The private equity playbook is a simple one: find what works, measure it, and replicate it across the portfolio.
AI turns out to make the first two of those surprisingly difficult, largely because an operating partner is accountable for it across a set of companies they do not run day to day.
The pattern across private equity at the moment is that each portfolio company runs its own experiments, buys its own tools on its own cards, and writes its own policy or copies one off the internet, with the result that nobody at the fund can produce a single figure for group AI spend, a single audit trail, or a reliable view of which use cases are earning their keep. Twenty companies generate twenty separate learning curves, and very little of what is working or failing in one of them ever reaches the other nineteen.
The research points the same way. FTI’s 2026 Private Equity AI Radar found most fund and operating leaders reporting positive financial impact from AI in their portfolio companies, while adoption remained uneven and the distance between an isolated success and a portfolio-scale advantage stayed wide. Bain puts roughly a fifth of portfolio companies at the point of having operationalised generative AI with measurable results, and McKinsey found 70% of GPs expecting high AI impact within three to five years against the 6% who see it today.
Just over half of limited partners now rank a GP’s AI value-creation strategy among their top five manager-selection criteria, so most funds have already settled the question of whether to run a portfolio AI programme and are now working out what order to run it in.
Order of operations
Most programmes begin with a use case, usually in the largest portfolio company, on the assumption that governance will catch up later. We think that sequence is backwards, for three reasons.
First, oversight gets cheaper as you add entities. A use case built for a manufacturer does not transfer to an insurance broker. A policy engine, audit trail and cost attribution do.
Second, the control plane creates the baseline for value. Before you can credibly claim AI value creation to an LP or buyer, you need to know what is being spent, by whom, on which models and for which workloads. Almost nobody has that baseline. Unattributed AI spend is one of the most common finance complaints we hear, and it makes ROI impossible to prove.
Third, oversight makes rollout easier. A portfolio CEO asked to adopt a fund-mandated AI tool will push back. Give that same CEO visibility into what their teams are already spending, plus a way to keep client data out of public models, and you have a much easier conversation. Oversight is the logical place to start because it serves the portfolio company as much as the fund.
Oversight is the logical place to start because it serves the portfolio company as much as the fund.
What the group layer actually gives you
The group layer gives the fund one consistent policy, with entity-level exceptions where regulation requires them; one audit trail showing who used which model, on what data, and whether the request was allowed; and one cost view that rolls from the individual user to the portfolio and fund.
That answers the diligence question, the LP question and the CFO question from the same system. That is why we would put the control layer in place before building anything on top of it.
Part two is about what you put on top of it, and what the whole thing is worth at exit.
MisaLabs builds the enterprise AI control plane. Govern AI usage. Prove the ROI. In your environment.
Talk to us →