in borrowings (the minimum draw on its credit facility), that's a net cash position of $17.4 million. Cash per share is roughly $2.02, meaning at the current $3.29 price you’re paying about $1.27 per share for the operating business.

Management’s compensation is tied to these numbers. The proxy filing (DEF 14A, filed April 2025) shows Dillon’s performance stock units are linked to adjusted EBITDA targets, and the 2024 PSU awards were completely forfeited because the company missed threshold. When management misses, they don’t get paid.

Enterprise value is approximately $15 million. On $51.5 million in trailing revenue, that’s 0.29x EV/Sales. To show how disconnected this is from restaurant tech peers:

Toast trades at 2.4x revenue on $6.2 billion in sales with roughly 10% EBITDA margins. PAR Technology commands nearly 5x on $340 million in revenue and is still unprofitable. Lightspeed sits around 1x on $900 million with low single-digit margins. TransAct trades at 0.29x on $51.5 million with a 2.3% EBITDA margin that just turned positive. The margin is thin, but Toast ran negative EBITDA for years before its inflection. PAR is still there. The market gives those companies credit for future margin expansion. TransAct hasn’t received any… yet.

Management guided fiscal 2026 to $55-57 million in revenue and $0.8-1.5 million in adjusted EBITDA. That implies continued top-line growth of 7-11% with profitability holding steady as the company invests in its software platform launch.

The 10-K comparison between fiscal 2024 and 2025 also shows a clean remediation. The material weakness in internal controls that was previously disclosed has been resolved, and all restatement language from December 2024 has been removed. Inventory dropped 33% from $16.2 million to $10.9 million while revenue grew 19%. Cash grew 42% from $14.4 million to $20.4 million. A new intangible asset of $1.5 million appeared on the balance sheet, the capitalized value of the source code acquisition.

When the Market Catches Up

The mid-2026 software platform launch is the event that could reprice this stock. When TransAct starts reporting its own hosted SaaS revenue at full margins instead of sharing economics with a third-party licensor, the financial profile changes. Software gross margins in food service tech typically run 70-80%, well above TransAct’s current blended 48%. Every incremental dollar of recurring software revenue should flow through at meaningfully higher margins.

The BOHA! economics work like a razor-and-blade model. A restaurant buys a terminal for $500-800 upfront, then pays ongoing fees for software subscriptions, custom labels, and services. Contracts typically run three years. The installed base of terminals is growing 36% annually, and each unit placed is a seed for recurring revenue. Management has stated that a large customer initially carries no recurring revenue, but they plan to change the selling model in 2026. If that customer converts, ARPU should inflect upward.

Meanwhile, the casino gaming business keeps generating cash. The company hired a new Chief Marketing Officer in January 2026 specifically to accelerate platform growth. This isn’t a business that needs to raise capital or take on debt to fund the transition. The $20 million cash balance and casino cash flows provide the runway.

Dillon purchased 29,098 shares at $3.56 on March 13, 2026, the day earnings were released, then came back three days later and bought another 70,902 shares at $3.49 on March 16 (per Form 4 filings on March 14 and March 18, 2026). That’s 100,000 shares for $351,037 in total, more than tripling his position in a single week while the stock traded near its 52-week low. Director Daniel Friedberg holds 10.13% of shares outstanding. Total insider ownership is 16.46%. These are people with real skin in the game.

What It Could Be Worth

You can’t put a P/E on a company that just crossed into positive EBITDA for the first time. The question is what multiple the market assigns to the revenue once the margin profile changes. EV/Revenue is the right framework at this stage, and the range of outcomes is wide.

Floor (~$2.00, cash per share). If BOHA! fails entirely and the operating business is worth zero, you recover $20.4 million in cash, or $2.02 per share. That’s 61% of today’s stock price backed by cash alone. Your total exposure to the operating business is $1.27.

Base (~$4.50-5.00, 0.5x EV/Revenue). TransAct keeps growing modestly and holds positive EBITDA, but the market never re-rates it. 0.5x revenue is where low-margin industrial hardware companies land. On guided $56 million in 2026 revenue, that’s roughly $28 million EV plus $17 million net cash, divided by 10.1 million shares. You land around $4.50, which lines up with the single analyst’s $5 target and Morningstar’s $5.93 fair value. No re-rating required, just the market catching up to the financials as reported.

