KKR's Pete Stavros on the Reality of Portfolio AI

Key Takeaways

• Pete Stavros reports that across roughly 250 portfolio companies, AI serves as an incremental operational lever rather than a core thesis for underwriting buyouts.
• Sitting management teams frequently lack visibility into AI opportunities, prompting KKR to bring in external specialists for asset-level operational diagnostics.
• KKR runs concurrent multi-vendor trials across its portfolio to identify which software providers deliver actual margin gains before buying at scale.
• The firm embeds technical talent directly alongside deal teams and directs portfolio companies to hire internal AI engineers instead of outsourcing tech strategy.
• KKR manages portfolio rollouts through a structured process known as KKR's Portfolio AI Experimentation Grid.

The KKR's Portfolio AI Experimentation Grid

• Step 1: External AI Diagnostic: Bring in external AI experts to audit each portfolio company's operational risks and technological opportunities because sitting management teams often lack visibility into rapid AI advances.
• Step 2: Mandatory Baseline Experimentation: Require every portfolio company to actively pilot at least one concrete AI experiment.
• Step 3: Multi-Vendor Matrix Testing: Plot companies along one axis and specific AI applications along the other axis, testing dozens of competing vendors concurrently across the portfolio grid.
• Step 4: Scale Validated Vendor-Application Matches: Identify top-performing matches of applications, vendors, and industry niches, then systematically roll out the validated solutions across all relevant portfolio assets.
• Step 5: Direct Technical Talent In-Housing: Embed dedicated AI engineers alongside deal teams and direct portfolio companies to hire internal engineering talent rather than relying exclusively on third-party vendors.

When This Works (and When It Doesn't)

This framework fits mega-cap buyout sponsors with massive, diversified holdings. With 250 assets, KKR holds enough statistical volume to run vendor bake-offs in parallel. When two dozen companies test customer service tools simultaneously, the sponsor gathers comparative performance data in months rather than years. Scale turns vendor selection into an empirical process.

For mid-market or sector-focused funds with 15 to 20 assets, running dozens of concurrent experiments is rarely feasible. Smaller sponsors lack the balance sheet to retain outside diagnostic teams for every platform asset. They also lack the purchasing power to extract favorable enterprise pricing when rolling a winner across the group. In specialized portfolios, forcing generic experiments can distract management from sector-specific execution.

Why It Matters

Stavros provides a candid counterweight to market marketing. Despite widespread commentary about instant operational disruption, top-tier buyout shops find that actual institutional overhauls remain rare. Software vendors promise immediate EBITDA expansion, yet real-world rollouts still face technical integration friction and management inertia.

Treating AI as a series of controlled experiments rather than an underwriting driver signals a disciplined shift in private equity operations. Instead of paying entry multiples that price in speculative margin expansion, sponsors use their portfolio scale as an R&D testbed. The edge does not come from believing the hype; it comes from running disciplined tests to separate functional tools from expensive vendor promises.