Methodology
Evidence, translated
Two rules govern the work. Every claim is traced to evidence or marked unsupported. Every finding is written through to its investment implication.
Engagement
How an engagement runs
Five steps, scoped to the decision in front of you. Material findings are raised as they emerge, not held for the read-out.
- 01
Scope to the thesis
We start from what you are underwriting, and set the questions the work has to answer.
- 02
Evidence request
A specific list: eval sets, traces, provider invoices, telemetry, repository access.
- 03
Testing & interviews
Claims tested against the evidence, with engineering and product interviews alongside.
- 04
Synthesis
Findings ranked by materiality, each written through to its investment implication.
- 05
Read-out
A working session with the deal team, plus the written report and open questions.
Frame
The eight questions
Does it work?
Measured performance against the task the buyer is actually paying for.
Is it real?
Claims traced to evidence: evals, logs, code, not demo footage.
Is it proprietary?
What is owned versus rented from a foundation model provider.
Is it scalable?
Behaviour under load, latency budgets, failure modes.
Is it economical?
Inference economics and gross margin at realistic usage.
Is it valuable?
Whether AI features drive adoption, retention and price.
Is it durable?
Data moat, switching costs, model-provider commoditisation.
Can the org execute?
Team depth, evaluation discipline, AI governance maturity.
Writing standard
Technical observation → business consequence → investment implication
A bare technical fact is not a finding. It becomes one only when its commercial and investment consequences are stated.
Not a finding
“The product calls a third-party foundation model over an API.”
True, and useless. It tells an investment committee nothing about price, risk or structure.
A finding
- Technical observation
- The core capability is produced by a single third-party foundation model, called directly, with no evaluated alternative.
- Business consequence
- Output quality, latency and unit cost are set by a supplier the company does not control, and a provider price change flows straight to gross margin.
- Investment implication
- Supplier concentration and pricing risk are material to the margin case. Underwrite a margin band, and make provider abstraction and a second evaluated model conditions of the value creation plan.
Standards
What we will and will not do
We will
- State confidence on every finding, and say plainly when evidence does not exist.
- Rank findings by materiality to your thesis rather than by technical interest.
- Raise material findings during the work, not at the read-out.
- Separate evidence, assumptions and unresolved questions in every deliverable.
We will not
- Name clients, disclose transaction details, or publish testimonials from engagements.
- Claim certifications, volumes or track record we do not have.
- Promise that every material risk will be found, or predict an investment outcome.
- Convert an absence of evidence into a favourable inference.
Scope
Three ways to engage
A rapid pre-LOI red-flag review, full AI due diligence, or deep-dive transaction support.
FAQ
Common questions
Considering an AI investment?
Bring the thesis, the data room and the timeline. We will tell you what evidence exists, what is missing, and what it means for the deal.