Bull (~$10.00, 1.5x EV/Revenue). The software platform launches, recurring revenue scales, margins expand, and the market starts applying a partial software multiple. 1.5x on $56 million is roughly $84 million EV plus $17 million net cash, for about $10 per share. That’s still deeply discounted to Toast at 2.4x and PAR at 5x. It just requires the market to acknowledge the business model is changing.

At $3.29, the downside to the cash floor is $1.27. The upside to the bull case is $6.71. The base case alone nearly covers the risk.

The Risks

• The BOHA! software transition could stumble. Migrating from third-party hosting to an internally managed platform by mid-2026 is execution risk. If the transition causes service disruptions, customer retention and new sales could suffer. The 10-K explicitly flags this as a risk factor.

• ARPU has been declining, not growing. Q4 ARPU fell 14% year-over-year to $756, driven by a large customer that currently generates no recurring revenue. If TransAct can’t convert that customer to a paying software model in 2026, the metric will keep deteriorating even as terminal sales grow.

• Casino and gaming revenue could soften. Management noted domestic demand weakness in Q4, with a large customer working down inventory. If the gaming cycle turns, the cash engine funding the FST investment gets smaller.

• This is a $34 million market cap stock with average daily volume of roughly 43,000 shares. Getting in and out of a meaningful position takes patience and the bid-ask spread can be wide.

• The company remains unprofitable on a GAAP basis. There is no guarantee the business reaches sustained profitability if the software launch underperforms.

The Bottom Line

TransAct is a $51 million revenue business with $20 million in cash, no meaningful debt, and an enterprise value of ~$15 million. The filings reveal a company that just took ownership of its core software, resolved its internal controls issues, and is investing to turn a hardware product into a recurring-revenue platform. The CEO is buying stock at the low. If the mid-2026 software launch executes, the market will have to decide whether 0.3x EV/Revenue is the right multiple for a growing food service SaaS business with expanding margins and a 36% unit growth rate. At $3.29, the risk-reward says it is not.

Disclaimer: This is not investment advice. The author may hold positions in securities discussed. All information is based on publicly available SEC filings, insider transaction records, and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.

Sometimes that’s true. But most of the time, the headline is noise. The SEC filing that triggered it could be anything from a CEO loading up on shares with their own savings to a routine option exercise that a computer executed on a schedule six months ago. Those two events carry completely different information. Learning to tell them apart is one of the simplest edges a retail investor can develop.

What the SEC Actually Requires

Every officer, director, and 10%+ shareholder at a public company has to report their transactions to the SEC on Form 4. These filings hit EDGAR within two business days of the transaction. They are public, free, and searchable. The filing tells you exactly who traded, how many shares, at what price, and on what date.

What the filing does not tell you, at least not in bold text, is why. That part you have to figure out by looking at the transaction code. There are about a dozen different codes the SEC uses to classify insider transactions, and the differences between them are everything.

The Transaction Codes That Matter

Every transaction on a Form 4 carries a single-letter code. Here are the ones you will see most often.

Code P: Open-Market Purchase. This is the one you care about. Code P means the insider went to their broker and bought shares on the open market, with their own money, at the prevailing market price, just like you would. No discount. No corporate subsidy. No obligation. They chose to buy because they wanted to own more of the company. This is the only transaction code that consistently carries predictive value.

Code S: Open-Market Sale. The mirror image of P. The insider sold shares on the open market. Sales get less attention than purchases for a good reason. Insiders sell for all kinds of non-informational reasons: diversification, a new house, a divorce settlement, tuition payments. Selling tells you something, but it tells you less than buying.

Code M: Exercise of Derivative Security. This is where most of the noise lives. Code M means the insider converted stock options into actual shares. In many cases, it is a “cashless exercise” where they simultaneously sell enough shares to cover the cost, meaning their net position does not change at all. The SEC reports this as both a purchase and a sale. Headlines report the purchase side and leave out the rest.

Codes A, F, and G: Award, Tax Withholding, Gift. These are all non-decisions. Code A means the company granted shares as compensation. Code F means shares were automatically sold to cover a tax bill on vesting stock. Code G means the insider gave shares away for estate or tax planning purposes. In all three cases, the insider made no active choice to increase or decrease their exposure to the stock. Zero signal across the board.

The 10b5-1 Problem

Even within Code P purchases, there is a critical distinction most people miss. Some open-market purchases are pre-planned under what is called a Rule 10b5-1 trading plan.

