Short answer
Deterministic deal screening means the AI applies the same written mandate rules to every company and records why each one passed or failed. A deal team gets a consistent, auditable list, not a pile of chat answers that change with each prompt. Lattice Partners is an AI implementation partner based in Los Angeles and New York that helps companies build an AI strategy, trains their teams on AI tools, and builds custom agents and software, including screening tools for private equity (PE) firms.
- The mandate is written down as explicit criteria, so the rules are the same on every run.
- Each call comes with a short reason, which lets an analyst check it in seconds.
- Edge cases go to a human queue, and the analyst's judgment stays in charge.
What is deterministic screening?
Deterministic screening is a method where fixed criteria decide each result, so the same company gets the same answer every time. The AI reads company data and applies the rules. It doesn't improvise a new standard on each run.
A plain chat prompt such as "is this company a fit for us?" can return different answers on different days. It rarely explains the standard it used. Deterministic screening removes that problem. The criteria are stored, every call is logged with its reason, and a rerun on the same data gives the same list. The glossary has a short definition.
How does a PE firm turn its mandate into screening rules?
A PE firm turns its mandate into screening rules by writing each criterion as a clear test with a pass, fail, or review result. Writing the criteria takes more thought than the AI does, and it's where most of the value comes from.
| Criterion | Example rule format | Result types |
|---|---|---|
| Revenue band | Revenue between a low and a high figure the firm sets | Pass, fail, or review if revenue is unknown |
| Sector | Company description matches a defined sector list | Pass, fail, or review if the description is unclear |
| Geography | Headquarters or main operations in approved regions | Pass or fail |
| Ownership | Founder-owned, family-owned, or sponsor-backed, per the mandate | Pass, fail, or review |
| Hard exclusions | Industries or business models the firm never buys | Fail, with the matched reason |
Firms often find their mandate has gaps when they write it this way. Two partners may define a sector differently, or an exclusion may rest on an unwritten habit. Closing those gaps is a benefit in itself.
How does the AI score each company?
The AI scores each company against every criterion using its description and available data, then writes a short reason for each score. It reads text the way an analyst would, and it applies the rules the same way every time.
For example, the tool reads a company description, decides which sector definition it matches, and records the sentence that supports the call. If revenue is missing, it marks the criterion "review" instead of guessing. The output is a table with one row per company, one column per criterion, and a reason next to each result.
Where the data comes from matters. PitchBook says its API lets customers call specific PitchBook data into internal business systems on demand (PitchBook). Crunchbase's API documentation describes read-only access to the data behind crunchbase.com, with access tied to the license tier (Crunchbase). Check each provider's license before connecting anything. MCP for Private Equity Finance Teams covers how these connections work.
Where do analysts step in?
Analysts review the companies that pass, the companies marked "review", and any call they disagree with. The tool narrows a long list so analyst time goes to the companies where judgment matters.
If an analyst disagrees with a call, they change the rule behind it and rerun the whole list. That's an advantage over a chat answer, because the fix applies to every company, not just the one in front of them. Investment decisions stay with the partners, which AI for Private Equity Firms explains in more detail.
What goes wrong with AI screening?
The common failures are vague criteria, missing data, and a rule set nobody maintains. Each has a fix.
- Vague criteria. "Good management team" can't be tested. Replace it with something the data can show, or move it to the human review step.
- Missing data. Mark the criterion "review" and keep the company in the queue. Don't let the tool fill the gap with a guess.
- Stale rules. The mandate changes. Assign an owner, and rerun the list when a rule changes.
- Over-trusting the output. Spot-check a sample of passes and fails each cycle against an analyst's own call.
Pilots that skip these steps tend to stall, which Why AI Pilots Stall explains.
How Lattice Partners does this
Lattice Partners builds deterministic screening tools for mid-market PE firms. In Learn, we sit with the deal team to write the mandate as explicit criteria and test it on companies the team already knows. If the mandate isn't clear enough to screen, we help tighten it first.
In Tailor and Build and Ship, we connect the firm's CRM and licensed data sources and build the review queue. When the project is well scoped, the first version usually ships within a few weeks, with hands-on training for the analysts who use it. Most products go through 2 to 3 iterations as the team refines the rules.
Custom software is a one-time build and the firm owns the code. We don't charge a recurring per-seat or per-module license.
Frequently asked questions
What does deterministic mean in AI deal screening?
It means fixed rules decide each result, so the same company gets the same answer every run. The AI applies the written criteria and records a reason for each call.
Can AI replace an analyst in deal screening?
No. The tool sorts a long list against written rules, and analysts review the companies that pass and the edge cases. Partners make the investment decisions.
What data does a screening tool need?
It needs the written mandate, a company description for each company, and the financial fields the rules use, such as revenue. Data usually comes from the firm's CRM and licensed providers.
How is this different from asking ChatGPT or Claude about each company?
A chat prompt can give different answers on different days and doesn't store the standard it used. A screening tool stores the rules and the reasons, so results are consistent and auditable.
How long does it take to build a screening tool?
A focused first version usually ships in a few weeks. Writing the mandate as clear criteria often takes as long as the build, so start that step early.
Related reading
About Lattice Partners
Lattice Partners is an AI implementation partner based in Los Angeles and New York. We help companies build an AI strategy, train their employees on AI tools like ChatGPT and Claude, and build custom AI agents and software. Clients own the code we build. Lattice Partners was founded by Colin McDonnell, Martin Anquetil, and Christian Yeghnazar. Contact us to talk about your use case.