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

The three levers of AI value creation#

The three levers of AI value creation are cost out, tech-enabled re-rating, and owned data and AI assets. Most playbooks stop at the first, which is where software vendors focus; the second and third depend on records the company already holds and on its right to use them.

Each lever has a decision rule. Applying the rules company by company keeps the program from funding tools where records are thin, or promising an AI story the company cannot support in diligence.

Lever one: cost out#

Cost out is the lever most operating partners start with: support deflection, faster quoting, dispatch optimization, invoice processing and drafting help for professional staff. It works when the company has years of records showing how the work was done and how it turned out.

The common failure is buying a tool before checking the records. A support assistant needs resolved tickets with clear outcomes; a quoting assistant needs past quotes tied to won or lost jobs. Without them, the tool is configured on guesses and there is no baseline to measure savings against.

Lever two: tech-enabled re-rating#

Tech-enabled re-rating is the argument that a company has become more technology-driven and deserves to be valued that way. Buyers test it in diligence by asking where AI runs in daily work, which records it relies on and whether those records belong to the company.

The decision rule keeps the claim honest. Present AI enablement only where it is embedded in workflows and supported by proprietary history the company has the right to use. A pilot in one department, or a feature built on a general model with no company records behind it, is easy for a buyer to discount.

Lever three: owned data and AI assets#

Owned data and AI assets is the lever that treats operational records as something the company can license, not only use. Support conversations, CRM histories, engineering workflows, dispatch records and quality logs can be licensed to AI developers for defined uses while the company keeps ownership.

This lever can produce license revenue inside the hold and a documented asset at exit, though neither is assured until a buyer engages with a defined scope. It also disciplines the other two levers, because the rights review it requires tells the company which records it can safely use for internal AI as well.

Sequencing the playbook across a hold#

Sequencing matters because the three levers share one input: the company's records. Run the record review first, then fund the levers in the order each company's records support.

• 1Review records at each company: systems, accessible history, linkage to outcomes and rights.
• 2Rank companies by record strength and by the lever each one fits best.
• 3Pick one cost-out project and one licensing candidate as portfolio pilots.
• 4Set governance: an AI owner at each company, a tool inventory and approval rules.
• 5Report progress to the board with record-based measures, not tool counts.
• 6Prepare evidence for exit: rights reviews, privacy records and license documentation.

Illustrative: one operating partner, four companies, three levers#

Illustrative: a fictional operating partner covers four portfolio companies: a construction scheduling software vendor, a food-grade contract packager, an operations consulting firm and a commercial electrical contractor. The record review places each one differently.

The software vendor's linked Jira issues, pull requests and support tickets make it the licensing candidate and the strongest re-rating case. The contract packager's ERP order exceptions and changeover logs support a cost-out project in production scheduling. The consulting firm's proposals and project reviews look promising, but many records sit inside client deliverables, so rights work comes first. The electrical contractor's estimates could support a quoting assistant, except that won and lost outcomes were never recorded, so the first step is fixing how estimates are closed out.

The partner funds one cost-out pilot at the packager, starts a fit check at the software vendor and defers the rest until both pilots report back to the board.

What to report to the investment committee#

Investment committee reporting should tie each lever to a measure drawn from company records, not to the number of tools deployed. Counts of pilots and licenses say little on their own; changes in work measured against history say more.

Agree the measures before a project starts, and pull the baseline from the same system that will report the result. A help desk that changes ticket categories midway through a pilot makes the before-and-after comparison unreliable.

Mistakes that stall AI value creation programs#

The mistake that stalls most AI value creation programs is funding tools before anyone has checked the records they depend on. A portfolio-wide license for a drafting assistant shows activity on a board slide, but without linked history there is nothing to ground the tool in and no baseline to prove savings.

The other mistakes follow from the same root: treating AI as a purchase rather than as a use of records the company holds and has the right to use.

• 1Counting pilots and seats instead of measuring work against history.
• 2Calling a company AI-enabled in the equity story when AI runs in one department or on a general model with no company records behind it.
• 3Letting integration teams retire acquired systems before their history is exported.
• 4Signing an exclusive data license that blocks the company's own internal AI plans.
• 5Running a separate record review for each lever, so the same contracts are read three times.

How SourceX supports the third lever#

SourceX supports the owned data and AI assets lever. The SourceX Enterprise Data Value Framework, a SourceX methodology with qualitative ratings rather than prices, weighs drivers such as uniqueness, domain expertise, human-generated signal, recency, rights and AI utility against preparation cost and privacy burden, which shows which record families are worth taking further.

The SourceX five-step transaction, Supply, Rights, Preparation, Approval and Delivery, then carries a license from fit check to delivery with the company approving every step. Nothing is shared during the initial assessment, and SourceX's dataset rights are set out in the signed supplier agreement.