The first page of Google for "proprietary deal flow" splits cleanly into two camps. The vendors say proprietary deal flow is what you get when you pay for their database. The practitioners say it's a relationship game that no software solves, and that "true proprietary" deals barely exist on an exposure basis once a company has been profiled enough times.
We sit in a different seat. We build the data-extraction systems the sourcing teams actually run on, so the argument looks different from the data layer than it does from either side of the SERP.
The system underneath the word
A proprietary deal, in the strict sense the CapTarget piece uses, is one no other buyer has seen. Almost nothing meets that bar. Any LMM operator above five million in revenue has been pinged by someone. The vendors know this and finesse the definition: Grata calls it "deals sourced directly", Falcon River calls it "private and closed", Carta calls it "sourced directly by the firm". None of those definitions say anything about how the firm gets there.
From the systems side, the thing that compounds at LMM scale has four pieces wired together. A signal layer that surfaces a target before the banker does. A scoring layer that decides which of those targets is worth a partner-hour. An outreach cadence the firm runs on its own clock. And a memory layer that captures every "not yet" so the next touch is smarter than the last.
A vendor sells the first piece, badly, and rents a partial second. The other two aren't in the SKU because they cannot be.
What the system looks like in practice
The pieces are familiar individually. Permit-issuance data from county portals. License-renewal filings from state boards. Payroll-tax filings cross-referenced against Secretary of State records. A graph over the firm's portfolio plus second-degree connections. None of those sources are secret, and the vendors index a couple of them in shallow form. The owned version is what happens when the firm runs the join itself, on its own cadence, against its own thesis.
A signal layer in this shape is a query plus plumbing plus a feedback loop. The query says "operators in trade X whose permit velocity has compressed for three straight months against the metro baseline." The plumbing says "pull weekly from the county portals that publish, scrape monthly from the ones that don't, dedupe against the active universe." The feedback loop says "the operators who replied last quarter, and the ones who said 'talk to me in a year,' come back into the scoring weight." A vendor sells the query — sometimes — and never the plumbing or the loop, because both are the firm's intellectual property by the time they've been tuned to a thesis for two cycles.
The other thing worth naming: the universe these queries run against is bigger than the deal universe by orders of magnitude. Across our own operator index of 63,774 small businesses scraped from Google Places — predominantly plumbers, general contractors, roofers, and electricians — only about 24% have had their public-web content actually crawled, and in the subset where the acquisition-signal field is populated, every entry surfaces at least one phrase like "acquired," "merger," "rebranded," or "formerly known as" in an About or ownership-history page. The signal is rare and high-precision when it shows up, not high-coverage. The index gets sharper every campaign as more operators get crawled and more signals get tagged. A rented vendor list does not.
Rented flow versus owned flow
The rented version is what a six-figure PitchBook subscription plus a junior associate running the same filters every Monday plus an SDR shop dialing the export adds up to. The firm calls the output proprietary because it ran the query in-house. The output isn't proprietary in any sense that matters. Any other PitchBook subscriber pointing at the same NAICS and the same headcount band lands on substantially the same operators in the same window. The vendor's incentive is the renewal, not the compounding.
What it actually costs to run
The cost shape comes in two versions. The rented version is what most firms drift into: a handful of vendor subscriptions, one or two enrichment tools, a sequencer, a contractor or two, plus the analyst time to coordinate all of it. A six-figure data subscription is the line item the firm sees on the invoice; the real cost is the headcount stack around it. That stack scales linearly with the number of theses the firm runs.
The build-once version replaces some of the org chart with code. Primary-source queries running on a regular cadence against the same registries the vendors paywall. A scoring service that re-ranks the universe on a schedule the firm controls. A feedback loop that pipes outreach outcomes back into the score. The same plumbing supports the next thesis without a parallel contractor stack.
This isn't the right answer for every firm. Some are better off with a vendor stack and a heavy SDR layer, particularly firms running a single thesis in a sector vendors cover well. The owned system pays back where the firm runs two or more concurrent theses, sources in industries whose meaningful signals aren't in any vendor dataset, and can absorb a build window before the first campaign ships.
The ninety-day mirror
Here's the part most of the SERP misses. Owned flow and rented flow look identical for the first ninety days: both produce a target list, both get a partner on a call, both look like they're working.
The divergence shows up in month four, and it isn't subtle. The rented system restarts from the same baseline every campaign. Same filters, same exports, same set of other subscribers hitting the same operators on the same Tuesday. The owned system has spent ninety days getting sharper: the disqualifications captured last quarter prune next quarter's list, the signals that didn't predict fit get downweighted, the operators who said "talk to me in a year" surface again with the re-touch already scheduled and the prior thread loaded.
At day ninety the two look the same on a slide. At month twelve one of them is producing a target list the firm couldn't have written down on day one, and the other is producing the same export it produced in January. That's the trade. It looks like a tooling choice. It's a compounding choice, and the compounding doesn't bend back once it diverges.
