Most pages that rank for "lower middle market private equity" answer a definition question. The answer is always a revenue band, sometimes an EBITDA band, occasionally a fund-size band, and a paragraph about how fragmented and founder-led the targets are. None of that is wrong. None of it is useful to somebody already running a $200M to $800M fund and wondering why the deal flow looks the way it does this quarter.
We build the data layer that sits underneath LMM sourcing: scrapers, scoring, the daily refresh. From the engineering vantage, what's interesting about LMM is mechanical. The targets are usually founder-owned, often weak on reporting, and rarely on a banker's existing rolodex. How does anybody keep the funnel full?
The target firm at $5M-$15M EBITDA
Up at $50M of EBITDA, the seller has a CFO, an audited three-year history, a banker on retainer, and a process. The information asymmetry is small. The work is winning the process.
Drop a tier. At $5M to $15M EBITDA the seller is the founder, the books are reviewed at best, and the "process" is often two phone calls and a regional accountant. Many of these companies have never been contacted by a sponsor at all. The asymmetry runs the other way from upper-middle: the firm can know more about the operator than the operator knows about the firm. The bottleneck is finding them.
That is why the upper-middle playbook breaks when it's lifted down a tier. Bankers cover the upper tier while direct sourcing covers LMM, and the two motions look similar in a pitch deck but run on completely different plumbing.
LMM funnel math
An LMM sponsor closes maybe two to five platforms per fund cycle. To get there, the funnel has to support a lot of upstream activity: thousands of operators surfaced, hundreds touched, dozens in real conversation, a handful under LOI, and the platforms that close. The drop-off from "operators identified" to "platforms closed" is steep — several stages of attrition, each one pruning the population by an order of magnitude or more.
That shape sets the infrastructure requirement before anyone hires a banker.
Intermediated deal flow does not cover it. The serious LMM banks, regional accountants, and broker networks together produce a meaningful share of the at-bats, but nothing close to all of them. The rest is on the firm. In practice that means somebody is identifying operators, pulling the signal that says now versus next year, and starting the conversation before anyone else does. That somebody is usually one or two people on a sourcing team sized for a fund half the current vintage. The math doesn't close.
The data layer for LMM sourcing
The LMM default is a firm paying real money for vendor data and the people to run it, and still not getting a clean list out the door weekly. The vendors built their products for upper-middle, where the filings are clean, the CRM data is decent, and the three-year financials exist. The LMM target is exactly where their inputs are weakest, which is exactly why a sourcing layer built for LMM looks different from what the vendors sell.
Three things have to be true for the funnel to keep moving.
First, the operator universe has to be built, not licensed. There is no canonical list of $5M to $15M EBITDA companies in HVAC, specialty chemicals, ag services, or any other LMM-heavy vertical. You assemble it from contractor registries, BBB filings, state corporate records, LinkedIn, web-stack signals, and a handful of vertical-specific sources. That assembly is the work. The population always comes out smaller than people guessed and dirtier than the vendors implied.
Second, the freshness has to be daily, not quarterly. Founder-led businesses change faster than their public footprint does. A new senior hire, a permit filing, a website migration, an exit by a competitor down the road: those are leading indicators that the conversation is worth having this month rather than in 2027. A quarterly refresh misses the window.
Third, the firm has to own it. A vendor list of LMM operators starts depreciating the day the contract begins. Six months in, the contacts have rotted, the qualification criteria have drifted, and the firm is paying again for the same rows. A pipeline the firm owns runs the other direction: every campaign leaves behind more verified contacts, more disqualification reasons captured cleanly, and a slightly better ranking function than the campaign before it.
Banker incentives in LMM
Bankers in this segment do real work. They also have to clear their own number, which means they prioritize the sellers closest to running a process. A first-call-with-the-founder conversation eighteen months before the founder is ready is not the banker's economics. It is the sponsor's economics, because the sponsor is buying the relationship at a discount to where a process would price it.
So the sourcing function in LMM isn't really a cost center the firm should be trying to minimize. It is the part of the investment process that produces the differentiated entry price. Firms that treat it as overhead pay for it twice: once in vendor fees and analyst headcount, and again at close in multiples bid up by a competitive process they could have avoided.
How the vendor stack accretes
The funnel at LMM is steep and the bandwidth is one or two analysts. So firms buy. They license a data sub to widen the top, then a second sub when the first one misses the verticals they care about, then an enrichment vendor to fill the contact gaps, then a sequencer, then a contractor in another time zone to actually run the lists. None of those decisions is wrong on its own. Each one is a rational fix to the bandwidth problem in front of the analyst that week. The drift is what gets expensive.
Eighteen months in, the firm is paying six figures a year across the stack, the analyst is still running campaigns one at a time, and the LMM-specific inputs the vendors were always weakest on are still weak. The system that would actually fix it costs less to build than the firm is paying to rent the version that doesn't. Four to eight weeks of build, the firm owns the code and the data, and the throughput a single analyst can sustain goes up. Almost nobody gets there by buying more of what they already have.
