Deal sourcing and screening for SaaS is the work of moving from a universe of thousands of software companies to a short list you would actually underwrite, and then proving each name's reported numbers are real before a partner spends time on it. The first half is a search problem. The second half is a judgment problem. Almost every tool on the market solves the first half and quietly hands you the second.

Here is the contestable claim this page is built on: in software, finding targets stopped being the constraint years ago. The constraint is screening at the speed of your funnel without a senior person reading every CIM and every data tape, and that screening judgment is the part nobody has actually automated. If your last sourcing tool gave you more names and your team is more underwater, you bought a bigger search and skipped the screen.

This page is written first for the PE operating partner who owns deal-flow throughput across a portfolio, and then for the corp-dev-minded CEO running programmatic acquisitions inside a platform. The CFO's question about predictability and risk gets a direct answer near the end.

Why SaaS sourcing breaks at the screen, not the search

Walk the actual chain. A name comes in, from a banker, a scraper, an inbound, or a proprietary outreach list. An analyst opens a data tape that is formatted differently than the last twelve. They rebuild the same model: normalize the revenue, split recurring from one-time, pull retention cohorts, check concentration, sanity-check the growth story. Thirty to fifty manual steps, most of them copy-paste, before anyone senior decides whether the company is even worth a call.

That chain is where the program actually dies. Not because the model is wrong, but because the most junior person in the room is making the most consequential read in the room: is this ARR real. When throughput matters, the analyst either rushes (and a bad number advances a bad deal, or kills a good one) or the funnel backs up and proprietary names go stale while a competitor moves. Both failures are expensive and neither is a tooling problem.

The sourcing software vendors are not lying when they say AI finds more targets. They are just solving the cheap half. The names were never scarce. The reliable, repeatable screen, run the same way on the hundredth company as on the first, is the scarce thing. The tool is maybe five percent of that gap. The other ninety-five is owning the judgment at volume.

What a real SaaS screen actually checks (the metrics that lie)

Software screening is specific in a way that generic deal sourcing advice never captures, because the headline numbers in software are stories, not facts, until you decompose them.

Start with ARR, the number every seller leads with and the number that lies most often. "ARR" can mean contracted ARR, last-month-annualized run-rate, or a blended figure that quietly includes implementation, services, and one-time fees. A clean screen separates true recurring subscription revenue from services and usage volatility before it computes a single multiple. A platform reporting strong ARR growth on the back of services revenue is not the asset the teaser describes.

Then retention, where the difference between two similar-sounding metrics decides the thesis. Gross revenue retention tells you how much you keep before upsell. Net revenue retention tells you the story after expansion. A company with net retention above one hundred percent and gross retention in the low eighties is leaking customers and masking it with land-and-expand on the survivors. That gap is the whole quality-of-revenue question, and it is invisible if you only read the net number.

The rest of the screen is concentration and durability: logo concentration (does one whale carry the book), cohort decay (are recent cohorts retaining worse than older ones, the early sign of a deteriorating product), CAC payback computed on gross margin not on revenue, and the spread between billings, bookings, and recognized revenue that tells you whether multi-year deals are propping up a single period. On the technical side, the screen flags founder-engineer dependency, monolith risk, and open-source license exposure, because a clean P&L sitting on unmaintainable code is a different deal than it looks.

What good looks like: a one-page screen memo per target that states normalized recurring ARR, gross and net retention with the gap called out, concentration, cohort trend, and the two or three things that would kill the thesis, produced fast enough that a partner reads it the day the name arrives, not the week after.

The sourcing-to-screen sequence that scales without more analysts

The move is not "buy a sourcing platform." The move is to put the hands-on-keyboard steps on agents and keep your experts on the judgment calls, in this order:

First, fix data coverage before you fix speed. The reason screens are inconsistent is that the inputs are inconsistent: every data tape, CRM export, and billing extract arrives in a different shape. The first job is normalizing ingestion so the same fields mean the same thing every time. On an adjacent problem, a PE-owned software platform we worked on moved data coverage from 53 percent to 81 percent before any downstream automation was trusted, because automation on incomplete data just produces wrong answers faster.

Second, automate the decomposition, not the decision. Agents pull and normalize the revenue, split recurring from services, build the cohort and retention tables, and compute the ratios the same way every time. The senior human reads the result and decides. This is the line most pilots cross in the wrong direction: they try to automate the go/no-go and then nobody trusts it. Automate the arithmetic, keep the human on the call.

Third, generate the screen memo as an artifact, not a dashboard. A dashboard is something someone has to go look at. A memo is something that lands in front of a partner. As a reference point for what is possible, a research firm we worked with turns a full research document into a client-ready deck in about 25 minutes at over 90 percent accuracy. The screening memo is the same shape of problem: structured inputs, a known output format, an expert reviewing rather than building.

The outcome of this sequence is throughput without headcount. The analyst stops rebuilding the same model and starts reviewing forty normalized screens instead of building eight from scratch. The expensive six-figure misread on a checkbox stops being one tired junior's responsibility at 11pm.

Is this predictable, and where is the risk?

The honest risk in this work is not the model. It is the last mile and the data. A screening engine that produces a confident wrong number because it ate a dirty data tape is worse than no engine, because someone will trust it. So the discipline is to gate on coverage first, run every change against a real copy of your actual deal data, and replay the screen against deals you already underwrote to confirm the engine reaches the conclusions your best people reached. That replay against known outcomes is what makes the result predictable rather than hopeful. If it cannot reproduce your team's past good calls, it does not ship.

The other real risk is the internal-build trap every operator has lived: you hire a sharp person to build the screening process, they map it, restructure it, and leave before the fix actually ships. The fix has to be owned end to end by someone accountable for it shipping, and it has to be owned by you afterward. A screening engine that lives in one departed analyst's head or one vendor's black box is not an asset.

How Salfati Group would approach this

We would scope this as a Mandate: a fixed-price, KPI-anchored outcome, screening throughput and screen accuracy against your own past calls, with a named architect owning it end to end. Agents do the ingestion, normalization, and memo generation; your senior people own every go/no-go; and you own the system that ships, not a license to someone else's. It is backed by an Outcome SLA, which means if we miss the named KPI we keep working at no additional cost until it ships. If your sourcing problem is really a screening problem, that is the shape of the fix.

Apply for a Mandate

Reviewed by Elon Salfati·15 June 2026