AI value creation in private equity portfolios: where data fits

What are the three AI value creation levers in a portfolio?#

The three AI value creation levers are cost savings, AI revenue and data as an asset. Cost savings come from automating work such as ticket triage, invoice matching or dispatch scheduling. AI revenue comes from new features or services a portfolio company can charge for. Data as an asset treats the company's operating records as something that can be documented, cleared and licensed to AI developers.

General partners mostly expect the first lever to show up first. In the 2026 StepStone and Bain GP outlook survey, expected 2026 AI outcomes at portfolio companies were 46% cost savings or efficiency, 10% revenue growth and 39% too early to tell. Bain's September 2026 analysis of the AI value paradox found little correlation between AI spend and value at most portfolio companies, and named scattered experiments and weak links to value-creation plans as the main causes.

The third lever gets less attention because it does not come with a software demo. It asks a different question: can years of support tickets, job records and quality logs that the company already holds be documented, rights-cleared and licensed under terms the company controls? For some companies the answer is yes; for others, rights or privacy burden rule it out.

Why do operating records sit underneath every lever?#

Operating records sit underneath every lever because each one asks the same first questions: which systems hold the history, how many years can still be exported, whether requests link to decisions and outcomes, and who controls the rights. An automation project needs clean history to configure and test. An AI feature needs permission to use customer content. A license needs rights evidence and privacy preparation. Surveys point the same way: in RSM's 2026 middle-market survey, data quality and availability issues were the top inhibitor to AI deployment, cited by 34% of respondents, ahead of security and privacy concerns at 30%.

When each workstream runs its own discovery, three teams interview the same IT lead about the same Zendesk instance and the same ERP. A single inventory, kept per company and refreshed at each board cycle, removes that duplication and gives the operating partner one view of what every company actually holds.

What does treating data as an asset involve?#

Treating data as an asset involves three pieces of work: an inventory of record families, a rights review, and a decision about licensing. AI developers license records of real business work, such as support conversations, code reviews, dispatch histories and nonconformance reports, because those records show how experienced people make decisions under real constraints.

The company keeps ownership. A license grants defined use for a defined scope, and the supplier approves what leaves the building. There is no published price list for this kind of record; value becomes clear only when a buyer engages with a specific, documented package.

• 1Record families named by system, with years of accessible history.
• 2Rights evidence: customer contracts, vendor terms, employee notices and confidentiality obligations reviewed.
• 3A privacy record showing how personal and confidential details are removed.
• 4A named signer for the supplier entity and a map of sponsor, board and lender consents.
• 5A delivery route that keeps large archives in the company's own storage.

How should an operating partner sequence the portfolio?#

An operating partner should sequence the portfolio by matching each company to the lever its records support best, not by running every lever everywhere at once. Companies with deep, linked operating history and clean rights are natural candidates for the data lever. Companies with thin history but repetitive manual work usually start with cost savings.

Timing triggers also matter. A planned ERP or helpdesk migration puts history at risk, so a company facing one should have its data reviewed before the old system is switched off, even if it would otherwise rank lower.

Where the three levers can conflict#

The three levers can conflict when a decision on one narrows the others. An exclusive data license may limit how the same records can be used to build a product feature later, so exclusivity, field of use and term need to be weighed against the product roadmap before signing.

AI vendor contracts are the other common collision. Some tools adopted for cost savings include terms that let the vendor use customer inputs to improve its models, which can quietly give away the records the data lever depends on. Review those terms during procurement, not after the pilot.

Management attention is the third constraint. A portfolio CEO running an ERP cutover, a pricing reset and an automation pilot has little room for a rights review as well. Spreading the levers across the hold, with the inventory done once at the start, keeps each company focused on one or two priorities at a time.

Illustrative: a lower-middle-market fund maps its levers#

Illustrative: a fictional fund holds four companies: a property management software vendor, a commercial roofing contractor, a specialty chemicals distributor and a precision machining shop. The operating partner asks each CEO for a one-page record inventory before approving any AI budget.

The software vendor's Jira and Zendesk history links customer issues to fixes across many releases, so it pursues AI features and a licensing review together. The roofing contractor's estimates, inspection photos and warranty claims sit in one field service system, so it starts with estimating automation and schedules a licensing review once records are mapped. The distributor is midway through an ERP migration, so it exports and documents its order exception history before cutover. The machining shop's most detailed records are customer-owned drawings, so it focuses on internal scheduling tools and parks licensing.

How SourceX fits into a portfolio AI plan#

SourceX works only on the data-as-an-asset lever. It does not build internal AI tools, and its own rights in a deidentified dataset are set out in the signed supplier agreement. For each company that proceeds, SourceX runs the SourceX five-step transaction: Supply, Rights, Preparation, Approval and Delivery, and the supplier approves every step.

Records are assessed with the SourceX Enterprise Data Value Framework, a SourceX methodology that gives qualitative ratings on drivers such as uniqueness, domain expertise, human-generated signal, recency, rights and AI utility, while preparation cost and privacy burden reduce net value. Each completed package carries a SourceX Evidence Packet recording provenance, licensing rights, permitted use, the privacy record and release authorization.