The Cross-Portfolio AI Playbook: Triage, Prove, Scale

The unit is the portfolio, not the company

The cheap version of “AI value creation” is 40 portcos each hiring a consultant and running a pilot. The expensive-but-real version pools the work: one team finds a bet, proves it once, and ships it across the book so company #2 through #40 pay zero discovery cost.

KKR is the cleanest example. Its Capstone operating team explicitly names cross-portfolio synergies as the mandate — codifying a win in one company into a repeatable playbook deployed across 225+ businesses, with conversational agents that recommend pricing or lean-manufacturing tactics learned from prior engagements (Umbrex profile of KKR Capstone). Hg runs the same logic in software: 60+ portfolio companies treated as one of the largest software ecosystems deploying AI at scale (Hg).

The thesis is simple: discovery is the expensive part. A firm that pays it once and reuses it 40 times beats a firm that pays it 40 times. That arbitrage is the program.

Stage 1 — Triage: not every portco gets AI, and that’s the point

The first discipline is saying no. Blackstone sorts its book into three buckets: AI is central, AI is meaningful upside, or AI is immaterial — and only the first two get serious operating attention (Blackstone). Vista goes further and weaponizes the triage: it now scores every portfolio company on AI-adoption velocity and ties that score to future capital allocation (Vista).

McKinsey’s independent read backs the discipline: winning value-creation plans identify two to three high-impact use cases per company, not a sprawling “digital transformation” (McKinsey). Spreading a thin AI budget across 40 companies produces 40 performative pilots and zero P&L. Concentrating it on the 8 where AI is central produces wins big enough to measure.

Stage 2 — Prove on a fraction, with a control group

This is the section every “AI strategy” skips and the one that separates the 5% from the 95%. MIT’s 2025 NANDA study found 95% of enterprise GenAI pilots delivered zero measurable P&L impact despite $30–40B spent — and the surviving 5% did the unglamorous work of baselining, holdouts, total-cost transparency, and portfolio discipline (MIT NANDA via Fortune). Holdouts means a control group. No control, no proof.

Blackstone’s own examples show the shape when it works: at Renaissance Learning, a lead-gen model doubled average order value; at Liftoff, AI pricing and bidding drove a 10% revenue increase — and Blackstone is blunt that “too many companies run performative pilots that never show up in EBITDA” (Blackstone Data Science). Bain’s survey of investors managing $3.2T AUM found only ~20% of portfolio companies have operationalized GenAI with concrete results (Bain Global PE Report 2025).

The discipline: baseline the metric, run the bet on a fraction of accounts, reps, or regions, hold the rest as control, read the delta. If you can’t see it against the control, it didn’t happen.

Stage 3 — Roll out: turn one proven bet into a portfolio asset

Once a bet clears the control group, the value-creation team’s job is replication, not reinvention. Hg’s Catalyst incubator embeds tiger teams inside portcos and has shipped 100+ AI products across the portfolio, compressing product build time from nine months to under three (Hg Catalyst). Hg’s value-creation team is 100+ AI experts — a shared resource no single portco could afford alone (Hg).

Vista’s Agentic AI Factory is the same move at the infrastructure layer: built on Azure, Google, and AWS partnerships, it gives portcos priority tooling and exclusive cloud-pricing discounts — Vista reports 80% of portfolio companies deploying GenAI (Bain). Treat the firm-reported headline numbers as exactly that — but the structural pattern is well-corroborated.

Stage 4 — Shared leverage: the portfolio’s quiet edge

The most under-discussed play isn’t a model — it’s buying power. KKR’s Capstone uses group-purchasing organizations to consolidate demand across 230+ portcos and $500B+ AUM (Art of Procurement). Apply that to AI: one negotiated rate for the model API, the data platform, and the dev-tools license, across the whole book. Thoma Bravo’s Google Cloud partnership is the explicit version — a firm-level deal letting $180B+ AUM of enterprise-software portcos embed Google’s models at negotiated terms (Thoma Bravo).

Stage 5 — AI in diligence and the exit narrative

The playbook bookends the hold. On entry, Deloitte’s 2025 study found 86% of M&A organizations now use GenAI in deal workflows and 35% in target screening and diligence; Bain notes AI-assisted diligence can compress timelines 30–40% (Deloitte). EQT’s Motherbrain platform — the firm holds the first AI patent in PE — screens 50M+ companies to source targets and bolt-ons (EQT).

On exit, the proven AI program becomes the story. The skeptic’s caveat: a roadmap is not realized EBITDA. A control-group-proven, replicated win underwrites the multiple; a deck of pilots does not.

What kills these programs

• Performative pilots with no holdout. No control group = no proof = it never shows up in EBITDA. The single most common killer.
• Spreading the budget evenly. Funding the AI-immaterial bucket starves the AI-central bucket. Triage exists to be acted on.
• Reinventing per portco. If the second deployment costs as much as the first, you have 40 projects, not a portfolio asset.
• Buying tech before deciding how to capture value. MIT’s literal diagnosis — tools bought before defining the P&L line to move.
• No CEO ownership. Blackstone names CEO-level ownership as a precondition; delegated-to-IT pilots stall.
• Roadmap-as-exit-story with nothing realized. Buyers — and diligence AI — increasingly see through a deck of pilots.