AI-assisted deal screening, live in 11 weeks.

Scope

Replace manual first-pass underwriting and deal screening with an AI workflow that ingests inbound deal materials, calculates the standard sensitivities, scores against the firm's investment mandate, and routes a one-page memo to the right partner — with full human approval before any decision.

• Inbound deal flow had grown faster than analyst capacity. The team was screening maybe 40% of incoming opportunities and triaging the rest by gut feel from the subject line.
• Every analyst built their first-pass model slightly differently. Comparing deals across analysts required senior partners to mentally re-normalize the work, which they often did not have time for.
• Institutional knowledge about why prior deals were passed lived in email threads and one partner's memory, not in any system the team could query.

• Mapped the existing deal-screening workflow with the partners and two analysts. Identified the four steps that ate the most time: document intake, sensitivity modeling, mandate-fit scoring, and memo writing.
• Designed a workflow spec with explicit approval gates: AI does the parsing, modeling, and first-draft memo; the analyst reviews and edits; a partner approves before anything goes to IC.
• Built against the firm's real historical deals as the test set. Tuned the mandate-fit scoring against 18 months of prior pass/proceed decisions until partners agreed with the routing.
• Deployed in stages — first as a parallel system that analysts could compare to their own work, then as the default with the analyst in the loop, then with auto-routing for clear-cut declines.

Measured after the first 90 days.