AI value creation in private equity: a playbook for operating partners

Two tracks of AI value creation

AI creates value in a portfolio in two different ways, and most playbooks cover only the first. On the buy side, a company adopts AI tools to do existing work faster or cheaper, or to sell more. On the supply side, a company with years of recorded work licenses those records to AI developers who need them to train and test their systems.

The tracks draw on different resources. Buy-side projects need usable data, a budget, an internal owner and months of change management. The supply side needs records the company already holds, the rights to license them and an executive willing to sign. A 200-person IT services firm with no data team can be a weak candidate for an agent rollout and a strong candidate for a license at the same time.

Which buy-side AI levers work in portfolio companies

The levers that pay back soonest sit in high-volume, rules-heavy workflows where the company already measures throughput. Start where a baseline exists, so savings can be proven rather than estimated.

• Customer support: drafting replies, summarizing case history and routing tickets, measured by handle time and first-contact resolution.
• Finance operations: invoice capture, cash application, close checklists and variance commentary, measured by days to close.
• Sales: proposal and RFP drafting, call summaries and CRM hygiene, measured by cycle time and win rate.
• Engineering: coding assistants and test generation in software-heavy companies, measured by cycle time and defect rates.
• Back-office documents: pulling fields from contracts, bills of lading, claims forms or work orders.

Each lever needs a named owner inside the company and a before-and-after metric written into the value creation plan. Without both, the board sees activity rather than results.

Why AI projects stall in the lower middle market

Lower-middle-market companies rarely run short of AI ideas. They stall on plumbing and people.

• Records are spread across many disconnected tools, often 10-15 or more in a mature company, with no warehouse joining them.
• IT is run by one manager or an outsourced MSP whose contract does not cover data engineering.
• Pilots launch without a baseline, so nobody can show the CFO what changed.
• Seat licenses get bought for everyone and used by a few.
• The managers who own the workflow were never asked to redesign it.

The practical response is to fund fewer projects, each with a clear owner and metric, and to share vendors, playbooks and prompts across the portfolio instead of letting every company start from scratch. The value creation levers overview shows where AI sits next to pricing, procurement and working capital.

The supply-side lever most playbooks skip

AI developers need records of real work: support cases with resolutions, project files with outcomes, engineering reviews, approvals and exceptions. As AI shifts from systems that answer questions to agents that complete tasks, records of multi-step workflows have become a scarce input, because they live inside companies rather than on the public web. Researchers at Epoch AI project that language models could fully use the stock of public human-written text between 2026 and 2032 if current trends continue. It is a forecast with wide uncertainty, but it explains why permissioned non-public data has gained value.

For an operating partner, the attraction is that this lever does not depend on AI talent. The company licenses records it already holds, keeps ownership and approves the scope. It is paid one all-in price as a single payment, typically within about 60 days of invoicing once a buyer selects the data. Licenses are typically exclusive for AI training for an agreed term, and nothing is binding until the company signs.

Which portfolio companies fit the supply side

The supply side favors companies whose work leaves a written trail. Industry matters less than headcount, history and how many systems capture the work.

Companies that have been acquired, merged into a platform or wound down can still qualify if the archives survive. The who qualifies page sets out the full baseline.

The dual-track AI screen

Ask both sets of questions in the same portfolio review. A company can pass one track and fail the other.

Buy-side questions

• Is there one high-volume workflow with a measured baseline?
• Is there a named owner who will run the change, not just sponsor it?
• Can the tool reach the data it needs without a new integration project?
• Is the budget for licenses, implementation and training approved?

Supply-side questions

• Does the company run many business systems with years of retained history?
• Did the company create those records, and do client contracts and privacy promises allow licensing?
• Can you reach an executive with authority to sign?
• Would that executive consider an exclusive AI-training license for an agreed term in return for a one-time payment?

The network opportunity finder helps you think through which companies and executives in your wider network deserve the supply-side questions, including businesses outside the current fund.

When to raise AI during the hold

Raise both tracks at moments when leadership is already reviewing systems, budgets or the equity story.

How the introduction works without touching data

You open the door; the company and SourceX do the work behind it.

1. Register as a partner, then send the CEO your referral link or submit the company through the referral form.
2. SourceX reviews size, history, data breadth and rights with the company's sponsor.
3. The company builds a data inventory: its systems, the years each one covers and what can be exported.
4. The company and SourceX settle price and licensing terms before any buyer sees the opportunity.
5. AI labs and data buyers review it; once a company is deal-ready, buyers typically respond within about two weeks.
6. The license is signed, records are prepared under redaction and de-identification rules agreed at the start, delivered with the company's authorization, and the company is paid.

At no point do you export, upload or describe the company's confidential records.

What to say to a portfolio CEO

Place the supply side next to the buy side so it reads as part of the AI agenda rather than a sales pitch.

For written versions, adapt the introduction email templates for operating partners.

How rewards work for operating partners

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. The reward is paid only after the buyer pays and SourceX receives its fee; an introduction, meeting or signed agreement alone does not trigger payment, and no reward is guaranteed.

The reward comes out of SourceX's fee, so it never reduces what the portfolio company receives. Your firm may restrict personal fees connected to portfolio companies; check its policy, and read the referral opportunities for operating partners page, before you register.

When to leave the data lever alone

Skip the supply side when the records mainly belong to the company's clients, when the data is mostly consumer personal information or patient records, when archives were deleted or tools cancelled without an export, when the same data has already been licensed for AI training, or when the owner will not consider an exclusive license. A company that fails today may qualify later, for example once a migration project preserves its history.

Next step

Put the dual-track screen on the agenda of your next portfolio review. For each company that clears the supply-side questions, register as a partner and make the introduction, or ask the CEO to apply directly at sourcex.si/apply through your referral link.