A recent Accordion survey of PE-backed CFOs found that 85 percent of buyers now assess AI-enabled finance capabilities when valuing portfolio companies, and that CFOs with those capabilities are twice as likely to report smoother exits and higher perceived valuations.¹ Meanwhile, West Monroe's 2026 PE outlook notes that tightening debt markets and LP pressure now demand verifiable operational improvements at exit, "with firms lacking credible AI roadmaps facing valuation penalties."²

The pressure to show value from AI is showing up in transactions, but much of the conversation around "AI as valuation driver" is still vague. Sponsors know it matters, but what are acquirers looking for in the data?

Operationalizing AI Efficiency in PortCos

The GTreasury deal provides a visible example of AI value creation. Hg's Catalyst incubator built GSmart AI, an agentic product layer on the core treasury platform. Within the product, AI agents forecast cash positions, detect FX and interest rate exposure anomalies, and auto-generate board-ready risk narratives without leaving the application. Bookings growth increased roughly 40 percent year-on-year, and more than 40 percent of inbound leads before the sale came directly from the AI product. Ripple acquired the business for $1 billion.³

Lloyd Hilton, head of Hg Catalyst, said, "The fact that we were a forward-leaning, product-led company that was already integrating industry-leading AI capabilities into the platform made us a much more attractive proposition for a strategic acquirer."³

GTreasury is an example of a software business where AI drove measurable top-line growth, but sponsors are also claiming significant productivity gains beyond just AI-centric companies. PSG Equity reports that AI now handles up to 70 percent of customer support cases across companies in its portfolio.⁴ At Apollo-backed Cengage, costs are down 40 percent in select content production processes.⁵ Another Apollo portco, Shutterfly, realized $5 million in new revenue in the first year after embedding AI directly into the core product workflow, auto-generating personalized layouts from uploaded images and turning a manual design step into a one-click conversion path.⁵ These are financial metrics that would show up directly in margin expansion and can be tested during diligence.

None of these are moonshot AI projects. They are focused deployments in support, content, and transaction processing workflows that show up in a buyer's operating model with clear before-and-after numbers.

What Moves the Needle in a Buyer's Model

Here are some sharper distinctions that are becoming more important in underwriting:

AI-sourced revenue, not AI-adjacent revenue. If you can't isolate what your AI capabilities are generating in revenue or pipeline, a buyer will discount accordingly.

Embedded capability, not bolt-on tools. GTreasury's GSmart AI didn't just surface dashboards — it deployed agentic workers that forecast cash positions, flag exposure anomalies, and generate board-ready risk narratives, all running inside the treasury workflow customers use daily. That's different from a chatbot stapled onto a settings page. A standalone tool bolted onto an existing product is a quarter's worth of engineering for a competitor; a deeply integrated capability is not.

A team that can keep building. If the AI capability was built by two contractors who left six months ago, the premium disappears. Even a small embedded team with a forward roadmap changes the buyer's perception of what they're acquiring.

Operational gains with baselines, not claims. "We use AI in customer support" is a slide. "AI handles 70 percent of tier-1 cases, resolution time dropped from 4 hours to 12 minutes, baseline from Q2 2025" is a diligence exhibit. The difference between those two sentences could be a turn of multiple — the first invites skepticism, the second invites modeling a new forecast.

The 18-Month Build

Portfolio companies need actionable plans to create value with AI, with at least an 18-month runway to realize reportable results:

Months 1–3: Instrument and baseline. Pick two or three workflows where AI can have measurable impact, such as support resolution and routing, data reconciliation, or development velocity. Establish baselines at the start of the project to drive specificity and make the numbers diligence-ready versus anecdotal. Powerdot, an Antin portfolio company, tracked a 35 percent reduction in development cycle time after introducing AI into its engineering workflow.⁴

Months 4–9: Build, ship, and track. Deploy AI into those workflows and measure impact with the same rigor you'd apply to any value creation plan.

Months 10–18: Package for diligence. Compile AI-sourced revenue attribution, operational efficiency gains with baselines, product architecture documentation, and a forward roadmap.

The work here isn't technically difficult, but it requires organizational mobilization to put measurement in place early. When a buyer shows up, you will have 12+ months of data instead of a deck full of projections.