Here is how it works. An insider sets up a plan in advance that says “buy 5,000 shares on the first trading day of every month for the next 12 months.” Once the plan is in place, the trades execute automatically regardless of what is happening at the company. The insider could know terrible news is coming and the plan still runs. That is actually the legal protection the plan provides: the insider can argue the trades were pre-committed and not based on material nonpublic information.

These planned purchases still show up as Code P on Form 4. Since April 2023, the SEC requires filers to check a box at the top of the form indicating whether the transaction was made under a 10b5-1 plan and to disclose the date the plan was adopted. Before that rule change, the only way to spot them was a small footnote buried at the bottom of the filing. Either way, most screeners and headline aggregators ignore both the checkbox and the footnotes entirely. They just see Code P and report “insider buying.”

A 10b5-1 purchase is not worthless information. An insider had to decide to set the plan up at some point. But it carries far less signal than a discretionary purchase, where the insider looked at the current price, looked at what they knew about the business, and decided to write a check.

The same logic applies in reverse. When you see an insider sale that looks alarming, check whether it was pre-planned under a 10b5-1 arrangement before you react. A director selling 5,000 shares on a schedule set up six months ago is not telling you anything about the company today.

What Makes a Strong Buy Signal

So you have confirmed it is a Code P, open-market, discretionary purchase. Not every one of these is equally meaningful. Here is what separates a compelling signal from a routine one.

Size matters, but relative size matters more. A CEO who owns $20 million in stock buying another $50,000 is barely rounding up. That same CEO buying $2 million is putting real skin in the game. Look at the purchase as a percentage of what the insider already holds and what they earn. An executive making $500,000 a year who voluntarily puts $200,000 into the stock is telling you something worth listening to.

Seniority of the insider. The CEO and CFO have the clearest picture of the company’s future. A director on the audit committee has a decent view. A vice president of regional sales has a narrower lens. Weight the signal accordingly.

First purchase in a long time. An insider who buys stock every quarter like clockwork is giving you less new information than an insider who has not purchased shares in three years and suddenly shows up with a six-figure buy. The break in pattern is what carries the signal.

Purchases during bad news. If the stock just dropped 30% on an earnings miss and the CEO steps in and buys, that is someone telling you with their wallet that the market overreacted. Insiders buying into weakness is considerably more informative than insiders buying into strength.

Cluster Buying: When Multiple Insiders Agree

One insider buying is interesting. Three insiders buying within the same two-week window is a pattern.

This is called cluster buying, and it is one of the strongest signals in the insider data. When a CEO, a CFO, and a board member all independently decide to purchase shares around the same time, the odds that all three are wrong go down considerably. Each of these people has a slightly different vantage point on the business. The CEO sees the strategic picture. The CFO sees the numbers. The director sees governance and board-level discussions. When all three reach the same conclusion, and back it with their own money, they are collectively telling you something that no single purchase could.

The key word is “independently.” If a company announces a new insider buying program or the board passes a resolution encouraging directors to increase their holdings, that is coordinated, not independent. Check the 8-K filings around the same period. If there is a formal program, the purchases are compliance, not conviction.

What you want to see is discretionary purchases by different insiders at slightly different times, at slightly different prices, with no corporate mandate behind them. That pattern suggests each person looked at the situation and individually decided the stock was cheap.

Pay attention to the composition of the cluster too. Three directors buying is good. The CEO and CFO buying together is better. A mix of C-suite executives and directors is the strongest version, because it means people with different information sets are all reaching the same conclusion.

If you feel this ticker looks familiar, it’s because I wrote about it in my last post.

The Signal Most People Miss: Cross-Industry Insider Buying

There is a second layer to cluster buying that is rarely discussed, and it is arguably more powerful than the single-company version, but more difficult to track.

Instead of looking at multiple insiders buying within one company, look at insiders across different companies in the same industry all buying around the same time. If the CFO of a specialty chemical company buys stock in the same two-week window that the CEO of a competitor does, and then a director at a third company in the same space follows a few days later, you are looking at something much bigger than one company’s story.

These people do not coordinate with each other. They are competitors. But they are all seeing the same industry-level dynamics: input costs falling, customer orders picking up, a regulatory headwind clearing, a destocking cycle ending. When multiple insiders across an industry independently reach for their checkbooks within a narrow window, they are collectively signaling a sector-level turn that the broader market has not priced in.

This is especially powerful in small-cap and micro-cap industries where analyst coverage is thin. In sectors with heavy Wall Street coverage, an industry turn gets flagged quickly through earnings estimate revisions and sector notes from analysts. In the neglected corners of the market, there may be no analyst paying attention at all. The insiders are often the first people to see the turn, and their Form 4 filings are the only public trace.

