AI Questions for Portfolio Company Boards: What PE Directors Should Ask Management Before 2027

1. The Short Answer: 20 Questions in Four Groups

A portfolio company board can cover AI with 20 questions to management, five in each of four groups: value, spend, risk and people. Ask all 20 once, before the 2027 budget is approved. After that, ask for one page a quarter that reports five numbers against the same groups.

The groups follow the money. Value asks which P&L lines AI moves. Spend asks what it costs, all in. Risk asks what could go wrong, including whether AI threatens what the company sells. People asks who owns the work and who has learned to do it.

Questions alone change little. Written answers, with numbers and names, give the board something to check next quarter.

2. Why Ask Now

For a company on a calendar year, the 2027 budget is being written now. Whatever AI line goes into it will get spent, so the board's best chance to shape it comes before approval.

The record so far is mixed. Bain & Company wrote in September 2026: "While AI activity is everywhere in private equity, meaningful value creation isn't." In Bain's 2026 survey of 100 CEOs, 82% said their AI programs were at best partially delivering, and 18% had reached most or all of their ambition.

Investors are pushing anyway. On September 17, 2026, UK Private Capital, formerly the BVCA, published a survey of 76 senior private capital investors: "Nine in ten private capital firms are encouraging the use of AI across their portfolio companies." Guidance lags the push. While 93% encourage AI in the back office, 45% give portfolio companies guidance on AI governance, risk and compliance.

Boards talk about AI more than they steer it. In NACD's 2025 survey of 65 private company boards, 62% set aside full-board agenda time for AI, and 25% had adopted an AI governance framework. Directors expect a lot all the same. NACD's 2026 Governance Outlook reported: "Seventy-six percent of directors say AI will factor into their 2026 growth strategy, but most organizations report only slight or moderate success in realizing operational or financial benefits from these investments."

Agenda time is the easy part. Questions with numbers attached turn it into oversight.

3. Questions on Value

Start with value, because every other answer depends on it. Bain advises picking three to five opportunities from the value-creation plan, and it lists use-case swirl and the micro-productivity trap among its five failure modes.

One of Bain's examples shows the discipline. At a PE-owned hospitality property manager, "management had identified 30 possible processes that might benefit from AI intervention, it quickly narrowed the list to 4 linked closely to revenue improvement." Bain says the company is on track for a 40% to 50% EBITDA increase by 2030.

1. Which three to five AI use cases are in the value-creation plan, and which P&L line does each one move?
2. What was the baseline before each one started, and who measured it?
3. Which use case will show up in the numbers first, and in which quarter?
4. Which AI pilots did we stop this year, and why?
5. How much of the 2027 EBITDA plan depends on AI, and what happens to the plan if it slips?

Question 4 sounds odd until you see the alternative: a company that never stops a pilot is collecting them. MIT's Project NANDA, in a preliminary July 2025 report based on 153 surveys and 52 interviews, found: "Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact." BCG reported a similar split in September 2025: "Only 5% of companies in our 2025 study of more than 1,250 firms worldwide are achieving AI value at scale."

Samples matter. In RSM's 2026 survey of 1,030 middle-market companies already using AI, 54% said it beat their ROI expectations. Either way, the board's job is the same: ask which line moved.

The quote below sets the standard for the answers. One use case on the P&L outweighs a stack of pilots.

4. Questions on Spend

Bain found that "for most portfolio companies, there's little correlation between AI spend and value" (Bain & Company). A bigger AI budget proves nothing on its own, so ask what each dollar is for.

Few boards see the number. In NACD's 2025 survey, 17% of private company boards had approved an AI budget.

6. What do we spend on AI today, all in: licenses, usage, vendors, consultants and staff time?
7. How many of the seats we pay for are used every week?
8. Which of our existing software contracts now include AI features or AI charges, and who agreed to them?
9. What is the AI line in the 2027 budget, and which result from questions 1 to 5 is it tied to?
10. Who can approve new AI spend, and above what amount does it come to the board?

Question 7 is cheap to answer and often uncomfortable, because a seat nobody opens is pure cost. Bain lists tools-first bias among its failure modes; in a budget, it can look like licenses bought for everyone before anyone named the job.

Question 9 matters most. A budget line with no result attached will still get spent.

5. Questions on Risk and Policy

Risk starts with where company data goes. Anyone with a personal AI account can paste a customer list into it today, which is how shadow AI starts. In NACD's 2025 survey, 17% of private company boards had audited where AI is used in the company.

Investors push less here. UK Private Capital found that "the use of AI in compliance, legal and risk management functions was less widespread, reflecting the greater regulatory and governance considerations associated with these activities." Encouragement in those functions was 71%, against 93% for the back office.

The legal calendar is filling up, and counsel decides which of these apply to the company:

• EU AI Act. As amended by Regulation (EU) 2026/1744, in force since July 27, 2026, the high-risk dates are "2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III" and 2 August 2028 for systems under Article 6(1) and Annex I.
• Colorado. SB 26-189, signed May 14, 2026, repealed and replaced the Colorado AI Act, and "this act takes effect January 1, 2027."
• California. Civil Rights Council regulations on automated-decision systems in employment took effect October 1, 2025. Under the CPPA's rules, "A business that uses ADMT for a significant decision prior to January 1, 2027, must be in compliance with the requirements of this Article no later than January 1, 2027."
• Texas and Illinois. Texas TRAIGA (HB 149): "This Act takes effect January 1, 2026." Illinois HB 3773, which amends the Illinois Human Rights Act on AI in employment, took effect the same day (Illinois Human Rights Commission).

