How sponsors run a portfolio-wide AI program across a PE portfolio

What is a portfolio-wide AI program?

It is a sponsor-level function, often called an AI center of excellence, that does once what every portfolio company would otherwise do alone: choose vendors, negotiate terms, write usage policies, find scarce talent and set security rules. Each company still owns its roadmap and its results. The program supplies the shared parts and keeps the pace up.

The case for building one has grown as the source of returns has shifted. McKinsey's Global Private Markets Report says multiple expansion and cheap leverage, which accounted for 59 percent of private equity returns between 2010 and 2022, have faded, leaving operational value creation as the likely primary source of returns. The same report finds firms have more than doubled their operating groups since 2021, are applying AI to operating levers, and that 53 percent of 300 LPs surveyed ranked a GP's value-creation strategy among their top five selection metrics.

For a head of value creation, that last point matters twice: a credible AI program helps the portfolio and helps the next fundraise.

What does the central team provide?

Programs usually combine some or all of six components. The right-hand column shows how each one also builds a company's readiness to license its records, which is the module this guide adds.

For use cases in depth, see the 2026 playbook on AI value creation in private equity. This guide focuses on structure and on the supply-side module.

How to stand the program up in stages

1. Baseline, first quarter: survey each company's systems, AI pilots, security posture and data owners. A short questionnaire to each CEO and CIO is enough to start.
2. Shared services, first two quarters: sign group vendor terms, publish the acceptable-use policy and set security minimums.
3. Playbooks, ongoing: release two or three proven use cases at a time, each with a named owner at every participating company.
4. Records module, alongside the baseline: add five questions about operational records to the same survey, described below.
5. Reporting, quarterly: fold AI KPIs and records status into the portfolio review pack.

Sequencing matters more than speed. Group vendor terms signed before the security baseline exists may need to be reopened once it does.

The supply-side module: flagging licensable records

Most portfolio AI programs look only at demand: what AI can do for each company. A supply-side module asks the reverse question. Which companies hold records that AI developers need, and could license them for a one-time payment while keeping ownership?

The module reuses work the program already does. The systems survey, the security baseline and the governance rules produce most of the information. Add five questions per company:

• Has the company had 50+ full-time employees at peak, contractors excluded, and several years of documented operations?
• Which systems hold its operational history, such as email, Slack or Teams, CRM, finance, support, engineering and operations tools, and how far back does each go?
• Are archived or retired systems still exportable, and who can run the export?
• Did the company create these records itself, or does a large share belong to clients?
• Is there an owner, CEO, CFO or authorized representative who would consider an exclusive, time-limited license?

Keep the module at the level of metadata. It never collects, uploads or reviews actual records; it captures answers about systems, years and rights. To extend the same thinking beyond the current funds, for example to former holdings or management teams now running other businesses, use the network opportunity finder.

Why the security baseline and licensing readiness reinforce each other

The controls a security baseline asks for are the same ones that make a dataset licensable: documented access, defined retention, a known export owner and clear rules on what leaves the company. A company that cannot say who is able to export its support tickets has both a security gap and a licensing gap.

The reverse also holds. Retention schedules written purely to cut storage cost can delete the archives that give a company its value to AI buyers. Before the program standardizes deletion rules, ask each company whether long histories should be preserved and exported first, subject to any legal or contractual obligation to delete.

Who decides: the center or the company?

The center sets the frame, and each company makes its own decision. That matters most for licensing, because the company owns its records and is the party that signs. A head of value creation can flag candidates and make introductions but cannot license on a company's behalf.

Portfolio CFOs are well placed to own the inventory work at company level, which is why the guide to the PE portfolio CFO role is worth sharing with them before the module launches.

How introductions from the program work

1. The module flags a company that meets the baseline and holds deep records it owns.
2. The head of value creation or the deal's operating partner raises it with the CEO at the next operating review.
3. If the CEO wants to proceed, the sponsor registers as a partner and submits the company through the referral form, or the CEO applies using the sponsor's referral link.
4. SourceX qualifies the company on size, history, data breadth and rights, and the company completes a data inventory.
5. The company agrees price and terms, buyers review the opportunity, and the company signs only if the deal works for it.
6. Records move only after the executed agreement and the company's authorization, under redaction rules agreed at the outset.

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. Nothing is payable until the buyer has paid and SourceX has received its fee, rewards are not guaranteed, and the amount never comes out of the company's proceeds. Check the firm's own fee policies first; see the referral overview for PE operating partners for how sponsors take part.

Where portfolio programs fall short

• Central programs stall when companies see them as reporting overhead. Tie every module, including this one, to a line in a value creation plan.
• The records module will flag fewer companies than the adoption side. Many will lack the scale, the history or clean rights.
• Licensing is one-time money, so it sits next to recurring levers rather than replacing them; the revenue acceleration guide for PE-backed companies covers those.
• Companies holding mostly client-owned, consumer or health data will rarely qualify, and the module should say so early.

If the board is weighing licensing against building data products, the licensing vs data products comparison lays out what each path demands.

Next step

Add the five records questions to your next portfolio AI survey. For any company that passes, register as a partner and make the introduction; share the who qualifies criteria with each CEO before the first conversation.