AI training for private equity portfolio companies
Why the first AI dollars at a portfolio company belong in training on the team's own work, what it covers, who attends, and how to check it.
King & Company
In short
Surveys of sponsors and operating partners name talent as a main limit on AI at portfolio companies, and BCG's workforce research ties regular use to at least five hours of training with in-person instruction and coaching. A short, role-specific program run on the team's live work should come before licenses are rolled out widely or anything is built. Done that way, it also shows which workflows are worth building and who will own each one.
AI training for private equity portfolio companies should come before a wide license rollout and before any custom build, because sponsors keep naming the same constraint in their own surveys, which is people who know how to use the tools on their own work. A short, role-specific program run on the team's real documents fixes that, and it also tells you what to build next.
What happens when licenses arrive before training
A portfolio company buys seats for an AI assistant after a board meeting. Three months later the CFO looks at the invoice, asks the controller how the close is going, and hears that it takes the same number of days it took before. A few people use the tool to tidy up emails. Most opened it once.
That is an illustration, but operating partners will recognize the shape of it. The licenses show up as cost, and the way the work gets done has not moved. The tool is rarely the problem. Nobody showed the accounts payable lead what it can do with an aging report, nobody said whether customer contracts may be pasted into it, and nobody explained how to tell a right answer from a confident wrong one.
What the sponsor surveys say is holding portfolios back
The surveys sponsors answer themselves point to people. In FTI Consulting's 2026 Private Equity AI Radar, a survey of 200 fund and operating leaders, talent was the primary constraint on scaling adoption, cited by 35% of respondents. The same report puts portfolio companies using AI across use cases at 36% and those at enterprise scale at 7%.
Accordion's PE AI Adoption Benchmark looks at one function, portfolio company finance. Run with Wakefield Research among 150 technology and AI operating partners in May 2026, it ranks data infrastructure as the top barrier at 71%, followed by talent and capability (no AI-literate finance leadership in seat) at 63% and change management at 52%. The benchmark also reports that 63% of portfolio companies have no formal AI structure and work from informal guidance only, that fewer than one in ten PE firms has a fully operational AI center of excellence, and that operating partners rate the shortage of AI-literate finance talent at 4.1 out of 5 as a constraint.
Read together, those numbers describe portfolios where the people who would use the tools have mostly been left to work it out alone. Data quality is a real constraint too, and training does not fix it, but a trained team is the one that can tell you where the data breaks.
How much training does it take before people use AI every week?
The best public evidence on a threshold comes from BCG's 2025 AI at Work survey of more than 10,600 leaders, managers, and frontline employees. It found that regular usage is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching. It also found that only one-third of employees say they have been properly trained, and that regular use among frontline employees had stalled at 51%.
BCG's 2026 edition, covering close to 12,000 people, reports that 74% of frontline employees now describe themselves as AI users. The training gap did not close as usage rose. Close to three-quarters of respondents (72%) say the skills expected of them have shifted, and only 36% feel they have received adequate upskilling. Among frontline employees who use AI regularly, 42% report saving eight hours a week, yet 66% get limited or no guidance on what to do with the time they save.
That last figure matters to a sponsor. Hours saved and never redirected do not reach the P&L. Usage can climb while the value stays on the table, and the missing piece in both years of the survey is instruction and direction from the people who run the work.
We treat five hours as a floor, and we would spread it over several weeks instead of one afternoon, so that people try things on their own work between sessions and come back with questions.
What should the training cover before anything is built?
Before any build begins, the team needs three things settled.
- What the tool is and is not reliable for. People should see it do well on drafting, summarizing, and pulling fields out of a document, and see it fail on a task where it guesses. Both demonstrations should use the company's own material.
- What may and may not be put into it. Customer data, employee data, anything under an NDA, and anything the lender or the sponsor considers sensitive each need a clear answer. That answer belongs in a written policy the company's counsel has reviewed, and our guide to writing an AI acceptable use policy covers how to make it short enough to follow.
- How to check output before relying on it. For a finance team that means tying every number back to the source document and knowing which outputs need a second reviewer before they leave the department.
For a finance function specifically, the material is the close checklist, the variance commentary, the board package, and the reconciliations that feed them.
