The AI due diligence checklist every PE operating partner needs
Most private equity operating partners are walking into management presentations where the word "AI" appears on every third slide. Some of those claims are real. Most are not. The problem is that very few deal teams have a structured way to tell the difference.
The mistake firms make at this stage is treating AI capability as a qualitative signal - something you either believe or you do not. That is the wrong frame. AI maturity is measurable. The gap between a business that has deployed AI at any meaningful scale and one that has bolted a ChatGPT wrapper onto its website is enormous, and it has direct implications for value creation potential, integration risk and management bandwidth post-close.
This article gives operating partners a practical due diligence framework for assessing AI capability in target companies - what to look for, what questions to ask, and how to weight what you find. It covers data infrastructure, talent, deployment maturity and the governance questions that most deal teams skip entirely.
Why AI due diligence is different from tech due diligence
Standard tech due diligence looks at architecture, security posture, code quality, infrastructure costs and technical debt. Those things still matter. But AI capability does not sit cleanly inside any of them.
A business can have clean, modern infrastructure and still have no meaningful AI capability. Equally, a business can be running on legacy systems but have a data science team that is generating real commercial output. The infrastructure assessment tells you about risk. The AI assessment tells you about opportunity and, critically, about whether the management team's claims hold up.
The questions are also different. Tech due diligence asks: "Is this system stable?" AI due diligence asks: "Does this capability actually do what they say it does, and would it survive a change in headcount, a change in vendor or a change in strategic direction?"
That last question matters in PE specifically. You are not just buying a snapshot. You are buying a trajectory, and AI capability is one of the few areas where the distance between "we are building this" and "this is generating returns" can be measured in years.
The four domains you need to assess
Treat AI due diligence across four distinct domains. Each one has a different failure mode and requires a different type of evidence.
1. Data infrastructure
AI capability is downstream of data quality. Before you assess any model or system, you need to understand what data the business actually holds, whether it is clean enough to be useful and whether the architecture allows for it to be used at speed.
Ask the target to show you, concretely: where customer and transactional data is stored, how it is structured, what the latency is between event and availability, and whether they have a data warehouse or data lakehouse that a data science team can actually query without raising a service ticket with IT.
A consumer goods business with £600m revenue that stores all its customer data in siloed ERP modules and cannot produce a unified customer view without a manual data pull is not AI-ready, regardless of what the management deck says. That is a remediation programme, not a capability.
2. Deployed AI versus experimental AI
This distinction is the most important one to make, and it is the one that gets obscured most often in presentations.
Deployed AI means a model or system is in production, running against live data, influencing a business decision or process on a repeatable basis. Experimental AI means someone ran a proof of concept, it showed promising results and it has not moved since.
Ask directly: what AI systems are currently in production? What decisions do they influence? What would happen if you turned them off tomorrow? If the answer to the last question is "not much", that tells you everything.
A UK logistics business at the lower end of the mid-market might present a demand forecasting model as a core capability. The right follow-up is: how long has it been running, what was forecast accuracy before and after, and is the operations team actually using its outputs? If the operations director cannot answer that question, the model is not deployed - it is a slide.
3. Talent structure and dependency
AI capability in mid-market businesses is almost always concentrated in one or two individuals. That creates a specific kind of key-person risk that does not show up in standard HR due diligence.
Find out who built the systems, who maintains them and who in the business can extend them. If the answer to all three is the same person, and that person is already fielding calls from London recruiters, price that into your assessment.
More importantly, understand the structural position of data and AI talent within the organisation. Is the data science function embedded in commercial teams with a mandate to drive decisions, or is it a central function doing ad hoc analysis on request? The former scales. The latter does not.
4. Governance, vendor dependency and model risk
This is where most deal teams stop asking questions too early. Three specific areas deserve scrutiny.
First, vendor dependency. A business that has outsourced its AI capability to a single third-party platform - whether that is a specialist SaaS tool or a hyperscaler service - carries concentration risk that is easy to miss. What happens to the capability if that vendor changes its pricing model, gets acquired or deprecates the feature?
Second, data governance. Are there clear policies around what data can be used to train models, particularly where customer data is involved? With UK GDPR still very much in force and ICO enforcement activity increasing, this is not a theoretical question.
Third, model documentation. Can the target produce documentation for its AI systems - what the model does, what data it was trained on, how it performs across different segments? Absence of documentation is a strong signal that the capability is not institutionalised.
The questions to bring into management interviews
The management interview is where AI claims either survive scrutiny or collapse. These six questions will tell you most of what you need to know.
- What is the single AI system you are most commercially dependent on, and what did it cost to build?
- How do you measure the performance of your AI systems - what metrics, and how frequently are they reviewed?
- What AI initiatives have you killed in the last 18 months, and why?
- Who owns AI strategy - a named individual with a mandate, or a committee?
- What would your AI capability look like in 12 months if your two most senior data people left?
- What does your board know about AI risk, and when did they last discuss it formally?
Question three is particularly revealing. A management team that has never killed an AI project has either never tried anything ambitious enough to fail, or is not honest about what has not worked.
Translating findings into the investment case
AI due diligence should feed three outputs: a risk-adjusted view of existing capability, a remediation cost estimate and a value creation roadmap.
If the target has genuine, deployed AI capability that is generating measurable commercial returns, that is an asset and should be valued accordingly. If the capability is real but fragile - dependent on one person, undocumented, running on a vendor relationship with no contractual protection - quantify the remediation cost before you get to heads of terms.
If the capability is aspirational, that is not automatically a problem. Many of the best AI value creation plays in PE start with a business that has the data and the commercial need but has not yet had the resource or the expertise to execute. In that case, the question shifts from "what have they built?" to "what could we build, and how quickly?"
That second question is where an advisory partner with deployment experience, not just assessment experience, changes the quality of the answer.
The cost of skipping this work
A business with overstated AI capability is not just a disappointment post-close. It is a management distraction, a capital reallocation problem and, in some cases, a material misrepresentation risk. The firms that are getting this right are treating AI due diligence as a first-class workstream, not an appendix to the tech review.
The firms that are not will keep buying slides.
If you are preparing for a process where AI capability is a material part of the investment thesis, Rodan offers a structured tech and AI due diligence diagnostic. It is scoped, time-bounded and priced to fit within a deal process. Book a diagnostic conversation at rodan.io.
