Last verified 2026-07-12

Deal sourcing & pipeline screening

Can AI find and screen deals? Yes for surfacing and ranking candidates from messy signals a database doesn't hold — before they show up in any vendor feed. No for conviction: the model orders the pipeline, it does not decide what you back.

The pain	A sourcing team wants the right companies in front of it before a competitor gets there. That means watching a firehose — news, hiring pages, product launches, reviews, filings, sector chatter — and matching it against a thesis, then keeping a pipeline of who was seen, scored and passed. Done by hand it is a few analysts reading the internet and maintaining a spreadsheet that is stale the day it is built.
What AI does today	Read the unstructured firehose and surface companies that fit a thesis, rank them against your criteria in natural language, and keep the pipeline updated with fresh signals. The boundary: a database filter answers a structured question deterministically; AI earns its place on the signal that isn't in any field — a hiring surge, a founder's post, a review trend — and on ranking fit in language rather than hard thresholds.
Proof it's real	Production-verified (deployer-confirmed). EQT's Motherbrain has run since 2016 — the firm's own platform combining external and internal data to source and assess opportunities, publicly credited with sourcing multiple completed investments (Peakon among them), with years of independent press coverage. This is the deployer speaking about its own system in production — the strongest evidence class in this library. Newest evidence: 2026.
What it can't do	It cannot form conviction — that is where the human sits: the diligence, the relationship and the price are the dealmaker's work, and EQT itself frames the system as supporting, not replacing, the decision. A model that looks confident about a market it half-understands will surface plausible-but-wrong candidates, and it inherits its data's blind spots: a company invisible to its sources is invisible to it.
The real alternatives	Data vendors + analyst screens — PitchBook/Preqin-class subscriptions filtered by sector, size and growth; the industry default. Wins when your thesis maps to structured criteria; five-figure subscriptions either way. Origination networks and advisors — boutique banks, sector advisors, placement relationships; wins on proprietary access, not coverage. An in-house platform on your own data — the EQT route: an internal build joining licensed feeds to your CRM and pass history. The strongest version and the most committing; only firms treating it as core infrastructure have made it pay. Associates reading the internet — the status quo; flexible, expensive per insight, stale fast. AI sourcing is essentially the last route industrialised — worth it only if non-obvious, early candidates are genuinely your edge.
What you need in place	Your CRM and pass-history as training context (the system learns from what you've seen and declined); licensed data feeds (the signal layer is mostly paid data); and honest ownership — the sourcing lead owns the pipeline, and the firm decides upfront how a model-surfaced lead enters IC materials, so provenance is never fuzzy. In a multi-strategy house, data licences and CRM choices are usually central decisions — check what the centre already pays for before pricing anything new.
Effort & cost	Weekend script — an LLM scoring a batch of company descriptions against a written thesis; free, a fair test of whether language-ranking beats your current screen. Off-the-shelf — sourcing and market-intelligence platforms with signal feeds; five figures a year, comparable to the data subscriptions you likely hold. Real project — an EQT-style internal platform joining external signals to your CRM; a six-figure-plus, multi-year commitment with data engineering. If your deal flow is intermediated (auctions, bankers), sourcing AI solves a problem you don't have.
What to watch	Check what it misses, not just what it ranks. A sourcing tool that only surfaces the obvious names is repackaging a database; the value is the non-obvious candidate found early. And watch for a confident score on a company the underlying data barely covers.

Questions operators ask

Does AI actually find deals, or just rank lists?

There is at least one deployer-confirmed answer: EQT publicly credits its Motherbrain platform with sourcing completed investments over nearly a decade of production use. The honest caveat: that took an in-house platform and a data team, not a subscription — most "AI sourcing" products are closer to smarter list-ranking.

Is AI sourcing better than PitchBook screening?

Different questions. A vendor screen answers structured criteria over covered companies, deterministically. AI adds value on uncovered signals (hiring, product traction, chatter) and thesis-fit in language — if your edge is early, non-obvious names. If your criteria are structured, the screen you already pay for is the right tool.

Changelog

2026-07-12 — published.