Agentic AI in private equity portfolios: deploy agents, and value the records behind them

Two sides of the same technology

An AI agent is software that plans and carries out a multi-step task using tools: it reads a customer's email, checks the order in the ERP, applies the refund policy, issues a credit and updates the CRM. For a private equity sponsor, agents show up twice in the same portfolio.

On the cost side, portfolio companies deploy agents in support, finance, sales operations and the back office to handle more volume without adding headcount. On the asset side, the AI labs building those agents need training and evaluation material that shows how real work gets done, step by step, with the outcome attached. That material sits inside companies like yours, in ticketing systems, ERPs, CRMs and email archives.

Sponsors are already pushing the first lever. McKinsey's Global Private Markets Report 2026 says multiple expansion and cheap leverage, which accounted for 59 percent of private equity returns between 2010 and 2022, have faded, so operational value creation is now likely the primary source of returns. It also finds firms have more than doubled their operating groups since 2021 and are applying AI to operating levers. The second lever, licensing rights-cleared historical records through SourceX, is the one most AI plans leave out.

Why do AI labs need records of real multi-step work?

Agents learn tasks from examples of tasks, and those examples are scarce outside companies. Public web text taught models to answer questions; agents need sequences of tools, decisions, exceptions and results from inside real businesses.

The public supply is also finite. Epoch AI researchers project that, if current trends continue, language models will fully use the stock of human-generated public text sometime between 2026 and 2032. The forecast carries wide uncertainty, but it explains why non-public, permissioned business records have become a scarce input.

The most useful records combine three features: several steps, more than one system and a recorded outcome.

Cost lever or records asset: how do the two compare?

They are different decisions with different owners, and a portfolio can pursue both.

One overlap needs checking. If an agent vendor's contract already lets the vendor train on the company's data, raise it early: data already licensed for AI training is a red flag, and the exclusive term of a new license has to be consistent with what earlier contracts granted.

The agent-records screen for a portfolio review

Use these six questions per company. One that passes most of them is worth an introduction.

• Recorded work: does the company handle repeatable multi-step work, such as tickets, approvals, claims, orders or change requests, and record how each one ended?
• Scale: did it reach 50+ full-time employees at peak, contractors excluded, with several years of documented operations?
• System depth: does work run across many systems, often 10-15+ in strong companies, with archives that can still be exported?
• Rights: did the company create the records, and do customer contracts and employee notices allow licensing? Are they mostly business records rather than consumer personal data or health records?
• No prior grant: has no agent vendor or earlier licensee already received AI-training rights to the same data?
• Sponsor: will the CEO, CFO or owner consider an exclusive license for an agreed term?

The company fit checker runs a preliminary, non-binding version with no contact details required, and who qualifies sets out the full baseline.

When should an operating partner raise it?

The right moments are the ones where agents are already on the agenda.

Timing matters more when holds run long; PE exit backlog 2026 covers that pressure, and AI in private equity 2026 covers adoption across portfolios. Portfolio CFOs are building the same systems inventory for their own reasons, as CFO priorities 2026 describes.

How the introduction works

1. You register as a partner and either send the portfolio company your referral link or submit it through the referral form.
2. SourceX qualifies the company on headcount at peak, operating history, breadth of records and rights.
3. The company completes a data inventory covering systems, years of history and export options.
4. Price and terms are agreed with the company first; AI labs and data buyers then review, and once the company is deal-ready, buyers typically respond within about two weeks.
5. After signing, the data is prepared under the de-identification and redaction rules agreed before any work began, delivered with the company's authorization, and the company is paid.

The operating team never exports, uploads or handles the records.

What to tell a portfolio CEO

Ready-made versions are in the introduction email templates for operating partners.

How rewards work for sponsors

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. It is paid only after the buyer pays and SourceX receives its fee, and no reward is guaranteed.

The reward is a share of SourceX's fee and is never deducted from what the portfolio company receives. Fund documents may say how fees linked to portfolio companies are treated, for example whether they offset management fees, so confirm that with your fund CFO or counsel first. The operating partner page covers the program for this role.

When the records lever does not apply

• The company never reached 50 full-time employees at peak, or has only a short operating history.
• Its records mainly belong to clients, as at many agencies, outsourcers and BPOs, and those clients have not agreed.
• Most of the data is consumer personal information or protected health information.
• Archives were deleted during earlier migrations, or nobody can run exports.
• The material was generated with AI in order to sell it; buyers want records of real work.

Next step

Add the agent-records screen to your next portfolio review and pick the two companies with the longest histories of recorded work. When one passes, register as a partner and introduce it, or have the CEO apply at sourcex.si/apply through your referral link.