In the P&L of the portfolio companies, and only where someone can prove it.Most sponsors now have an AI line in their value creation plans, but too little of it reaches profit. We'd start with one company, size the use cases in pounds, pick the few that move margin and fund those first. Then measure them against a baseline the deal team agrees to, so the gain still stands up when a buyer checks it in diligence. Portfolio-wide programmes help later, once one company has shown the number.
Share of executives who say AI has not yet boosted productivity, Atlanta Federal Reserve study
fortune.comSize of the Ode AI services joint venture between Anthropic and Wall Street firms
wsj.comMargin over Euribor on European mid-market software direct lending deals, up from about E+475
pitchbook.com1 briefing this month, with 1 new figure that passed our source checks.
subscribers read every briefing in full, and get each one on WhatsApp. subscribe free · how we research and check sources
| element | what good looks like |
|---|---|
| Value | Sized in pounds of EBITDA, use case by use case |
| Owner | A named person in the portfolio company |
| Phasing | Value by year across the hold |
| Cost | One-off and running costs stated |
| Evidence | Measured against an agreed baseline, ready for diligence |
AI value in private equity shows up in EBITDA, not in the number of pilots. Every use case should be sized in pounds, owned by a named person in the portfolio company and tracked against a baseline the investments team outlines in the value creation plan.
That value has to last until exit and survive diligence. A buyer will test the AI story at exit, so the evidence trail matters as much as the gain. We help define it, deliver it, and measure it.
how we help: private equity. →our frameworks: the Lumo method · the value framework
In two places: inside portfolio companies, to lift margin, revenue and cash, and at fund level, in deal sourcing, due diligence and portfolio monitoring. The portfolio side is where most of the value sits, and where most plans under-deliver.
As named use cases, each with a value in pounds, an owner, phasing by year, the one-off cost and the measure that proves it. A single line for AI efficiencies is hard to deliver and harder to defend at exit.
It can, when a buyer can see the gain in the numbers and trust how it was measured. Claims without an evidence trail tend to be discounted in diligence, and a business model that AI could undercut can weigh on the price.
Both have a place. A central team sets standards, shares platforms and buys tools once, but the value is still created company by company, by changing how each one works. Many mid-market sponsors start with one company and build the central model from what works.