AI Diligence
Bringing rigor to a frothy market
We go deep when the value of the deal rests on the target’s AI or its AI talent. The job is to get past the hype to tell you whether the AI is a real and durable strategic advantage or a thin layer applied for optics.
Where does the target fit
Three kinds of AI company
Each pattern requires different, specific analysis.
Built an application on existing models
The most common pattern. Value lives in the product surface, the proprietary data loop and the workflow, so we test how much survives a model swap and what breaks when the underlying provider changes terms or prices.
Builds tooling for AI
Infrastructure, evaluation, orchestration, safety. Here the moat question is velocity: the field is moving fast in exactly this layer, and we assess whether the target is ahead of it, keeping pace, or already commoditized.
The three questions
Provenance, durability, viability
Provenance
Is this an AI play, or a company that merely uses AI? We trace where the models, the data and the differentiating work came from: built in-house, fine-tuned, or bought from a vendor with an API key.
Durability
Is the capability both differentiated and built to emerging norms for quality control? It may have been differentiated once. We ask how well it holds up as the field keeps moving.
Viability
The business lens: so what? We assess the capability against an AI backdrop that moves every quarter, and against what the acquirer actually intends to do with it.
What we assess
Six angles on the same claim
Defensibility and moat
Take advantage of one of the few firms that has seen under the hood of a great many AI startups.
Team and talent
We identify key-person dependencies through interviews, source-code trails and engineering controls.
Risk
Security flaws, regulatory compliance and bias, assessed against the target’s readiness to expand.
Data and IP ownership
Do they own the training data? What rights do they hold to customer data, and what survives when a customer leaves? These come up in nearly every AI deal, and we answer them from the contracts and the code, not the pitch.
How do you know what you know?
Accuracy tracking, drift resistance, bias, overfit. The most common failure we find: they happen to be right, but they can’t prove it. Quality control is what separates a capability from a demo.
We diligence AI technologies for some of the largest companies in the world, including some of the largest AI companies.
And the question behind every AI deal now: could frontier models erase this company’s reason to exist? We answer it through a computer-science lens, asking whether this is at heart a data problem, a vision problem or a workflow problem, rather than through a market survey. We draw that line on purpose.
Questions we get
Frequently asked
Is it real AI, or a wrapper on someone else’s model?
Does the moat survive better base models?
Can we train on the data we’re acquiring?
Is AI-written code a risk?
Ready when you are