The challenge is that this pattern is nearly impossible to spot manually. You would need to track Form 4 filings across dozens of companies, group them by industry, and look for temporal clustering. Nobody does this by scrolling through EDGAR. But the data is all there, sitting in public filings, waiting for someone to connect the dots.

A practical way to think about it: if your normal process flags insider buying at one company, take five minutes to check whether any competitors have had insider purchases in the last month. If they have, the signal just got a lot stronger. You are no longer betting on one management team being right about their own company. You are betting on an entire industry’s insiders being right about their shared operating environment.

Once you start spending enough time in these filings, you start to notice patterns most people never look for.

The Three-Question Filter

When you see an insider buying headline, run it through three questions before you do anything else. Was it a Code P open-market purchase? Was it discretionary, or was it a pre-committed 10b5-1 plan? And was the size meaningful relative to what this person already owns and earns?

If the answer to any of those is no, keep scrolling. You are looking at noise. Option exercises, awards, tax withholding sales, gifts, directors meeting minimum ownership thresholds, and stale 10b5-1 plans all show up in the same filings and the same screeners as genuine conviction purchases. They will fool you if you do not check the 10b5-1 checkbox and the footnotes.

If the answer to all three is yes, you have something worth paying attention to. And if you find that same pattern repeating across multiple insiders, or across an entire industry, you might be looking at something the rest of the market has not figured out yet.

Disclaimer: This post is for educational and informational purposes only. Nothing here constitutes investment advice, a recommendation, or a solicitation to buy or sell any security. Always do your own research before making any investment decisions.

Clearance Rack is a series that finds structurally mispriced stocks. This could be due to forced selling, spin-offs, or institutional dumping created a gap between what the filings say and what the market believes. Good businesses that didn’t fit the funds that owned them. The full breakdown of the shop lives on the about page.

Grocery Outlet Holding Corp. (NASDAQ: GO) | ~$5.99 | Market Cap: ~$570M

Between March 9 and March 19, five senior executives at Grocery Outlet collectively spent over $5 million of their own money buying shares in a ten-day window. They bought at prices between $5.82 and $6.26, near the stock’s all-time low. Meanwhile, a third of the company’s publicly available shares are sold short and Wall Street analysts are tripping over each other to downgrade it.

Somebody is wrong. The insiders are betting it isn’t them.

What Grocery Outlet Actually Does

We spend our time digging through overlooked filings for undervalued finds, so there’s something fitting about our first pick being a company that does the same thing with groceries. Grocery Outlet is America’s largest “extreme value” grocer. Think TJ Maxx, but for food. The company buys overstock, closeouts, and packaging changes from major brand manufacturers at steep discounts and sells them through roughly 530 stores across 16 states. If Kraft made too much mac and cheese, or a cereal brand redesigned its box, that surplus ends up on Grocery Outlet shelves at 40-70% below what you’d pay at Kroger or Safeway. They spot value where the big brands see a problem, and pass the savings on.

The model has an unusual twist. Most stores aren’t run by corporate employees. They’re operated by independent entrepreneurs (called IOs) who function like local franchise owners. They pick their own product mix based on what sells in their neighborhood, they manage their own staff, and their income is directly tied to how well their store performs. Corporate handles the buying muscle and the brand. It’s a “small business at scale” setup, founded in 1946 and still carrying the fingerprints of the Read family, who built it from a surplus cannery business in San Francisco.

The company is now led by Jason Potter, who took over as CEO in February 2025 after the prior CEO departed abruptly in late 2024. Potter spent 26 years at Sobeys, a major Canadian grocer where he oversaw 1,500+ stores, then ran The Fresh Market for five years and is credited with turning that business around. The new CFO, Christopher Miller, joined from Shamrock Foods at roughly the same time. Both have been on the job for about 13 months. Enough time to diagnose the problems, not yet enough to prove they can fix them.

The Numbers

Grocery Outlet’s fiscal 2025 results were brutal on a headline basis. The company reported a net loss of $224.9 million versus net income of $39.5 million the prior year. The stock fell 28% in a single session after Q4 earnings came in below expectations and management guided FY2026 revenue and EPS well below what Wall Street was modeling.

