Private equity and portfolios
AI diligence questions acquirers ask in 2026
By SourceX Editorial · Updated
Short answer
AI due diligence questions in M&A fall into four groups: data rights, third-party restrictions, AI use and licensing history. Acquirers ask them to learn whether a target's records and AI tools can be used, transferred and built on after closing. The working rule: a gap with a document trail is fixable; a gap without one becomes a deal term.
Key takeaways
- Ask what records the target controls before asking about its AI features, because rights decide what AI work is possible.
- Change-of-control clauses in data feeds, platform APIs and vendor contracts can cut off data the business depends on.
- Licensing history matters as much as AI use, because exclusivity and deletion obligations transfer with the company.
- Trades contractors, distributors and manufacturers need the same questions, phrased in operational language.
Why acquirers now ask separate AI diligence questions#
Acquirers ask separate AI diligence questions because traditional technology and IP reviews rarely ask what records the target holds, what it has promised about them, or which AI tools already process them. Those answers shape both the risk picture and the value creation plan after closing.
The question bank below is grouped the way findings usually turn into deal work: data rights, third-party restrictions, AI use and licensing history. Each group can go into the request list as one item with follow-ups, which spares the target's management team from answering the same question four times in different workstreams.
Data rights questions#
Data rights questions establish who owns or controls the target's records and under which terms they were collected. Strong answers come as tables and documents; weak answers come as reassurance.
| Question | What a strong answer shows | Follow-up if the answer is weak |
|---|---|---|
| Which systems hold customer, operational and engineering records, and since when? | A named system list with date coverage per system | Ask which history was lost in migrations |
| What do customer contracts say about your use of their data? | Contract types mapped to data clauses, with exceptions listed | Request the clauses for the largest accounts |
| Which records belong to customers rather than to you? | A clear line between customer content and company records | Ask how deletion and return requests are handled |
| Which privacy notices applied when personal data was collected? | Notice versions with effective dates | Ask which records predate the current notice |
| Do records from acquired businesses carry their own terms? | A per-entity list of legacy terms and notices | Request the legacy documents directly |
Third-party restriction questions#
Third-party restriction questions find data the business depends on but does not fully control. Data feeds, software vendors, platform APIs and channel partners often supply data on terms that limit use, prohibit model training or end on a change of control.
For operators, the franchise and OEM questions often matter most. A home services platform may learn that job records sit in a franchisor-controlled system, and a parts distributor may learn that much of its catalog content belongs to manufacturers.
- Which outside data sources feed your product, pricing or operations, and under what license?
- Do any of those licenses prohibit machine learning, model training or derivative datasets?
- Which vendor, partner or customer contracts terminate, or require consent, on a change of control?
- Do your software vendors claim any right to train on data you store in their systems?
- Are any records governed by franchise, dealer or OEM agreements that give another party control?
AI use questions#
AI use questions map where AI already touches the business and whether that use was governed. They apply to every target, including those with no AI in the product.
Ask for the tool list in the same format as the IT systems inventory so the two can be reconciled. Tools that show up in expense or single sign-on data but not on the list are the ones to follow up, because they were adopted without review.
| Question | Why it matters |
|---|---|
| Which AI tools and features are in use, by team? | Shows exposure and the integration work after closing |
| Has customer, employee or confidential data gone into tools that train on inputs? | Identifies possible loss of confidentiality |
| Does the product include AI features, and which model providers power them? | Reveals dependencies and subprocessor obligations |
| Has the company trained or fine-tuned any model, and on what data? | Tests rights in both the training data and the model |
| Is there an AI use policy, and is it followed in practice? | Separates governance on paper from governance in use |
| Have customers complained about or restricted AI use? | Signals coming contract changes and trust issues |
Licensing history questions#
Licensing history questions uncover what the target has already granted to others. A prior data license can be an asset, but exclusivity, broad permitted use or ongoing delivery obligations transfer with the business and may narrow the acquirer's plans.
