AI in Private Equity Due Diligence: What Changes for the Associate

In private equity due diligence, AI now does the first read. At TPG, an assistant built by Lab 39, the firm's in-house team of data analysts and scientists, takes in documents and data every week and produces a report for the investment committee to review: the risks, the opportunities and the net takeaways on a deal. The investment decision stays with the people in the room.

For the associate, who used to do that first read by hand, the job moves from reading and summarising to checking and judging. Each finding the machine surfaces has to be traced to its source, tested in the model and defended in front of the committee. That shift is what the job now rewards, and it can be practised long before an interview.

2019

TPG sets up Lab 39

its in-house team of data analysts and scientists

Minutes

Blackstone deal-document review

once a weekend-long job, its CTO said in May 2026

15,000

Software purchase agreements analysed in minutes

by an Apollo AI system, across its portfolio, June 2025

What does a private equity associate do in due diligence?

Due diligence is the stretch between a firm deciding a company deserves a serious look and its investment committee approving a binding offer. The firm pays lawyers, accountants and consultants to dig. The associate turns what they find into a single case the committee can vote on.

The data room

A sale usually opens with a confidential information memorandum, or CIM, the case for the business that the seller's bankers put together. Then comes a virtual data room: the company's contracts, accounts, customer data, board minutes and litigation files. The associate reads the CIM, builds the valuation or buyout model and researches whether the company can grow, while coordinating the outside advisors and checking the analysts' work.

The workstreams

In practice the work runs in parallel streams, most of them staffed by outside advisors:

• Commercial. The market, the customers and the competitors: whether the company can keep growing where it sells.
• Financial. Accountants test the quality of earnings: whether the adjusted profit the seller presents will repeat, and what the net debt and working capital really are.
• Legal. Lawyers read the contracts, leases and litigation for anything that changes the price or ends the deal.
• Management. Meetings, site visits and background checks on the people who will run the company.

The associate sits in the middle of it, on the calls where the lawyers and accountants report what they found, carrying each finding into the model.

The investment committee memo

Everything converges on the investment committee memo, the document the deal team writes to win approval for a binding offer. It pulls the diligence findings, the model and the investment case into one argument, and it is where every number an associate touched has to hold up.

How is AI changing private equity due diligence?

The change is in the order of the work. Reading the data room, once the associate's opening job on every deal, now falls to the machine, and the associate's work starts where the machine stops.

The first read moves to the machine

John Stecher, Blackstone's chief technology officer, put a scale on it in May 2026: "Think about doing a deal: you're looking at thousands of documents, and what used to be a weekend-long activity can now be carried out in minutes."

He was as clear about where the time goes instead: "That frees our teams to focus on judgment: understanding risk, improving structures, and driving better outcomes for our LPs." Blackstone's private equity and real estate teams, he added, use targeted tools to "help move quicker from raw deal documents to financial models."

One week of a deal, with AI in it

Put the capabilities the firms describe in one place, and a week of diligence looks something like this:

1. Monday. The data room opens and the firm's platform reads it. The associate queries it for every customer contract with a change-of-control clause, three years of revenue by customer and every mention of litigation. Each answer points to the page it came from.
2. Tuesday and Wednesday. Figures move from the raw documents into the model, and the associate reconciles each one to the management accounts instead of retyping it.
3. Thursday. The platform compares documents side by side: the CIM against the contracts behind it, this year's accounts against last year's. The associate decides which differences matter.
4. Friday. A report for the committee sets out the week's risks, opportunities and net takeaways. The associate's job is to know which of them are real.

None of those steps removes the associate. Each one moves the effort from producing the first version to deciding whether it is right.

How does TPG use AI in due diligence?

TPG has described its tooling in public, in a December 2025 newsletter built from interviews with its leaders. It shows what a firm builds when it wants AI inside the deal process rather than beside it.

Lab 39 and Samantha

The work started in 2019. Ayanna Clunis, a TPG partner and its head of operations, told the story: "At TPG, we started our AI journey in 2019, when we set up an internal team called Lab 39, which is a group of data analysts and scientists. We knew we had this wonderful set of data that we were sitting on, and didn't have the power to really harness it."

The answer was Samantha, which Clunis called TPG's "agentic platform or orchestrator layer." It sits on top of outside models rather than replacing them: she called it hybrid because it draws on "best-in-class external tools like Grok, OpenAI, Copilot, and others," so TPG's teams can "perform complex research, query information, compare documents, etc."

Teams like Lab 39 are also a way into a buyout firm for engineers and data scientists; the forward deployed engineer in finance guide maps the related engineering seats.

A weekly report for the investment committee

Tim Millikin, a TPG partner who co-leads its software and enterprise technology investing, described what Lab 39 built for the committee: "Our team at Lab 39 has developed a purpose-built TPG AI assistant. This is a tool that will ingest documents and data every week and output a report for the investment committee to review that includes risks, opportunities, and net takeaways around any specific investment opportunity we're assessing."

The committee that reviews the report is the one TPG's annual report describes: senior leaders and investment professionals who first decide whether to give a deal team preliminary approval to keep doing diligence, may meet several times on a single deal, and direct the team on terms, strategy and process. Their review covers the transaction and the investment thesis, the business, the risk factors and diligence issues, and the financial models.

It can also look back, Millikin said, and "normalize, look back at data over the past several years from investment memos, and our own portfolio performance." The decision does not move. "Ultimately, the investment decision remains with the humans in the room," he said, "but it's a very powerful supplement to the discussion."

What TPG asks of its AI

TPG frames the whole programme around people. Paul Daugherty, who joined TPG as its AI advisory chair after 38 years at Accenture, said "the best path to AI success is a human plus machine approach." TPG's July 2025 newsletter put the limit plainly: "It's not a replacement for people and decision-making, but it's an accelerator for idea generation."

