An acquisition thesis rests on assumptions about the future. Buying cash flow assumes the cash flow persists, and buying growth assumes the growth rate holds. The assumption that matters most is often the one nobody at the investment committee debated.
For the last decade, that assumption in the lower middle market was that a good business stays good. A profitable, essential company with loyal customers and a 20-year operating history was treated as stable. Buyers paid a multiple for the stability and spent the diligence budget on what might undermine it: customer concentration, key-person risk, deferred maintenance, and the quality of the earnings.
That assumption no longer holds by itself. Essential businesses remain essential, but the cost structure underneath them is changing for the first time in a generation, and it is changing unevenly. Two companies with identical financials and identical customer bases can have materially different futures, depending on how much of their labor follows a repeatable pattern and how much requires judgment. A quality of earnings report does not show which is which.
The Intersection is our name for asking both questions at once: is the demand durable, and can the delivery be improved? Each answer is more informative when it is read alongside the other.
Enduring demand depends on who makes the purchase decision
Enduring demand is a narrower claim than essential. Most owners consider their company essential, and in a limited sense they are right, because someone is paying for the product or service today. “Essential” describes the present. “Enduring” is a claim about the future, and a claim about the future needs a mechanism.
The mechanism is straightforward to test. Identify who makes the purchase decision, and whether that person is permitted to decline.
A restaurant owner does not decide whether to service the fire suppression system over the range. NFPA 96 governs fire protection for commercial cooking operations, and the local fire marshal enforces it. The purchase decision belongs to a code official, and a weak quarter does not change the requirement. That revenue stream behaves differently from one that depends on a customer’s budget meeting, even where the two look identical on a trailing 12-month income statement.
Historical financials show that customers bought. They do not show whether customers could have chosen not to.
That test sits inside the first of five screening criteria. Non-discretionary need is the foundational criterion: a business serving purely discretionary demand cannot qualify, whatever its other scores.
Non-discretionary need. Customers buy because they must, not because they want to. The product or service fills an essential need: regulatory, safety, health, infrastructure, supply chain, or basic operations. The test is whether customers would still need it if a recession began tomorrow.
Recession-resilient revenue. Demand may flex in a downturn but does not collapse, because the underlying need is structural rather than cyclical. Performance through 2008 and 2009, and through 2020, is the available evidence, and we give it more weight than any forecast.
Technology-proof core demand. Software can change how a product is made or a service is delivered, but it cannot remove the need for the product or service itself. The physical world requires physical products, physical movement, and physical work. The test is whether a purely digital offering could replace the business entirely; for the businesses we pursue, it could not.
Recurring or repeat revenue. Contracts, maintenance cycles, regulatory schedules, supply agreements, or natural repurchase patterns make revenue predictable, reduce the cost of winning each additional dollar, and increase the lifetime value of each customer.
Favorable market structure. Conditions that leave room to build an advantage: fragmentation, local dominance, underserved niches, or limited competition from institutional capital. Fragmentation is one positive indicator among several and is not a pass-or-fail gate. A dominant position in a small market can be worth as much as a share of a fragmented market with 100 competitors.
The five criteria are a filter. Mandated demand is visible to every buyer, which attracts competition and limits price. It removes one risk, that customers stop buying, and leaves the others in place. A business can meet all five criteria and still be a poor acquisition because of customer concentration, labor availability, or an owner who is also the company’s best salesperson. Enduring demand identifies which businesses deserve further diligence, and the rest of the diligence still decides.
Intelligent automation depends on which hours follow a pattern
The second half of the thesis is often framed as a technology question. We treat it as an operations question, one an industrial engineer in the 1950s would have recognized.
Inside any business, some hours are judgment and some hours are pattern. Judgment hours are the ones in which a person weighs competing considerations and decides: what to bid, whom to hire, whether a customer’s account of a problem is accurate, how hard to negotiate a term. Pattern hours follow a recognizable sequence, and the difficulty is volume rather than insight: assembling the same report from the same inputs, re-entering the same data into a second system, or drafting the first version of a document that has been drafted hundreds of times before.
In the companies we review, both kinds of hours usually sit in the same role and are paid at the same rate, because separating them used to require another hire. Separating them now costs much less than it did, and the constraint has moved elsewhere.
The adoption data show where. As of May 3, 2026, the U.S. Census Bureau’s Business Trends and Outlook Survey put the national AI use rate at 19.8%. Among firms with at least 250 employees, 37% reported using AI; among firms with four or fewer employees, under 20% did. A Census working paper drawing on the survey’s AI supplement found that 57% of adopting firms had integrated AI into three or fewer business functions.
Low adoption among small firms is often attributed to cost or access. The breadth figure points to a different constraint: many firms that have bought the tools have not written down how their work should be done, so the tools have little to apply. Converting a license into a working capability requires someone to document, carefully and accurately, how a recurring piece of work is done. That takes several uninterrupted days, and in an owner-led business with no internal technology function, it is nobody’s assigned job.
That raises the question of how to assess the opportunity accurately. An earlier version of this framework scored each business against a fixed checklist of six improvement categories. The method had a flaw: it rewarded businesses for being behind. The more analog a company looked, the higher it scored, whether or not a credible intervention existed, and the write-ups it produced were generic.
