Skip to content

Case study

Private Equity & Portfolio Value Creation

97%

agreement with blind partner review on live deal flow

97% agreement with blind partner review, trained on 32,000 deals, validated across a six-week rollout

A global private equity firm had two decades of deal history, 32,000 logged deal records: screening records on every logged opportunity, plus memos, data-room exports, and post-mortems on the deals that went deep. We trained a model on all of it. By the end of the six-week rollout, it was matching partner-level judgment on 97% of assessments across live deal flow.

Situation

The situation

The firm screens thousands of opportunities a year across offices in three regions. Screening quality depended on who caught the deal: a senior associate in one office would kill a target that another office would take to IC. Analysts spent the first week of every diligence sprint re-deriving risk factors the firm had already paid to learn across 32,000 prior deals.

The cost showed up twice: in analyst hours, and in the occasional deal that reached IC carrying a risk the archive already knew about.

The build

What we built

Three connected systems, trained on the firm's 32,000-deal history and calibrated against how its investment committees had actually graded risk over two decades.

  • Compliance assessment

    A model that reads target documentation and flags exposure against the regulatory and contractual patterns that had hurt the firm's past deals, not a generic checklist.

  • Risk analysis

    Scoring across financial, operational, and counterparty dimensions, benchmarked to the firm's own IC decisions.

  • Deal scoring

    A ranking layer that compares each new target against the closed record, what the firm bid, what it passed on, and how those calls aged.

Deployment

How it deployed inside the workflow

No new interface to log into. Scores and flags land in the pipeline system the deal teams already use, and every flag links back to its source documents, so an associate can verify a finding in minutes rather than taking the model's word for it. Analysts remain the decision-makers; the model does the first read.

Rollout ran office by office over six weeks, with blind partner-review sampling to establish trust before anyone was asked to rely on the output.

Results

Results

Agreement was measured the hard way: model assessments compared against blind partner review on live deal flow, not a held-out test set. Screening that consumed an analyst's first week now takes an afternoon of verification. The firm estimates roughly 12,000 analyst hours a year returned to the deal teams, capacity that went into deeper work on the deals worth it rather than headcount reduction.

97%
agreement with blind partner review on live deal flow
6 weeks
from first office deployment to full three-region rollout, validated by blind partner review
32,000
historical deals in the training corpus
~12,000 hours
analyst hours returned to deal teams annually (client estimate)

From the partner who signed off

The model does the first read the way our best associate would on their best day, except it does it on every deal, including the ones that come in at six on a Friday. The IC gets a cleaner risk picture than we have ever had.

Partner, global private equity firm (name withheld)

Discuss a diligence system for your fund

Your deal archive is a training set nobody else has. We will assess what it could do to your screening quality and cycle time, and tell you whether the build is worth it.