Private equity and portfolios
Operating partner questions to ask every portfolio CEO about data and AI
By SourceX Editorial · Updated
Short answer
An operating partner should ask every portfolio CEO fifteen questions about data and AI in three groups: records, rights and readiness. Good answers name systems, years and owners; weak answers stay general. Ask the same questions at every company in each review cycle, so answers can be compared across the portfolio and tracked over time.
Key takeaways
- A fixed question set turns scattered AI conversations into a portfolio view that can be compared company by company.
- Records questions come first, because what a company holds limits every AI project and every licensing option.
- A good answer names a system, a range of years or a person; a vague answer is itself a finding.
- Written answers before the review leave meeting time for changes and warning signs.
Why ask every portfolio CEO the same questions?#
Asking every portfolio CEO the same questions turns scattered AI conversations into a portfolio view the fund can act on. Operating partners often cover several companies at once, and without a fixed set each review drifts toward whichever tool a CEO saw at the last conference. Bain's September 2026 piece on the AI value paradox in private equity found little correlation between AI spend and value at most portfolio companies, and named scattered experiments and weak links to value-creation plans as the main causes.
The fifteen questions below sit in three groups. Records questions test what the company holds, rights questions test what it may do with those records, and readiness questions test whether anyone owns the work. Each comes with what a good answer sounds like and a warning sign to listen for.
Records: five questions about what the company holds#
Records questions establish what operating history exists, how deep it runs and whether it connects work to results. A CEO who cannot answer them needs a short inventory exercise, not an AI strategy.
| Question | Good answer sounds like | Warning sign |
|---|---|---|
| Which systems hold our core operating history? | Named systems per function, each with an owner | Only the ERP is named; support and project tools are unknown |
| How many years of that history can we export today? | A range of years per system, confirmed by a test export | The company's age quoted instead |
| Do records link requests to decisions and outcomes? | Tickets link to fixes, jobs to invoices and callbacks | Outcomes live in email or in people's heads |
| Which history did past migrations or acquisitions lose? | A specific list of gaps and where old data still sits | No one has checked |
| Which systems are due to be retired or replaced? | Dates, plus an export plan for each | A retirement planned with no archive step |
Rights: five questions about what the company may do#
Rights questions establish whether the company may use its records for AI projects or licensing, and what it has already promised to customers, vendors and others. Many CEOs have never been asked them, so expect the first round to produce follow-ups for counsel.
| Question | Good answer sounds like | Warning sign |
|---|---|---|
| What do customer contracts say about our use of their data? | Contract types mapped to their data clauses | Contracts never reviewed for data use |
| Which records belong to customers rather than to us? | A clear line, especially for files and content customers create | Everything assumed to be ours |
| Have we shared or licensed records outside ordinary service? | A register with terms and status | Unsure, or handled by someone who has left |
| Which AI tools have staff put company data into? | A tool list with training settings confirmed | Not tracked |
| Do lender or investor documents restrict licensing data? | Counsel has checked the relevant covenants | Never considered |
Readiness: five questions about who owns the work#
Readiness questions establish whether the company can act on what the first two groups reveal. Strong records and clear rights achieve nothing if no executive has the time or the mandate to use them.
| Question | Good answer sounds like | Warning sign |
|---|---|---|
| Who owns data and AI decisions here? | One named executive with time set aside | Everyone shares it, so no one owns it |
| Is there an AI use policy staff actually follow? | A short policy, an approval log and training records | A draft that was never rolled out |
| Which AI projects are live, and what did they change? | Named projects tied to an operating metric | Pilots with no measured outcome |
| What would we need to approve an outside data license? | A known signer, the consents and the review steps | No idea who would sign |
| What records work will we fund this year? | A specific export, cleanup or inventory task | Waiting for the market to settle |
How to run the questions in a quarterly review#
Running the questions in a quarterly operating review works best when CEOs receive them in advance and answer in writing, with the COO or CTO contributing.
Keep the wording stable for at least a full year. Changing the questions every quarter destroys the comparison that makes them useful, and CEOs learn to answer the new wording instead of fixing the old gaps.
- Send the questions with the review pack and ask for written answers.
- Spend meeting time only on changes since the last review and on warning signs.
- Score each group green, amber or red using the same rules at every company.
- Record agreed actions with an owner and a due date.
- Roll the scores into one portfolio view for the investment committee or board.
How to score each group consistently#
Scoring each group consistently means writing the rules down before the first review, so that green at the roofing contractor means the same as green at the software vendor. Score on evidence offered, not on enthusiasm in the room.
Amber is the most useful score. It marks work that is known and bounded, which is exactly what an operating partner can fund or unblock within a quarter.
| Group | Green | Amber | Red |
|---|---|---|---|
| Records | Named systems, tested exports, linked outcomes | Systems known, but exports untested or links partial | History unknown or already lost |
| Rights | Contracts reviewed, customer content separated, register kept | Review started, some contract types unchecked | No review and no register |
| Readiness | Named owner, policy in use, next step funded | Owner named but without time or budget | No owner and no policy |
Illustrative: an operating partner covering five companies#
Illustrative: a fictional operating partner at a lower-middle-market fund covers five companies: a vertical software vendor for property managers, a commercial roofing contractor, a regional food distributor, a precision machining shop and an engineering consultancy. Each CEO answers the fifteen questions before the quarterly review.
The software vendor scores green on records and amber on rights, because its support tickets contain customer exports. The roofer scores green on records and red on readiness, since no one below the owner has time for data work. The distributor reveals that order history from before its ERP migration sits on a legacy server nobody is responsible for. The machining shop's quality records are strong, but customer-owned drawings must be excluded from any outside use.
The operating partner funds an archive export at the distributor, has the roofer name a data owner, and puts the software vendor forward for a metadata-only licensing fit check.
Turning the answers into a portfolio view with SourceX#
A portfolio view built from these answers shows which companies have deep, linked records, clear rights and a named owner, the three conditions any data or AI program needs. The SourceX Enterprise Data Value Framework explains why linkage, depth of history and recorded outcomes drive how useful operating records are to AI developers.
For companies that move forward, SourceX runs each one through its own SourceX five-step transaction, with the portfolio company's authorized signer approving every step. Nothing is shared during the initial assessment, which makes it easy for a cautious CEO to take part.
Frequently asked questions
Should the CEO answer personally or delegate?
The CEO should own the answers, but the COO, CTO or controller will often draft them. What matters is that the CEO reads and stands behind them, because several questions concern priorities and authority that only the CEO can settle.
How often should the questions be asked?
Quarterly works for most portfolios, tied to the regular operating review. Ask the full set once a year and focus the other reviews on changes, open actions and warning signs, so the exercise stays light for management teams.
What if a portfolio CEO is skeptical about AI?
The questions still apply, because they concern records, rights and ownership, which matter for reporting, integration and exit as much as for AI. Framing them as records hygiene often draws better answers than framing them as an AI initiative.
Should the answers go into board materials?
A summary should. A board that oversees AI spending benefits from a view of data and AI risk and opportunity. A one-page roll-up of scores and actions is usually enough, while the detailed answers stay with management and the operating team.
Can answers be shared across portfolio companies?
Scores and lessons can, and sharing them helps CEOs see what good looks like. Detailed answers about contracts, customers or incidents should stay within each company and the operating team, because portfolio companies are separate businesses with their own confidentiality obligations.
Sources
- Bain's September 2026 piece says that for most portfolio companies there is little correlation between AI spend and value, naming scattered experiments and weak links to value-creation plans as main causes. Source
Related resources
See if your company qualifies
A short company assessment. No data uploads are needed.