- Has the company licensed, shared or provided access to any records, including through aggregated data programs?
- For each license: counterparty, record types, exclusivity, field of use, term and renewal.
- What happens to each license on a change of control?
- What deletion, return or audit obligations apply when a term ends?
- How is license revenue recognized, and is it recurring or one-time?
- Is there a document trail of rights review and approval for each license?
How to weigh what the answers reveal#
Weighing the answers means separating findings that can be fixed before closing from findings that must be priced or allocated in the purchase agreement. The difference is usually documentation, not the underlying facts.
How a finding becomes a representation, an indemnity or a closing condition is a drafting decision for counsel on each transaction. The table shows common patterns, not rules.
| Finding | Usually fixable before close | May become a deal term |
|---|---|---|
| AI tools used without a policy | Adopt a policy and switch off training settings | Rarely |
| Customer data entered into a training-enabled tool | Document it and request deletion | Targeted representation if the extent is unclear |
| Data feed prohibits model training | Plan around the restriction | If the value creation plan depends on that data |
| Prior exclusive data license | Seldom | Price, consent or carve-out of the licensed field |
| No record of rights review for a prior license | Reconstruct the file where possible | Representation on rights and permitted use |
| Legacy terms from acquired entities unknown | Collect the documents | Limits on post-close use of those records |
Illustrative: a fund tests a vertical SaaS target#
Illustrative: a fictional lower-middle-market fund is buying a vertical SaaS company that serves independent auto repair shops. The operating partner puts all four question groups into the first request list.
The answers show deep engineering and support history in Jira and Zendesk, a parts pricing feed whose license prohibits model training, and an existing non-exclusive license of de-identified support records to an AI developer, with a clean approval trail and terms that survive a change of control. The deal team treats the license as a documented asset, plans the pricing roadmap around the feed restriction, and asks counsel for a targeted representation on customer content inside support attachments.
Using diligence answers after closing with SourceX#
Diligence answers become the starting file for any post-close licensing work. When a portfolio company later considers licensing operational records, SourceX begins the SourceX five-step transaction from the same system list, contract map and licensing history, so the Rights step builds on diligence instead of repeating it.
For prior licenses, SourceX looks for the elements of a SourceX Evidence Packet: provenance, licensing rights, permitted use, the privacy record and release authorization. Where those elements are missing, they are reconstructed or the affected records stay out of scope.
Frequently asked questions
Who on the deal team should own AI diligence?
Usually the operating partner or the technology diligence lead coordinates, with legal counsel owning rights and contract questions. What matters is a single owner who merges findings from technology, legal and commercial reviews, because AI and data issues tend to fall between workstreams.
How is AI diligence different from technology diligence?
Technology diligence looks at architecture, security, scalability and technical debt. AI diligence asks what data the business may use, what it has promised about that data and how AI already processes it. The two overlap on systems, but the questions and the evidence differ.
Do these questions apply to non-software targets?
Yes. HVAC contractors, distributors and manufacturers use AI tools and hold job, order and quality records with real value. Phrase the questions in their terms, such as dispatch history, order exceptions or nonconformance reports, and expect less formal documentation.
What documents should we request first?
Start with the systems list, standard customer terms and major negotiated contracts, privacy notices with dates, the AI tool list, and any data license or data sharing agreement. Those documents answer most of the question bank and show where to dig further.
Should AI questions go to the seller before the letter of intent?
A short version can. Before a letter of intent, ask only for the systems list, any data licenses granted and whether AI tools have processed customer data. Those answers show whether deeper questions could change price or structure, which is worth knowing before exclusivity begins.
Related resources
- IndustryHealthcare data
- QuestionDo AI companies buy private business data?
- InsightCan restaurants and hospitality groups sell their data to AI companies?
- InsightCan SaaS companies sell their data to AI companies?
- InsightSelling internal documentation to AI companies: what to know
- SolutionData licensing: granting defined rights to use your data
See if your company qualifies
A short company assessment. No data uploads are needed.