Is AI a value creator or a deal risk? What PE investors now test for in diligence 

Leading PE investors on telling the two apart, asset by asset, and the readiness signals that decide it. 

IPEM Community brought together a panel of leading AI voices from Singulier, Apollo, Ardian, Fondo FSI and Goldman Sachs on a question now sitting on every investment committee’s agenda: is AI the next great value creator, or the biggest risk in the deal? The honest answer was both, and the work is in telling them apart asset by asset. The sharpest takeaways are below; the full discussion, including the portfolio examples, is worth watching in the replay. 

AI due diligence is now a test of business quality, not technology 

The narrow technology check is gone. The questions now split two ways: an outside-in view of where the sector is heading and how fast competitors are moving, and an inside-out view of whether this specific company has the leadership ownership, data foundations and operating capability to act. 

The two rarely move together, which is the whole point. A company can sit in a market with a high AI ceiling and still be the wrong bet because its foundations are weak or management hasn’t grasped the scale of change. One panellist walked away from a well-run accounting-services business that was vocal about AI in every meeting, because the strategy wasn’t owned at the top and business-unit leads weren’t involved in it at all. In a sector being reshaped this quickly, you don’t bet on the laggard. 

A consistent framework matters more than any single score 

Every investor in the room recognised the same need: a consistent way to judge an asset’s AI position that holds across very different sectors, so you’re not comparing a software business and a clinic network on instinct. 

This is the logic behind Singulier’s AI due diligence framework, which examines three things in parallel: AI risk and exposure (which parts of the business are vulnerable to disruption), internal readiness (how prepared the asset is to respond across technology, data and organisation), and impact and value-creation opportunity (where the upside sits, quantified in a way that holds through diligence and into post-close delivery).

Keeping the three separate is what makes them useful, because exposure and readiness so rarely line up. 

Real readiness is structural, not a list of tools 

Readiness is the layer most often overstated. It is not a tally of AI tools in use. It is whether the data foundations are good enough to build on, whether AI reaches into how decisions get made rather than sitting beside the workflow, whether it is embedded in the product or pricing, and whether there is real ownership and governance behind it. 

This is why “we use a few AI tools” keeps failing as evidence. Four years into this wave, investors want two or three examples that have scaled and produced EBITDA, not more pilots. The more useful definition of readiness is clarity: knowing where the gaps are and having a resourced plan to close them. 

The strongest value creation redesigns the model, not just the cost base 

Every panellist drew the same line between efficiency plays and reinvention. Out-of-the-box tools that take 5 to 10 per cent out of a process are worth doing, but the work that moves the needle starts from first principles: how would you rebuild this process if you designed it around AI from scratch? 

The session’s portfolio examples made this concrete, from a textbook business that turned an existential threat into an AI tutor built on its own proprietary content, to a sports agency using its untapped data to make its agents harder to compete with, to an audit firm that cut report drafting from hours to seconds and converted the freed capacity straight into revenue. Each one strengthened the core business rather than replacing it.

The full stories, with the numbers, are in the replay. 

How the value shows up: EBITDA now, multiple at exit 

Value lands either as margin (the same people doing more, leaning less on costly contractors) or as revenue (personalising service at a scale that wasn’t possible before, or capturing demand the business couldn’t serve).
One panellist’s longer-term point is worth holding onto: the biggest prize may not be the EBITDA uplift during the hold but the exit multiple, since a business with AI genuinely embedded in how it operates is worth more to the next buyer, if you can evidence it. 

The risks investors still underestimate 

Sophisticated investors now tend to overestimate the obvious risks, disintermediation and pricing pressure, and miss two subtler ones. The first is margin compression through pricing-model drift: as delivery gets more efficient, time-and-materials contracts mechanically shrink, so the mitigation is often a move to fixed-fee pricing or redeploying freed capacity into new work. The second is the table-stakes trap: a capability can fail to show up in the bottom line and still be necessary, because the market’s quality bar is rising and the real cost is falling behind. Some AI investment is defensive by nature, and the honest move is to size it that way. 

On model costs, the panel pushed back on the idea that providers will simply raise prices once dependency sets in. Costs have trended down and competition is real; the practical defences are architectural, building model-agnostic systems and not using frontier models for tasks a cheaper one handles fine. 

Where this leaves investors 

The AI due diligence question has moved on from whether a company is experimenting with AI. It is whether AI is material enough to change the investment thesis, whether the organisation can act on it, and whether risk mitigation is already built into the plan rather than deferred to post-close. 

Watch the full session for the portfolio examples and the panel’s take on the sectors most exposed to disruption : Is AI the Next Great Value Creator — Or the Biggest Risk in Your Deal?