But those headline numbers are almost entirely driven by non-cash write-downs: $149 million in goodwill impairment, $121.5 million in long-lived asset impairments (mostly stores marked for closure), and $45.9 million in restructuring charges. Strip all of that out and the underlying business looks different. Adjusted EBITDA actually grew 7.4% to $254.3 million. Adjusted net income was $75.2 million, essentially flat year-over-year. Cash from operations nearly doubled to $222 million.

That 5.4% EBITDA margin is worth putting in context. Conventional grocery chains like Kroger typically run at 4-5%, and the broader grocery industry average sits around 3-5%. Grocery Outlet is actually in line with or slightly above its grocery peers even in a down year. The company has historically operated closer to 6-7% before the operational problems stacked up, which means there’s a clear margin recovery path without needing to reinvent the model.

On the balance sheet, total debt stands at roughly $500 million against about $52 million in cash, leaving net debt of approximately $448 million. That’s about 2.0x EBITDA, which is manageable for a grocery retailer but not exactly comfortable. The debt-to-equity ratio is 0.49. This isn’t a balance sheet crisis, but it’s a constraint. The company doesn’t have much room to absorb another round of major setbacks.

The company generated $4.69 billion in revenue, gross margins edged up to 30.3%, and same-store sales were still positive at +0.5% on a 52-week basis, though trending in the wrong direction with Q4 declining 0.8%.

Metric Value Revenue (FY2025) $4.69B Adjusted EBITDA $254.3M (+7.4% YoY) Cash from Operations $222.1M Net Debt / EBITDA ~2.0x Price / Sales ~0.12x Market Cap ~$570M FY2026 EPS Guidance $0.45 to $0.55

At roughly 0.12x sales, the stock trades at about a third of Kroger’s P/S multiple and a fraction of what the market gives discount peers like Dollar General (~0.45x) or Ollie’s (~2.0x). Even on an enterprise value basis, which accounts for the company’s $448 million in net debt, GO trades at just 0.22x sales. The market is pricing this as if the business is structurally broken. The question is whether the write-downs represent past mistakes being flushed, or a preview of more to come.

What We Found in the Filings

CEO Jason Potter led the insider buying with a $1.69 million personal purchase, 286,097 shares at $5.90, on March 19. Chairman Eric Lindberg, who married into the founding Read family and has been with the company for 26 years, bought $1.64 million worth at $5.98. Three other executives bought between $250,000 and $750,000 each. Five buyers, $5 million total, all clustered in a ten-day window at nearly identical prices. Nobody does that for optics.

The name that stands out most is Erik Ragatz. He's a board director now, but more importantly, he's the former Hellman & Friedman Managing Director who led the private equity firm's acquisition of Grocery Outlet in 2014. He sat on the board through the IPO in 2019 and has watched every phase of this company's evolution. Ragatz personally bought nearly $2 million in shares across multiple transactions during this window, adding to an existing position worth over $5 million across trusts, partnerships, and personal accounts. When the person who underwrote the original buyout is reaching into his own pocket at $5.75-$6.09 to make a large position even larger, that tells you something about how he sees the gap between the current price and what the business is worth.

When we compared the current 10-K to the prior year filing, several significant changes stood out. In Potter’s first 13 months, the new leadership team has resolved the material weakness in internal controls (an adverse audit opinion that had been a serious governance red flag), fixed the ERP system implementation that had disrupted ordering and inventory management since August 2023, and expanded the private label program from roughly 180 products to 485 in about four quarters, a 170% increase. A prior federal securities class action that appeared in the previous filing has also been removed, presumably dismissed or settled favorably.

These aren’t cosmetic changes. They represent the clearing of multiple operational problems that had been dragging on the business simultaneously. And they happened under Potter’s watch, which is relevant when assessing whether this management team can execute.

The Catalyst

Consumer sentiment is deteriorating. The University of Michigan Consumer Sentiment Index fell to 55.5 in March 2026, with consumers across all income and age brackets reporting weaker expectations for personal finances, down 7.5% nationally. Year-ahead inflation expectations are holding at 3.4%, above pre-pandemic norms and elevated enough to keep household budgets under pressure. The Israel-US-Iran conflict is pushing gasoline prices higher. SNAP benefit disruptions during the government shutdown directly impacted Grocery Outlet’s core customer base in Q4.

This is the environment where extreme-value grocery formats should thrive. When consumers feel squeezed, they trade down. Grocery Outlet’s entire model is built for this moment: branded products at 40-70% off conventional retail pricing, sourced from the surplus that manufacturer distress and inventory volatility create. More economic stress means more closeout inventory flowing through the supply chain and more consumers hunting for bargains.