Train on the team's own work
A course built on generic examples treats training as a classroom event that happens before the real work starts. In our view that kind of training does not carry over to the job, because the distance between a sample exercise and the controller's own close checklist is where people give up.
So the controller brings the actual close checklist. The account manager brings the prep for a renewal coming up this month. The session works through that document, and the person leaves with something they can use the next morning. We cover the mechanics in how to run AI training on your team's own work.
How training tells you which workflows to build first
Training run on live work doubles as discovery. When each person on a finance team brings a task, it becomes clear within a few sessions which ones repeat every month, which depend on documents in a consistent format, and which have a person willing to own the result. Those are the candidates for a build, and the choice is made on evidence from the team instead of a vendor's list of use cases. Our guide to choosing the first AI workflow at a portfolio company picks up from there.
It also leaves the company able to run what gets built. BCG's survey of 100 senior PE investors found that only 45% of successful firms systematically ensure knowledge transfer from external partners to internal teams, and warns that firms without it stay dependent on outside support. A team that was trained before the build, and has a named owner for each workflow, can take the handover.
Who to train first in a portfolio company
| Group | Why they come first | What they work on in the room |
|---|---|---|
| CEO, CFO, or COO sponsoring the effort | They set the data rules and decide where saved time goes | Their own board materials and weekly reporting |
| Managers who own a recurring process | They will own the workflow after any build | The close checklist, renewal prep, the weekly pipeline review |
| The people who do that work each week | They surface where the process really breaks | The specific documents they handle |
Train one function together instead of a few volunteers from each department. A whole team that shares a process can change it, and scattered individuals cannot.
Newer staff deserve seats early. A National Bureau of Economic Research study of 5,179 customer support agents found that access to a generative AI assistant raised productivity by 14% on average and by 34% for novice and low-skilled workers, with minimal impact on the most experienced. That is one setting and one kind of work, so treat it as a reason to include junior people and not as a forecast.
How to tell whether it worked ninety days later
Seat logins will not answer the question. Ninety days out, check four things you can verify by looking:
- Named people are using the tool on a named recurring task, and their manager can point to the output.
- Each trained team can state the data rules without looking them up.
- Reviewers can show how they checked a recent AI-assisted document against its source.
- The team has produced a short list of workflows worth building, each with an owner.
We go further into this in how to measure AI adoption on a team without counting logins.
What the EU AI Act says about AI literacy
Sponsors with European operations have a regulatory reason to document training as well. Article 4 of the EU AI Act says providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and others who operate AI systems on their behalf, taking into account their technical knowledge, experience, education and training, and the context the systems are used in. The current text also states that this obligation does not require guaranteeing any specific level of AI literacy for any individual, which is softer than the earlier wording about ensuring a sufficient level. Whether and how the article applies to a given portfolio company is a question for that company's counsel.
If you are deciding where the first AI dollars go across a portfolio, we are glad to talk through it.
Common questions
Should a portfolio company train its team on AI before buying tools?
In our view the team needs a small number of licenses to train on, and the training should come before a company-wide rollout or any custom build. Sponsor surveys from FTI Consulting and Accordion both name talent and capability as a main constraint, which a license purchase alone does not address.
How many hours of AI training do employees need?
BCG's 2025 AI at Work survey found regular usage is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching. We treat five hours as a floor and spread it over several weeks so people practice on their own work between sessions.
Who should be trained first at a PE-backed company?
Start with the CEO, CFO, or COO who sponsors the effort, then the managers who own a recurring process such as the monthly close or renewal prep, then the people who do that work each week. Training one whole function together is more useful than training a few volunteers from every department.
Why are employees not using the AI licenses we bought?
The usual reasons are that nobody has shown people how the tool applies to their own tasks, nobody has said what data may go into it, and nobody has explained how to check the output. BCG's 2025 survey found only one-third of employees say they have been properly trained.
Does the EU AI Act require AI training for employees?
Article 4 of the EU AI Act, as currently worded, says providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff, and states that this does not require guaranteeing any specific level for any individual. Whether and how it applies to a given company is a question for that company's own counsel.