How the engine works
How the origination engine actually works.
Everyone claims AI now. Most are lying, and the rest point an unsupervised tool at a list and hope. Here is the real operating system, layer by layer, and the honest version of what AI does and what it cannot. AI does the work that scales. Operators do the work that matters.
One mandate, end to end
From a thesis to a founder conversation
A thesis enters on the left. It runs through five engine layers, each one either AI working at scale or an experienced operator applying judgment. Measurable outputs fan out on the right. The diagram below traces a single mandate through the whole machine.
Counts are a representative steady state for a single mandate. Real client numbers appear under Results.
The five layers, in plain terms
No layer is magic on its own. The system works because each one feeds the next, and because a human sits at the point where judgment decides whether a founder takes the call.
-
01 Data layer AI
16+ databases, custom scraping, and validation
We combine more than sixteen databases with our own web scraping, then add enrichment and validation on top. A single subscription goes stale and only covers part of the market. Breadth is what makes the map accurate and complete.
The output is a full universe of companies that fit your thesis, with verified owner and decision-maker data attached. You receive that map as an owned asset: your market, documented, not a list rented from one vendor.
-
02 Thesis-fit scoring AI
Your thesis, made machine-readable, 0 to 100
A proprietary model scores every company from 0 to 100 against 50+ signals: sector and sub-sector, size proxies, ownership profile, end markets, competitive position, and dozens more. Your investment thesis stops being a paragraph and becomes targeting logic the engine can act on.
Scoring is what lets us prioritise. The market map is large; the score decides where attention goes first and which segments to test against each other.
-
03 Trigger detection AI
Continuous monitoring for the events that mean now
Fit tells you who belongs in the universe. Triggers tell you who is ready to talk. We monitor continuously for ownership change, succession signals, leadership hiring, funding events, and growth inflections, the moments that move an owner from theoretical to reachable.
This is also where the honest limit lives: readiness is rarely fully visible online. So detection narrows the field, but it never replaces volume and relationship-building, which is how you reach the owner whose trigger never showed up in any feed.
-
04 Agentic outreach Operator-led
Drafted at scale, checked by experienced operators
Our system drafts and personalises outreach using everything the layers above know about a company. Then an operator with deal experience reviews every founder-facing message before it sends, and handles the replies. This is the guardrail an autonomous AI SDR does not have.
It is why the outreach reads like an investor writing to a founder rather than a broker blast, and why a founder takes the call. AI for the volume; judgment at the moment that decides the relationship.
-
05 Deliverability Infrastructure
The invisible layer that makes messages arrive
Dedicated domains, warmed inboxes, and deliverability monitoring at scale are the unglamorous infrastructure a thin AI tool never builds. Without it, even perfect outreach lands in spam and the whole engine produces nothing.
Alongside it sits the visibility layer: client portals and dashboards that show the mapped market, the live pipeline, and every conversation, so the firm owns the asset rather than renting a black box.
AI does the work that scales. Operators do the work that matters. Most vendors only have one of those, and it shows.
What this is not
-
Not
An AI SDR
An autonomous model pointed at a list, sending unsupervised, is what founders learn to smell and ignore. We use AI for scale and put an experienced operator in front of every founder-facing moment. Judgment is not optional at the points that decide a relationship.
-
Not
A database subscription
A list is a starting point, not a system. The data layer is one of five, and on its own it produces nothing. The value is the engine that scores, monitors, reaches out, and is delivered, plus the operators who run it.
-
Not
Ten outreaches a week
You cannot find readiness online, so precision-only theatre misses the deals that were never visible. Real coverage needs volume and relationship-building, so that when an owner becomes ready, you are already the name they know.
Questions on how the engine works
How does AI deal sourcing actually work?
It works in layers. First a data layer combines 16+ databases with custom web scraping and enrichment to map the full universe of companies that fit a thesis. A scoring model then rates each company from 0 to 100 against 50+ signals, turning the thesis into machine-readable targeting. Continuous trigger detection watches for ownership, succession, hiring, funding, and growth events that mean an owner is ready to talk now. Our system drafts and personalises outreach, and experienced operators review every founder-facing message before it sends. Dedicated deliverability infrastructure makes sure those messages actually arrive. AI does the work that scales; operators do the work that matters.
Is this just an AI SDR?
No. An AI SDR points an autonomous model at a list and lets it send unsupervised. That produces plausible-sounding outreach that founders can smell, and it never builds the deliverability layer that makes messages arrive. We use AI for the work that scales, mapping, scoring, monitoring, and drafting, and put experienced operators in front of every founder-facing moment. The difference is judgment at the points where judgment decides whether a founder takes the call.
Why not just target a small list of perfect-fit companies?
Because you cannot find readiness online. Whether an owner is actually ready to sell, who really decides, and what is happening inside the business is rarely in any database. A thin, precision-only list of ten outreaches a week misses the deals that were never visible to begin with. You need genuine volume to cover the market and the relationship-building that means when an owner does become ready, you are already the name they know.
What data sources does the engine use?
The data layer combines 16+ databases with our own custom web scraping, then adds enrichment and validation on top. That breadth matters because a single subscription goes stale and covers only part of the market. The result is a market map that is accurate and complete enough to score and monitor every company that fits the thesis, not just the ones one vendor happens to list.
Do experienced operators really check every message?
Yes. The system drafts and personalises at scale, but operators with deal experience review founder-facing outreach before it reaches anyone, and they handle the replies. This is the guardrail that separates institutional-grade origination from automated noise. It is also why our outreach reads like an investor writing to a founder rather than a broker blast.
See the engine run on your thesis
Thirty minutes on your mandate, your current origination coverage, and what this system would map for your market. If we are not confident it fits, we will say so.
Confidential, and handled by the team that would run your mandate. Or read the client results first.