AI Was in the Plan. It’s Not in the EBITDA.

Why AI initiatives are stalling in the middle market

Somewhere in the value creation plan for almost every deal closed since 2024 is a line about AI-driven margin expansion. Eighteen months later, a lot of those line items are being deferred, repurposed, or marked for a write-down. The pattern is hard to miss, but nobody wants to be the first to put it on the agenda for a Monday call. 

The thesis assumed the PortCo could absorb AI the way it absorbs a new pricing analyst: drop it in, train the team, see results in two quarters. What the thesis didn’t price in was the data infrastructure debt inherent in most middle-market companies. You’ve got three ERPs from three add-ons, a finance team running the close on a dozen linked workbooks, and a demand planner whose “system” is a SharePoint folder. Drop a machine learning model on top of that and the output will be confidently wrong. 

This is where the gap between deal team and operating team gets expensive. The deal team underwrites a margin lift. The operating team inherits a stack that cannot support the analytics the thesis assumed. A year later, the LP report has to reconcile the difference, and “AI initiative is in early stages” starts showing up in the language. 

What AI readiness means at the portfolio level

The harder conversation for operating partners right now isn’t vendor selection or who to hire as CIO. AI readiness at the portfolio level comes down to a much less glamorous question of whether the data your management teams depend on is reliable, current, and consistent across the entities you’ve stapled together through add-ons. If it isn’t, the AI line in the value creation plan is decorative. 

The exit premium has moved

The same dynamic is showing up on the exit side, faster than most expected. Buyers are starting to do real diligence on the operational evidence behind AI claims now, and the bar is higher than it was eighteen months ago. A target that says “we’ve deployed AI-driven forecasting” is being asked to show forecast accuracy over time, that the underlying data flows are auditable, and that performance survives without the consultant who built it. 

The targets that can answer those questions are clearing at full price. The ones that can’t are seeing the AI premium come off the multiple in negotiation because they can’t underwrite what they can’t verify. 

The premium is paid for verifiable, repeatable performance built on data infrastructure a buyer can underwrite. Buyers have gotten faster at separating capability from claim. When a management presentation leads with AI capability, buyers are asking the data room to confirm it. The firms that have done the foundational work can answer that request. The ones that haven’t are having a different conversation with their bankers. 

The work that must happen in year one

The firms that understand this are starting their data work in year one of the hold, not year four when the bank book is being prepared, because they understand what a buyer is actually paying for. 

There is no shortcut. The first move isn’t a model. It’s an honest assessment of whether the data foundation in each PortCo can carry the thesis you sold to your LPs. That assessment has to happen at the portfolio level, systematically, not company by company when it becomes urgent. If it hasn’t happened yet, the AI lever is theoretical. If it has, you’re already ahead of most of the market. 

Resultant and Liberty Advisor Group work with PE firms and their portfolio companies on the data infrastructure that holds up in a data room. If you’re evaluating where your portfolio stands, our approach is here.