Potter acknowledged on the earnings call that “consumer pressure intensified” and “basket pressure intensified for our core customers.” But he also noted that the 36 store closures, all identified as having “no viable path to sustained profitability,” are expected to deliver about $12 million in annualized EBITDA improvement. The company is still opening 30-33 new stores in 2026 under a more disciplined approach, including initially running some as company-operated locations before transferring them to IOs.

The near-term catalyst is the next quarterly earnings report, which should be the first clean comparison period with the store closures underway, the operational fixes in place, and the new leadership team fully embedded. If same-store sales stabilize and the margin impact of the private label expansion starts showing up, the gap between the stock price and the adjusted operating reality starts to narrow.

The short interest dynamics add another dimension. Roughly 24.7 million shares are sold short, about 34.5% of the float. At current trading volumes, it would take short sellers roughly two weeks to buy back all their borrowed shares, which means if the thesis starts breaking in the bulls’ favor, the unwinding of those short positions could itself become a catalyst as forced buying pushes the price higher. The most recent reporting period showed the first decline in short interest after months of steady increases. Short interest roughly quadrupled from around 6 million shares in early 2024 to nearly 25 million by late 2025. That trend may have just broken.

Between the class action uncertainty keeping institutional buyers away and the heavy short positioning suppressing the price, there’s a window here where the stock may be cheaper than the fundamentals warrant simply because of who isn’t allowed or willing to buy it yet. For the patient thrift shopper, that’s the discount.

The Q1 fiscal 2026 results, expected in late May or early June, should be the first real test of whether the turnaround is taking hold. We’ll be watching.

The Risks

• Turnaround execution is unproven. Potter and Miller have been on the job 13 months. The strategic plan sounds sensible, but the company has now been through two restructurings in 12 months, a CEO departure, a CFO departure, and a Chief Purchasing Officer retirement. The FY2024 bonus payout was just 18.3% of target, reflecting how badly management missed its own operational benchmarks. Plans are not results.

• Same-store sales are trending in the wrong direction. Q4 same-store sales declined 0.8% on a 13-week basis, and FY2026 guidance implies continued softness. In grocery retail, negative sales trends at existing stores compound quickly because the fixed cost base doesn’t shrink. If the consumer environment deteriorates further than expected, the stores that remain open could struggle too.

• New securities fraud class actions create uncertainty. Multiple law firms have filed class actions covering August 2025 to March 2026, alleging misstatements about the company’s expansion strategy. For context, these filings follow a familiar pattern: historically, about 43% of securities class actions are dismissed outright, and motions to dismiss succeed in whole or part roughly 80% of the time. Six different firms racing to file near-identical complaints within days of a stock drop is a playbook, not a prosecution. It’s also worth noting that a prior class action against the company has already been resolved and removed from the filings, which sets a precedent for this type of claim not sticking. Still, the lawsuits will keep risk-averse institutional capital on the sidelines in the short term and add legal costs regardless of outcome.

• The IO model limits corporate control. Independent operators are entrepreneurial and local, which is great when times are good. During a turnaround, corporate has limited ability to mandate operational changes at the store level. Pricing, merchandising, and customer experience all vary by operator. Consistency is harder to enforce than in a company-operated chain.

• Remaining goodwill risk. After the $149 million impairment, substantial goodwill remains on the balance sheet. The auditors flagged goodwill impairment as a new critical audit matter. If performance doesn’t meet projections, further write-downs are possible.

The Bottom Line

Grocery Outlet is a 79-year-old business with a differentiated sourcing model, trading at 0.12x sales and roughly 2x EBITDA, while five insiders, including the CEO, the founding family’s Chairman, and the private equity dealmaker who bought the company a decade ago, collectively put $5 million of personal capital on the table at prices near the all-time low. The write-downs look like a slate-clearing exercise, not an ongoing bleed. The macro environment is moving in the direction that should benefit an extreme-value grocer. And the first cracks in the short seller consensus are starting to show.

The risk is real. This is a turnaround, not a hidden gem, and turnarounds fail more often than they succeed. But when the people closest to the business are buying and everyone else is selling, it’s worth paying attention.

Disclaimer: This is not investment advice. The author may hold positions in securities discussed. All information is based on publicly available SEC filings, insider transaction records, and published financial data. Do your own research before making any investment decisions. Past performance is not indicative of future results.