Insights
Portfolio AssessmentNovember 14, 20264 min read

How to run an AI maturity assessment across a private equity portfolio.

Five dimensions, a separate rating for opportunity, and the grouping step that turns scores into a first wave.

An AI maturity assessment scores every portfolio company on how ready it is to get value from AI, then uses the scores to decide where the fund should put effort first. The scores are a means to an end. The output that matters is a ranked first wave of companies and use cases, and a short list of things to stop.

A fund with twenty or thirty companies cannot run an AI programme in each one at the same time. The assessment exists to answer a sequencing question: which three or four companies go first, with which use case, and why.

Score five dimensions.

Keep the model simple enough that a CEO can check their own score in a few minutes.

  • Data. Is operational data owned, accessible and reasonably clean? Can it be used under customer contracts?
  • Process. How much of the work is repetitive, high-volume and rule-bound? Are the processes documented or held in people’s heads?
  • People. Is there a leader who wants this, and engineers who have shipped with current models?
  • Product. For software and data businesses, does AI strengthen the product or threaten it?
  • Governance. Is there a policy, a view of which tools staff use, and a way to review model output?

A four-point scale per dimension is enough. Finer scales suggest a precision the evidence does not support.

Score opportunity separately from readiness.

Readiness alone ranks the wrong companies first. A well-run software business may score highly and have little cost to take out. A services business with a large back office may score poorly on data and still hold the largest prize.

So give each company a second rating for the size of the opportunity: the share of cost that sits in document, data and review work, and how exposed revenue is to competitors or customers using AI. The first wave comes from companies with a large opportunity and enough readiness to start, not from the top of the readiness table.

Where the evidence comes from.

Start outside-in. Websites, product pages, job postings, customer reviews and published materials are enough for a first score on every company and take days, not months. This first pass is useful mainly because it is fast and treats every company the same way.

Then test the scores with management. A structured hour with each CEO and one operations or technology lead corrects most errors. Say clearly which scores are evidence-based and which are still estimates, and keep that label on the page until someone has checked.

Group companies by pattern.

The step that separates a fund-level assessment from a set of company reports is grouping. Look for the same process appearing in several companies: document intake in services firms, scheduling in field businesses, support in software companies, data checks in information businesses.

Where a pattern repeats, one design can be built once and deployed several times. That is where a fund gets a return a single company cannot: the second and third deployments cost less and carry less risk than the first.

What the output should contain.

  • A one-page view of every company on readiness and opportunity.
  • The groups, with the repeatable use case for each.
  • A first wave of three or four companies, each with a named use case, a rough range and an owner.
  • A twelve-month sequence for the rest.
  • A stop list: pilots and tool spend that are not leading anywhere.

How long it takes.

Three to six weeks is realistic for a mid-sized portfolio: a week or two for the outside-in pass, two or three for management conversations, and a week to group, rank and write up. An assessment that takes a quarter has usually turned into a consulting project.

Common mistakes.

  • Treating the score as a league table. CEOs will argue about rank and ignore the plan.
  • Scoring only what companies say about themselves.
  • Ranking by readiness and missing the larger prize in less tidy businesses.
  • Producing a report per company with no view across the portfolio.
  • Ending with a roadmap and no owner for the first use case.

What happens next.

The assessment should lead straight into a build in the first-wave companies, on a timetable short enough that the fund sees a measured result within a quarter. If it ends in a presentation, it was the wrong exercise.

The Portfolio AI Assessment page describes how I run this for a fund.