The framework now works from plays. A play is a named intervention in one workflow at one business, stating the data it depends on, the EBITDA impact it is expected to produce, the time to reach that impact, and the reason the advantage persists. Naming a real play for a well-run business is harder than naming one for a poorly run business, and the framework credits the analyst who can do it. The six value levers from earlier versions remain as play classes, which help an analyst find candidates rather than grade a company:
Back-office automation. Taking on manual administrative work such as scheduling, order management, invoicing, logistics coordination, payroll, and receivables.
Customer intelligence and experience. Automated follow-up, predictive lead scoring, and intelligent routing that raise close rates and retention without adding headcount.
Dynamic pricing and estimation. Pricing jobs and orders against complexity, cost, and market data, in place of estimates based on experience alone.
Workforce augmentation. Real-time guidance, quality monitoring, and skills-based task assignment for frontline workers, whether they are technicians, machine operators, warehouse staff, or drivers. These plays give the existing team better tools and better information for the work that needs their judgment.
Predictive operations. Anticipating equipment failure, demand swings, and inventory needs, so the business can act before a failure or shortage instead of after it.
Data flywheel and compounding. Each improved process generates data that informs the next decision, and the business’s information advantage grows the longer it operates.
These six are illustrative rather than exhaustive. A play that fits none of them still counts if it is real, sized, and defensible.
Why the two compound
Each half of the thesis can be defended on its own, and each has a well-understood price when bought on its own.
Buying durable demand without improvable delivery means buying a stable cash flow at the price the market already sets for stable cash flows. The business will likely still be operating in 10 years and will look much as it does today. That is a legitimate investment, and the market prices it accurately.
Buying improvable delivery without durable demand means funding an efficiency project on revenue that may not last. Margin improvement on revenue that later disappears is an expensive form of operational success. It catches disciplined operators because the project can go exactly to plan until the demand falls away.
Together, the two behave differently, and the reason is arithmetic. Cost removed from a business with durable demand stays removed, because the revenue it supported remains. Margin gained on volatile revenue has to be rebuilt after each downturn, while margin gained on mandated revenue accumulates. Over a multi-year hold, that difference changes what the business is worth at exit, beyond the value of a one-time operating improvement.
The protection runs in both directions. In a downturn, demand-resilient revenue persists while operating improvements protect margins. In an expansion, the same infrastructure scales revenue without proportional cost. The demand protects the floor. The automation raises the ceiling.
A second effect follows. In the businesses we review, the ones that meet the demand criteria are disproportionately ones that never built a technology function, because steady demand never required one. Their pattern hours are still undocumented, in the companies whose revenue is most reliable. The durability that makes the demand attractive is also what left the operating slack in place.
Operational complexity adds to this. Businesses with multi-step workflows, compliance requirements, specialized logistics, or technical service delivery already have barriers that less capable competitors find hard to replicate. The more complex the operation, the more workflows there are to improve, and the harder it is for a competitor to match both the operation and the automation built on it.
What this screens out
The framework excludes three kinds of business.
The first is automation without demand. A business can be highly automatable while its demand is discretionary, cyclical, or exposed to digital substitution. Improving efficiency in a shrinking or volatile market produces a well-run company in a market that is going away. Purely discretionary demand disqualifies a business regardless of how it scores on everything else.
The second is demand without room to improve, at least as a platform. A resilient business that is already efficient, or too simple to benefit, has a solid floor and a low ceiling. These are worth monitoring, and some make sound add-on acquisitions for a platform already in place.
The third is a business whose price already reflects the improvement. A company can look right on both dimensions in a market where institutional capital has spent a decade consolidating and has inflated entry multiples to the point where risk-adjusted returns deteriorate. The thesis needs room between the entry price and the improvement to be worth executing.
The framework does not screen on size, sector, or growth rate, and it does not require fragmentation, although fragmentation often accompanies the other criteria for the same reason the pattern hours remain undocumented. It also replaces none of the ordinary diligence that ends most transactions.
What to do with this if you own one of these businesses
If your company meets the first criterion, the implication is not that you should sell. The operating slack in your business now has a specific, measurable value, and you are better placed if you know that figure before a buyer calculates it.
The exercise is simple to run. Take one week of your team’s time and sort it into two columns: hours in which a person decided something, and hours in which a person carried out a known sequence. The ratio appears nowhere in your financial statements, and it is the starting point for sizing any of the plays described above.
If you advise these companies, the same exercise is worth a half day with any client three to five years from a transition. If you allocate capital in this part of the market, add one question to the diligence list: which of this company’s hours follow a pattern, which require judgment, and is the company paying judgment wages for pattern work?
Works Cited
- NFPA 96, Standard for Ventilation Control and Fire Protection of Commercial Cooking Operations | National Fire Protection Association
- Large Firms With at Least 20 Employees Biggest AI Users | U.S. Census Bureau
- The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks | Kathryn Bonney, Cory Breaux, Emin Dinlersoz, Lucia Foster, John Haltiwanger and Aditya Pande, Center for Economic Studies Working Paper CES-26-25, U.S. Census Bureau, April 2026
- Business Trends and Outlook Survey | U.S. Census Bureau