AI-Assisted Private Equity Investing

FinTech · Expert Skepticism & Trust

FinTech · AI-Assisted Private Equity Investing · 2025

Role
Lead Product Designer
Surface
AI for private-equity investing
Screening · n=42 · 90-day window
−60% time
Status
Shipped · under NDA
Reconstruction — anonymized, rebuilt from memory for illustration · client under NDA
Deal Atlas-7 Risk 62 Medium confidence
Q3 board pack · p.12

Top customer is 38% of recurring revenue — concentration pushes the score up.

Weight 3
Audited statements · FY24

Margins hold steady across three years; working capital reads clean.

Weight 2
Founder interview · note 07

Churn figure couldn't be matched against the raw export.

“I'm not sure about this one” — low confidence, kept visible

Weight 1

Disagreement logged — training signal

Interactive · data shown is illustrative, not client data

I held the release until the model could defend its own scores. Screening ran 60% faster once it could. Client is under NDA — I'll walk through the methodology on a call.

PE analysts are paid to be suspicious of confident numbers. Doubt is the job, and they're good at it. A score they couldn't cross-examine wasn't a tool — it was a liability with a UI, and they treated it like one.

The model scored deal risk accurately. That was never the bottleneck. The bottleneck was that every score read the same on screen whether the model was ninety percent confident or fifty percent guessing. With no way to check the reasoning, analysts built parallel scorecards and let the AI rank as a sanity check at best.

So I built an "explain this score" surface. An analyst could pull any rating apart into the signals behind it, challenge the weighting, and watch the score answer back. I held the release until the model could survive that. Three things were non-negotiable: the primary sources behind each number, a visible "I'm not sure about this one" state for low-confidence cases, and a logged override when the analyst disagreed. The override drew the longest argument — logging disagreement felt exposing. What it actually built was the training signal that sharpened the model over time.

Interrogation turned into reliance. Screening ran 60% faster across 42 deals in a 90-day window, and analysts went from ignoring the model to leading their deal memos with it. The model didn't get more accurate — they could finally see where its confidence came from.

Diagram: a deal-risk score with an explain-this-score panel, traced back to its three source documents.
The argument arrives before the verdict — every score opens onto its three sources

Want the full version of this work?

Hiring? In an interview I'll walk you through the decisions, the artifacts I can't host publicly, and the numbers — under mutual NDA.

Send me the role

Design Patterns Used in This Case

This project produced two patterns I now reach for whenever AI meets domain experts:

  • ML Explainability Patterns: How to expose a reasoning chain — from the steps the model took to how this deal stacks against its peers — so a non-technical expert can read the why, not just the verdict.
  • Human-in-Loop Patterns: The override-and-explain loop, the visible weighting on each risk factor, and disagreement handled with respect — the moves that keep the expert in command.

I write about this on The Trust Layer ↗

Read 05 / OrgOS → Send me the role ↗