11. Which AI tools do our people use for company work, on which accounts, and with what data?
12. What does our AI policy say, who has signed it, and when was it last reviewed?
13. Which AI laws and customer contract terms apply to us, and by what dates? We want counsel's answer in writing.
14. When AI gets something wrong in front of a customer, a lender or a regulator, who catches it, and how would this board hear?
15. Could an AI agent do what we sell, and what is management's plan if customers start trying?

Many mid-sized companies already put controls first: in RSM's survey, 67% apply governance controls before a pilot or production. For the European rules in more depth, see the EU AI Act portfolio-company checklist.

Question 13 has a deadline inside it. Two of the dates above fall on January 1, 2027.

6. Inside Risk: Could an AI Agent Do What the Company Sells?

Question 15 looks outward. The other risk questions ask what AI could do to the company's data; this one asks what AI could do to its revenue.

For a software business, the threat is direct. The Bank for International Settlements wrote in July 2026 that "software firms' revenues are now highly vulnerable to AI, which can both substitute for existing products and lower development costs, reducing barriers to entry and intensifying competition." A services business that bills by the hour faces its own version of the question.

Ask management to bring evidence to the answer:

• Win-loss notes that mention AI tools or AI-native competitors.
• Renewal and churn reasons, in customers' own words, from the last four quarters.
• Pricing exposure: how much revenue is priced per seat, per hour or per head, and so shrinks if customers need fewer people.
• What the company owns that an agent would struggle to copy: proprietary data, regulatory approvals, workflow lock-in, relationships.
• The product plan: what the company is building with AI, and what it would cost a new entrant to build the same thing.

Deal teams run the same test before buying a company, and it belongs on the board agenda after closing too. The guide to assessing AI disruption risk in a target sets out the method.

A company can run a flawless AI policy and still lose its market to an agent. The board is the place to ask about both.

7. Questions on People

People is the group most boards skip, and the other three depend on it. In NACD's 2025 survey, 14% of private company boards had a director education plan. Among the investors UK Private Capital surveyed, 50% supported management education and 45% helped with talent.

Two of Bain's five failure modes, capability gaps and sponsorship gaps, are people problems. In Europe, part of this sits in law. Article 4 of the EU AI Act, as amended, reads: "Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf" (Regulation (EU) 2026/1744). Counsel decides whether it applies.

16. Who owns AI at the company, by name, and how many hours a week does it get?
17. How does the CEO sponsor it, and what did the leadership team use AI for last month?
18. Who has been trained, on what, using the company's own work, and what changed afterwards?
19. Which roles and hiring plans in the 2027 budget assume AI, and how will the people in those roles hear about it?
20. What does this board need to learn to oversee AI, and who will teach us?

The owner should be someone who already runs a process, such as the CFO, the controller or the head of operations, with the work written into their goals. Add a champion in each function who tries new tools first. The guide on who should own AI covers the choice. Question 19 touches the employment rules in section 5, so counsel should see that answer too.

Question 20 turns the list around. A board that cannot judge the answers to the first 19 should fix that before it asks them.

8. What Good Answers Sound Like, and the Red Flags

A good answer names a number, a person and a date: order inquiries, measured in minutes, owned by the head of customer service, live by March. A promise to roll out AI across customer service gives the board nothing to check.

Bain's parts distributor is the shape to look for: "AI radically shortened the average turnaround time for order inquiries from more than an hour to less than five minutes" (Bain & Company). One workflow, one measure, before and after.

Bain's five failure modes make a useful list of red flags. Here is how each might show up in a board meeting:

• Use-case swirl. Management lists pilots in every department, and none is tied to a P&L line.
• The micro-productivity trap. Hours saved are reported, and they never show up in cost or revenue.
• Tools-first bias. Licenses went to the whole company before anyone named the job.
• Capability gaps. People are left to figure it out, with no training on the company's own work.
• Sponsorship gaps. AI sits with IT, and nobody in the business owns the result.

Add one of your own: a report of no incidents. Ask how management would know.

Red flags help only if someone follows up. Put each follow-up in the minutes with a name and a date.

9. The One-Page Quarterly AI Report

Once the 20 questions have written answers, the board needs a way to track them without another long session. In NACD's 2025 survey, 15% of private company boards had set management reporting metrics, and among public company boards the figure was 6% (NACD).

The page carries the same five numbers every quarter:

1. Value delivered. The P&L effect of each named use case against its baseline, in dollars, or in hours on a measured task until the dollars show.
2. Spend against budget. Licenses, usage, vendors and consultants, all in, with the budget beside it.
3. Active users. Weekly active users against seats paid for.
4. Incidents. Policy breaches, data incidents and AI errors that reached a customer, with what changed as a result.
5. People trained. Staff trained on their own work against the plan, and whether the owner still has the hours.

Below the numbers, a few lines of news: a law that now applies, a tool approved or retired, a customer asking about AI. Operating partners who want the same numbers from every company can borrow the scorecard in the operating partner's Claude kit.

Read the page for trends. A number flat two quarters running is next quarter's agenda item.

10. Where to Start

Before the 2027 budget meeting, send management the 20 questions and ask for written answers.

1. Put AI on the budget meeting agenda as its own item.
2. Send the questions two weeks ahead, so the answers come with numbers.
3. Ask counsel to answer question 13 in writing.
4. Agree the five numbers and who reports each one.
5. Put the first one-page report on next quarter's agenda.

Short on time? Ask questions 1 and 16 first: what AI is for, and who owns it. For a newly acquired company, the 100-day playbook shows where AI fits in the value-creation plan.