AI for private equity portfolio companies: a 100-day plan

Start with how little is usually running. In the US Census Bureau's business survey (14 December 2025 to 3 May 2026), 37% of firms with at least 250 employees reported using AI in their business operations, and in the period ending 3 May 2026, 32% of firms with 100 to 249 employees did. Among firms that do use it, a Census working paper found that 57% use it in three or fewer business functions. So a new portfolio company is usually close to the start, which is useful: the first uses can be picked for value instead of inherited from whoever got excited first.

I screen uses against the value creation plan, not against a list of what AI can do. If a use does not move a line the board already tracks, it is a nice-to-have, and nice-to-haves are how a portfolio ends up with a drawer full of pilots and no change in EBITDA. The table is the screen I would run in the first three weeks.

Two uses often look good on a slide and disappoint in the P&L. One is a general assistant switched on for everyone with no training, no workflows and no measure attached. The other is a customer-facing chatbot put in front of a support process nobody has fixed. Both can be worth doing later, done properly. Neither belongs in the first 100 days.

The usual failure has little to do with the technology. A pilot gets chosen because someone senior saw a demonstration, it runs on sample data in a vendor's account, it reports usage instead of money, and nobody in the company owns it once the outside help leaves. Six months later it is still a pilot. Repeat that across a portfolio and the fund has paid for the same lesson several times.

Treat the phases as a shape to hold the work to. Access to systems and data sets the real pace far more often than engineering does, so the access requests go out in the first week. The plan assumes a company that has closed and has a value creation plan. For a company still in diligence, the questions further down this page come first.

Run the screen with the CEO, the CFO and the people who do the work. Sign-off runs in that order and usefulness runs the other way: the people doing the work know where the hours go, and the CFO knows which of those hours show up in the numbers.

Then measure the work as it runs today, on whatever the chosen measure is: time per case, cost per case, error rate or days sales outstanding. Take the baseline from the company's own systems where you can, and have finance agree it in writing. Everything you claim at day 100, and possibly at exit, gets compared with this number.

Send the access requests in the first week. Accounts, data extracts and security reviews take longer than the build, and a plan that asks for them in week four has already lost a month.

Build each use against real data in the company's own environment, and keep the code, prompts and test cases in the company's own repositories and cloud accounts. That matters more in a portfolio than anywhere else, because the company will be sold, and whatever it depends on has to go with it.

Build the controls with the first working version rather than after it: who can use it, what it can read and change, which actions need a person's approval, and a log of what it did. Our write-up of the pattern where the model proposes and a deterministic check decides shows one way to build the approval step.

Then measure against the baseline, on the measure agreed at the start, and take the cost of running each use off its benefit: licences, usage charges and the hours people spend reviewing its output. A use that saves time but costs more than the time it saves is a hobby.

Somebody will test your AI numbers: the investment committee, a lender or, at exit, the buyer's diligence team. Build the measurement to pass that test from the first week, because rebuilding a baseline after the fact is close to impossible.

Be as careful with the story as with the numbers. In March 2024 the SEC announced settled charges against two investment advisers for false and misleading statements about their purported use of AI. In September 2024 the FTC announced Operation AI Comply, five law enforcement actions against deceptive AI claims, and said there is no AI exemption from the laws on the books. An AI claim in a portfolio company's marketing, a fund's investor materials or an exit document needs the same evidence as any other claim. This is not legal advice: have counsel review the claims before they go out.

AI due diligence has two sides for a deal team. One is using AI in the diligence itself, to read a data room faster. That is a tooling decision, and the usual rule holds: a person checks what the model summarised before anyone relies on it. The other side is what AI means for the target, and that is the one that moves value. These are the questions I would add to the diligence list.

A portfolio company of a few hundred people does not need an AI ethics board. It needs a short policy its board adopts, a register of every AI use, a few risk tiers and controls on the risky uses. The fund's job is to write those once and let each company adopt them, instead of paying for the same document in every company.

For US companies, the NIST AI Risk Management Framework makes a sensible common spine. NIST describes it as intended for voluntary use. It has four functions, Govern, Map, Measure and Manage, with Govern running across the other three, and NIST says version 1.0 is being revised as part of the White House AI Action Plan. A shared policy mapped to those four functions gives every company the same structure without pretending they are the same business.

State rules apply company by company, and they are moving. In California, rules on automated decisionmaking technology (ADMT) were approved on 22 September 2025. A business subject to them that uses ADMT to make a significant decision about a consumer must comply with the ADMT requirements by 1 January 2027. The regulation defines a significant decision as “a decision that results in the provision or denial of financial or lending services, housing, education enrollment or opportunities, employment or independent contracting opportunities or compensation, or healthcare services.” The same rules made risk assessments a requirement from 1 January 2026. Ask counsel which state laws reach each company. This is not legal advice.

1AYM has worked with private equity, and the plan above is the shape I would bring to an operating partner. For one portfolio company, the first paid piece is usually our AI Opportunity & Feasibility Sprint: a fixed scope of two to four weeks that produces the scored opportunity map, a current-state analysis of the work and an investment case, which covers the screen in days 1 to 20 of the plan. A fixed-scope build can start within a day of the scope being signed. For a programme across several companies, a retained implementation team owns the shared architecture and builds alongside each company's own people.

Rolling out AI work tools and coding tools across a company, with the training, controls and measurement that make them pay, is core work for us, and our documented rollout across a global agency shows what that involves. If the fund already has a scoped job and needs it staffed, 1AYM can resource it on contract from the collective of associates who work with the firm, held to the same standard. US portfolio companies can contract with 1AYM's US entity. If you are planning the first 100 days for a new platform company, book a call below.