Where to start with AI at a portfolio company

Start with one recurring workflow, chosen because a named person owns it and wants it fixed, and build it against the files that person already uses. A readiness score and a ranked list of twenty use cases can come later, once one thing is working and the team has seen it.

That is a deliberately small starting point for a 100-day plan. We think it is the right one, and the published surveys of sponsors and operating partners help explain why.

What is the sponsor asking for when it asks for an AI plan?

The sponsor wants two things. It wants evidence at the next board meeting that the company is moving, and it wants a story at exit that a buyer will accept. Both depend on something specific and documented.

The obstacle is rarely a shortage of ideas. In a BCG survey of 100 senior PE investors published in January 2026, 90% named competing priorities as the top blocker to digital transformation and 76% pointed to unclear ROI. A management team in its first hundred days is also running a new reporting calendar, a lender relationship, and a value creation plan. An AI plan that asks for a quarter of executive attention before anything works will lose to those priorities.

Why is a readiness assessment the wrong first step?

Standard advice starts at the top. Bain's 2024 guidance to sponsors was to scan the portfolio and link AI initiatives to clear strategic objectives such as customer satisfaction, revenue generation, or cost reduction, and that is sound at the fund level. Inside one company, the same instinct usually becomes a scored assessment and a prioritized list of use cases with an estimated EBITDA figure beside each one.

The assessment is the inexpensive part of the work, and it leaves the hard question open. A ranked list cannot tell you whether the controller will use the thing on the third business day of close, when the old spreadsheet is sitting right there. Only a working draft, built on her files and put in front of her, can tell you that.

An assessment also tends to conclude that the data needs work first. That conclusion is usually true. Accordion, a finance consulting firm that works with sponsors, surveyed 150 operating partners in May 2026. They reported that 34% of portfolio companies arrive unprepared on data at acquisition and need twelve months or more of remediation, another 41% are only partially ready, and data infrastructure is the top barrier to scaling AI, named by 71%. A plan sequenced behind that remediation puts the first visible result a long way out.

Five tests for a good first workflow

Five tests separate a good first workflow from a tempting one. A candidate should pass all of them.

1. It recurs weekly or monthly. A task that happens once a year gives you one chance to tune it and nothing to measure.
2. Expensive people do it. The time you return should belong to a controller, an FP&A lead, an account manager, or an executive.
3. It runs on documents and exports the company already has. Trial balance exports, the prior board deck, the contract folder, and the aging report all qualify. Anything that needs a new integration before it can start does not.
4. A named person can check the output in minutes. If review takes as long as doing the work, nothing has been gained.
5. The owner wants it fixed. A workflow assigned to someone who did not ask for it gets used while the sponsor is watching and then dropped.

Where are portfolio companies starting?

The clearest published data is on the finance function, and within it the reporting cycle. Accordion's benchmark lists the use cases operating partners report as live in portfolio finance functions: covenant monitoring and alerts at 58%, budget variance detection at 53%, board package drafting at 44%, and close orchestration across ERP and consolidation systems at 39%.

Those map onto work a mid-market finance team will recognize.

Workflow	What goes in	Who checks it
Board package first draft	Monthly financials, KPI export, last month's package	CFO
Variance commentary	Budget and actuals by department, GL detail	Controller or FP&A lead
Contract and renewal prep	Customer contract, usage or billing export, account notes	Account owner
Collections follow-up	Aging report, invoice history, prior correspondence	AR lead

The first two rows pass the five tests at most companies because the inputs already exist in a folder and the reviewer is the person who would have written the document anyway. Close orchestration across systems is a larger build and usually belongs later.

What can be working in one quarter without a data project?

One workflow in daily or monthly use, with its owner trained, is a realistic result for a quarter. A second and third can follow in the same period if they share inputs with the first. Our own engagements run eight to twelve weeks with a working session each week, and each session centers on a draft of the workflow built from the company's real documents.

Three practices keep it inside the quarter. Build inside the company's own environment and AI workspace so the question of where the data lives is settled on the first day. Use the last three months of real files as the test set, so the draft can be compared with what the team produced by hand. Train the owner during the build, so the person is already running it by the time it is finished. We explain the reasoning in why AI training comes before AI tools at a portfolio company.

How do you design the review step so the output can go to the board?

Nothing produced by the workflow should reach the board, a lender, or a customer until a named person has reviewed it. In practice the draft arrives with its sources attached. Every figure in the variance commentary points to the export line it came from, and every sentence that makes a judgment is marked for the CFO to accept or rewrite.

