AI Due Diligence

AI Due Diligence for Investors and Acquirers

Engineers who build with AI every day tell you whether a target’s AI is a genuine asset, a thin wrapper or a hidden liability, before you commit.

Discuss your AI due diligence

JFDI provides AI due diligence for private equity and venture capital investors, acquirers and boards. Our engineers examine the artificial intelligence behind a target business and tell you, in plain English, what it really is: a defensible asset, a sensible use of someone else’s model, or a demo that won’t hold up once real customers use it.

Plenty of businesses now call themselves AI-powered. Some have built real technology. Others have put a user interface and a confident pitch deck around someone else’s model. In a demo the two can look the same, but they carry very different risks and are worth very different amounts. AI due diligence is the part of our technical due diligence work that tells them apart.

Who AI Due Diligence Is For

Private Equity

Evidence on the AI before the investment case relies on it.

Venture Capital

What the AI really does, before a funding round puts a price on it.

Acquirers

What the AI is built on, and what it will cost to run after the deal.

Boards

An independent check on your own AI claims before investors test them.

What AI Due Diligence Covers

We start from the code, the data and the running system, not the pitch deck.

What we look at

Ideally we see the source code, prompts and model configuration, the data pipelines and records of where training data came from, evaluation results and production monitoring, and the contracts and bills for any third-party models. We use the system ourselves and talk to the engineers who built it.

Early in a deal, access is often limited. We then work from documents, demonstrations and interviews, and the report says which findings we checked and which rest on what the target told us.

Are the AI claims real?

We trace what happens when a user triggers an “AI” feature: which code runs, which model is called, what data goes in and what comes back. That tells you whether the target owns meaningful technology, has built genuine value around a third-party model, or has very little beyond the prompt. The first two can both be sound businesses, but they are not worth the same.

Where the models and data come from

We trace where training and fine-tuning data came from, whether it includes customer or personal data, and which terms and licences the business relies on. Where it uses open-weight models, we identify them and their licences. Whether those terms allow what the business does with them is a question for the deal’s lawyers; we give them the facts.

Dependence on model providers

Many AI products are built on models from a small number of providers. That is often the right engineering decision, but it creates a dependency you are buying into. We look at how much of the product would stop working if a provider changed its pricing, its terms or its models, what the AI costs per customer today and how that scales, and how much work it would take to move to another provider.

Evaluation and failure modes

AI systems fail differently from conventional software: they produce confident, plausible answers that are wrong. We check whether the target measures quality properly, with test sets, automated evaluations and monitoring in production, and whether it would notice if a model update made things worse. If a team can’t tell you how often its AI is wrong, it doesn’t know.

AI-generated code quality

Many targets now write much of their software with AI coding tools. So do we, which is how we know where those tools cut corners. We assess whether the engineering around them holds up and whether the team you would inherit can maintain the result. For a single application, our code health check goes into this in more depth.

AI security

AI features open attack routes that traditional security reviews miss. Prompt injection can turn a helpful assistant into a way to extract data or trigger actions it should never take. Shared context can leak one customer’s information to another. AI agents wired into email, files or business systems can do real damage with the permissions they have been given. We review how the target handles each of these, in the code and in practice.

Regulatory exposure

What the EU AI Act requires depends on what a system is used for and whether the business builds it or only uses it, and the Act can apply to businesses outside the EU whose AI is used there. In the UK, existing law applies, including UK GDPR rules on automated decisions. We are engineers, not lawyers, so we don’t give a legal view. We document what each AI system does, who it affects and what logging and human oversight exist, so the deal’s lawyers can.

What You Receive

  • A plain-English account of what the target’s AI is, what it is built on and what it depends on
  • Graded findings across claims, provenance, dependence, evaluation, code quality and security, each marked verified or reported
  • Strengths as well as weaknesses, so the findings can be weighed against the investment case
  • Questions for your legal advisers on data rights and regulation, with the technical evidence behind each
  • Recommendations for remediation, and questions to put to the target before completion

Why JFDI for AI Due Diligence

We build AI systems ourselves, for our own engineering and for clients through our agentic AI development work. We know how they are put together, where they break and what they cost to run, which is what lets us tell real capability from a convincing demo.

Where a target runs on Microsoft, our Microsoft 365 and Power Platform due diligence covers the estate around the AI. For smaller companies, or founders preparing for investment, our code health check is a faster, fixed-scope review of an AI-built application.

AI Due Diligence FAQs

Is the AI as good as the pitch deck says?

Tell us about the target and your timeline, and we’ll scope an AI due diligence review around it.