Its annual report names the risks. Using AI, it says, brings "the risk that confidential information may be stolen, misappropriated or disclosed" and the risk of relying on "incorrect, unclear or biased outputs." TPG's test for the AI it uses meets both: it has to be secure, "tested, accurate, and compliant," and transparent, so the firm can "understand why AI made a certain decision."

Which other private equity firms use AI in diligence?

Several of the largest firms describe a similar setup: a central team working alongside the deal teams that use its tools.

Apollo's contract system shows the flow running the other way. Prices learned inside the companies it owns give Apollo, in MIT Sloan Management Review's words, "its own proprietary benchmark that it can use during the diligence process for future investments."

Where AI goes wrong in diligence

Whoever builds the tool, it fails in diligence in two familiar ways, with far more at stake than in a chat window: it cannot see what is missing, and it can sound right when it is wrong.

What the data room leaves out

A model can only read what the seller uploaded. The document that decides a deal can be the one that is not there: the side letter, the customer who has already given notice, the claim still in a lawyer's drafts. A fast, fluent summary of an incomplete data room is still incomplete, and it can read as if nothing is missing.

Confident, cited and wrong

A summary that cites a page is easier to trust and easier to check. The citation proves where a sentence came from, not that the sentence is true. Repeating a seller's claim with a page reference is still repeating the seller's argument.

Checking an AI-drafted finding: a worked example

A central exhibit in financial diligence is the bridge from reported profit to the adjusted EBITDA a seller asks the buyer to pay for. The numbers below are invented; the method is the one the financial workstream exists for.

A seller presents a business services company with reported EBITDA of $40.0m and four add-backs worth $8.5m, for adjusted EBITDA of $48.5m. The firm's AI tool reads the seller's documents and labels all four add-backs non-recurring, each with a page reference to the CIM. The associate opens the evidence behind each one.

Two of the four add-backs move, and they move the price. At an 11.0x purchase multiple, the seller's $48.5m implies an enterprise value of $533.5m. The checked $44.5m implies $489.5m.

On the seller adjusted EBITDA of $48.5m

$533.5m

After the associate checks, $44.5m

$489.5m

Illustrative. The company, its add-backs and the 11x multiple are invented to show the arithmetic: a $4.0m cut in adjusted EBITDA takes $44.0m off the implied price.

The tool did its job: it found every add-back and pointed to where the seller made each claim. What it could not do was doubt the seller's labels, so it missed that a cost called one-off shows up every year and that a saving built on three months of invoices is still a forecast.

Deloitte's advice to sellers makes the same point from the other side: substantiating each pro forma adjustment is what gives a buyer the confidence to accept it as the basis for a higher bid.

How do private equity firms assess a target's exposure to AI?

Checking the seller's numbers is half of the new work. The other half is a question the committee now expects answered about the company itself: will AI make this business stronger, or take its customers?

TPG has written the question into its paperwork. Peter Munzig, a TPG partner who heads business services for TPG Capital, said the firm has "a standard investment committee template where we spend time talking about the components of AI and how any given business model may be impacted over time." He added: "Those areas where we feel like it will be a beneficiary are where we tend to lean in much more."

Other firms come at it from different angles:

• Apollo starts with the industry. As MIT Sloan Management Review described it in June 2025, Apollo's process begins before any investment with an assessment of how AI is affecting the target's whole industry over the coming years. Detailed diligence on the company's own products and projects follows, often with outside consultants.
• Blackstone looks at the data. "When we diligence software companies, we look at how much proprietary data they have that can't be replicated from what's publicly available. That is a moat," Stecher said.

For the associate, the question turns into work. Someone has to write the section that says how exposed the company is, and back it with evidence: which products rest on data a rival could buy, which customers could do the job themselves with an AI tool, which costs AI could take out.

Will AI replace private equity associates?

The firms building these tools describe them as support for the people who decide, not a substitute for them. On jobs, Stecher did not claim to know where it ends: "There will be real disruption but long term we believe the job impact remains to be seen." Rodney Zemmel, who heads the same firm's operating team, called the data "still noisy," adding that "historically technology has created more jobs than it's destroyed."

Stecher's stated goal is to "eliminate rote work and amplify human insight." For an associate, the rote part is reading everything first and building the first pass of the model. The part the committee relies on, deciding which findings hold, stays with the associate.

The same question for bankers, with what the banks have said on the record, is answered in Will AI Replace Investment Banking Analysts.

How to prepare for AI questions in a private equity interview

The four questions below are not any firm's list. They follow from how the firms describe the work, and a strong answer to each can be built from public material.

Four questions to be ready for

What to do before the interview

Most of the preparation needs nothing more than public documents and an ordinary AI tool:

1. Write a two-page investment memo on a listed company from its annual report, by hand, including a paragraph on its exposure to AI. Then have an AI tool draft the same memo from the same filing and mark every place the two disagree.
2. Take a listed company that reports an adjusted profit figure, rebuild its reconciliation from reported profit, and decide which adjustments recur. It is the worked example above, on real numbers.
3. Before an interview, read the firm's own AI pages and its latest annual report, and note what it says about diligence in its own words.
4. Rehearse the checking habit out loud with the interview readiness tool, and use the guides to learn the tools you are likely to meet on the desk.

The bottom line

AI now does the first read of a private equity deal, from the seller's documents to the first cut of what the committee sees. The judgment has not moved: which risk ends the deal, which add-back is real, how exposed the company is to AI itself. The associate who can check the machine's work, and say exactly why a finding is wrong, is the one the committee still needs.

The same checking runs through the rest of the deal. Before this stage, AI builds and ranks the target list; after closing, it becomes a line in each company's value creation plan.