The reviewer's corrections are also the best input for tuning the workflow. A fuller treatment is in our guide to designing the human review step in an AI workflow.

How do you measure it in a way a buyer will believe?

Sponsors themselves say this part is hard. In the Q4 2025 wave of EY's AI Pulse survey, 31% of PE respondents strongly and another 31% somewhat acknowledge that their organizations grapple with linking specific productivity gains to AI adoption. Accordion finds 18% of operating partners are not yet measuring AI impact systematically. FTI Consulting's 2026 survey of 200 fund and operating leaders reports that 95% of funds say their AI initiatives met or exceeded the original business case, and notes that those cases were often conservatively scoped, which is a reason to read headline success rates with care.

The measurement that survives scrutiny is a before-and-after on one workflow, recorded at the time:

• Hours per cycle, by person, for the two or three cycles before the build
• Cycle time from inputs available to output delivered
• Corrections the reviewer makes per cycle, before and after
• What the returned hours were spent on

The last item matters most to a sponsor. EY's report says that value is created only when leaders deliberately redeploy the freed-up capacity to a defined and measurable KPI in the value creation plan.

The same record becomes exit evidence. BCG found that 82% of firms track ROI from digital initiatives while only 11% explicitly link digital progress to exit narratives. Buyers are already asking. In Accordion's survey, 44% of operating partners say buyers ask about AI in the finance function during diligence, though the premium is not yet in the price. A dated log of one workflow gives a buyer something to test, which a slide of estimates does not.

Keep the statements specific for a second reason. In March 2024 the SEC charged two investment advisers with making false and misleading statements about their use of AI, and the firms agreed to pay $400,000 in total civil penalties. That action concerned advisers and their marketing, and whether similar exposure applies to anything your company or sponsor says is a question for your own counsel. The practical habit is the same in any case: describe what the workflow does, who reviews it, and what was measured.

What should happen in days 100 to 365?

Extend from the first workflow to its neighbors. A company that started with variance commentary has most of what it needs for the board package draft, and then for the lender reporting that draws on the same numbers. Each addition gets its own baseline and its own owner.

This is also the point where data work earns its budget, because the team can now name the specific export or system that is holding the next workflow back. For the operating partner, a build that works at one company becomes a candidate pattern for others, which we cover in how to repeat one AI build across a private equity portfolio. The wider picture across deal teams and portfolio operations is in how private equity firms use AI.

Which mistakes slow the first workflow down?

• Sequencing everything behind a data platform. The first workflow can run on exports.
• Buying licenses for everyone and counting logins. Seats are an input, and the board will ask what changed.
• Starting with a customer-facing chatbot. It fails the review test, since nobody checks the output before a customer sees it.
• Assigning the work to IT alone. The owner has to be the person who does the work today.
• Reconstructing the baseline afterward. Record the hours before the build begins.
• Reporting an estimated EBITDA figure for work that is not yet in use. A measured result on one workflow is more credible than a projection across ten.

If you are writing the AI section of a 100-day plan and want a second opinion on which workflow to pick first, get in touch.

Common questions

What is the best first AI use case for a PE-backed company?

The best first use case is a piece of work that recurs weekly or monthly, is done by senior people, runs on documents and exports the company already has, and produces an output a named person can check in minutes. At many companies that is the board package first draft, variance commentary, contract or renewal preparation, or collections follow-up.

Do we need to fix our data before starting with AI?

Not for the first workflow, if you choose one that runs on the exports and documents the finance or operations team already works from. In Accordion's May 2026 survey, operating partners reported that 34% of portfolio companies arrive unprepared on data at acquisition and need twelve months or more of remediation, so a plan that waits for a clean data layer can wait a long time. Data work matters more once you try to scale past the first few workflows.

How should a portfolio company measure AI ROI?

Measure one workflow before and after: the hours it takes, the cycle time from start to finished output, and the number of corrections the reviewer makes. Record the baseline before the build starts and keep the record dated, because a baseline reconstructed later is much harder to defend.

Will a buyer pay more for a company that uses AI?

The evidence so far is mixed. In Accordion's May 2026 survey, which covered AI in the finance function, 44% of operating partners said buyers ask about it in diligence but the premium is not yet in the price, and 9% had seen a demonstrable premium in completed transactions. BCG's survey of PE investors found 40% had experienced a valuation haircut of 5% or more when digital maturity lagged or was underinvested, so documented evidence is worth keeping either way.

Who should own AI at a mid-market portfolio company?

Each workflow should be owned by the person who runs that work today, such as the controller for the close and the board package. One executive, usually the CFO or COO, should own the sequence of what gets built next and the reporting to the sponsor.