The conversation about AI inside PE-backed businesses has shifted in the last twelve months. Where the question used to be whether the AI thesis would deliver value during the hold period, the more pressing question is now whether the value being delivered will still be visible at exit. Operating partners are starting to ask it, CFOs are modelling it into their value-creation plans, and CTOs are being asked to account for it in board updates with the kind of scrutiny that used to be reserved for capital projects. The answer determines whether AI investment compounds into multiple expansions or disappears into the asking price.

The reason this matters is that AI capability does not behave like other technology investment on a sale process. A modernised data warehouse and a rebuilt customer platform are all assessed by buyers as durable assets with a clear cost base and a clear utility. AI is harder to read as some of itis durable capability that the next owner can operate, scale and extract value from, and some of it looks like capability inside the hold period but reveals itself as a cost saving the next buyer has already priced in. The difference between the two is usually not visible from the AI itself. It is visible in the architecture around it, the data underneath it, the governance built into it and the people responsible for it.

What Buyers pay for

The AI investments that hold up at exit share a small number of characteristics, and they have less to do with the sophistication of the model than with the conditions in which it operates.

AI capability that has shipped, been measured and continues to generate results is credited by the buyer because the buyer can see what they are paying for. Capability that exists as a roadmap, a pilot or a forward-looking projection is harder to value and harder to defend in the diligence conversation. The PE-backed businesses that get full credit for their AI investment are usually the ones that ran fewer initiatives, finished them, measured them and let the numbers speak.

Where the AI is running on data infrastructure that has been modernised, integrated and structured for the workload, the buyer assumes the capability will continue to perform after the sale. Where the AI is running on a tightly coupled monolith, fragmented data and brittle integrations, the buyer assumes the capability will degrade once the original engineering team is no longer holding it together. The second scenario is priced down accordingly.

‍

Lee Provoost, CTO of Flagstone, has made the elated point that the discipline required to operate in a regulated environment is the same discipline that produces durable systems:

‍

"I often find that constraints breed creativity. Some of the things we have to do to protect client data and client money are probably things that most companies should do anyway. I am quite horrified sometimes by the cavalier attitude towards sensitive data from consumer tech businesses."

‍

[Episode 50 – Leaders and Founders Podcast]

‍

The observation matters because the businesses that have already built under regulatory constraint tend to have the foundations AI needs in place by default. Data is protected, controls are embedded, risk is designed into the architecture rather than appended to it. The AI investment that follows starts from a stronger position and is more likely to be credited at full value in the diligence conversation. The businesses that have not built under those constraints find themselves doing the foundational work alongside the AI work, and the cost differential between the two paths shows up in the valuation either way.

‍

What Buyers Discount

‍

The AI investments that lose value at exit usually share a different set of characteristics, and they are easier to recognise once the pattern is named.

The most common is AI deployed against an existing process rather than designed into a new one. When automation is added to a workflow as it already exists, the buyer prices the resulting cost saving as efficiency already captured. The investment is not credited as a transformation because a transformation requires the underlying operating model to have changed. Efficiency does not multiply, but operating model change does. The same investment, made earlier in the hold and aimed at a redesigned process rather than the existing one, would be valued differently.

AI systems without embedded governance are increasingly treated as regulatory exposure rather than capability. The next buyer either prices the cost of retrofitting the governance, defers the purchase until the work is done, or refuses to absorb the risk at all. The cost of building governance at the design stage is lower than the cost of building it in after deployment, and that cost differential shows up in the valuation conversation either way.

The third is key-person dependency. Where the AI capability depends on the institutional knowledge of two or three engineers, the buyer prices the risk that those engineers will not stay through the integration period. The capability is treated as conditional rather than owned, and the discount reflects the uncertainty. The portfolio companies that get full credit for their AI investment are usually the ones that documented the capability and embedded it into the team's ways of working.

‍

What the Strongest AI Investments Have in Common

The PE-backed businesses generating compounding value from AI investment share a common operating pattern. The decision about how AI would change the business was made early in the hold and tied to a specific operating model change rather than a thematic line item in the value-creation plan. The foundational work needed to support the AI was assessed before the AI scope was fixed, and the cost of that foundational work was funded as part of the AI programme rather than treated as a separate cost. Governance was designed into the architecture from the start, with risk and control treated as a feature of the system rather than an appendix to it. Technical leadership had the authority and the brief to push back when the value-creation plan started to outpace what the foundations could support.

As Kirsty Rutter put it on Leaders and Founders:

"Innovation and change and ideas are fab, but unless you actually get executing, it counts for not very much."

[Episode 49 – Leaders and Founders Podcast]

‍

The observation lands particularly hard at exit, because buyers do not pay for AI ambition, only for AI capability that has shipped. The portfolio companies that have moved from idea to deployment to measurable outcome within the hold period are the ones whose AI investment compounds. The ones still describing what they intend to do are the ones whose AI investment is treated as a forward-looking projection.

‍

What the Next Twelve Months Look Like

Several PE portfolios have already locked their AI value-creation plans for current holdings. The businesses within those portfolios are about to start delivering against plans underwritten when both the technology and the operating model decisions were less well understood. Some of those plans will hold up. Some will need to be re-scoped mid-hold to recover value, and the cost of that re-scoping will be higher than the cost ofgetting the plan right at the outset. Some will be carried into a diligence conversation that will not credit them the way the value-creation plan assumed.

For deals being underwritten now, the question is whether the AI thesis is being built on a clear operating model change with foundations that can support it, or on a thematic assumption that AI will deliver margin improvement somewhere in the hold. The first is investable. The second is increasingly visible in technical due diligence and increasingly easy to discount.

If the argument in this issue resonates and you're not sure whether your current foundations would hold up in a diligence conversation, our Modernisation Readiness Assessment is a good place to start.