Investing in an AI product company, on evidence
Most AI diligence rests on a demo, a deck and a founder reference. None of those survive contact with an enterprise security review. huby evaluates the product itself - the thing the company is actually selling - against a published framework, and publishes the score and the evidence behind it. This page explains what that gives an investor and where it stops.
The problem with AI diligence
An AI product company is unusually hard to diligence from the outside. The demo is the artefact most investors see, and a demo is the single easiest thing in the category to make look finished. The gap between a product that demos well and a product that survives an enterprise security review is where a large share of AI write-downs live.
The conventional answers each have a hole in them. Founder references are selected. Customer calls reach the customers who stayed. Technical diligence is expensive and is usually spent re-deriving facts that could have been established before the meeting. None of it is comparable across two companies in the same cohort, which is the thing a portfolio decision actually needs.
What an independent evaluation adds
huby evaluates products, not companies, against a published framework of six categories broken into subcategories and then into atomic factors - specific, verifiable data points. Every factor score carries its evidence and a source tier, from primary testing we ran down to anecdotal community reports, and the tier is visible rather than averaged away.
Two properties make that useful in diligence. It is comparable: the same tests and the same framework are applied across every product in a cohort, so a 4.1 in one report means what a 4.1 means in another. And it is independent: we are not paid by the company being evaluated for the outcome, and the methodology is published in full so you can disagree with it specifically rather than generally.
Reading the six categories as investment signals
| Category | The question it answers | Why it moves a decision |
|---|---|---|
| Quality | Does the product do what the deck says it does? | Identical tests run across every product in a cohort. A score well below the cohort median on a capability the company sells as its edge is the gap between the pitch and the product. |
| Security | Can this company close an enterprise deal? | Weak access control and governance evidence is the most common reason a promising AI product stalls in procurement. It caps realistic deal size long before it shows up in revenue. |
| Privacy | Is the data position defensible? | How inference data is governed determines which markets and customer segments are reachable at all, and how exposed the company is to a regulatory change. |
| Sustainability & Reliability | Will it still be running in three years? | Operational reliability and business sustainability are scored separately. A product with strong uptime and a thin commercial base is a different risk from the reverse. |
| Use Cases & Pricing | Is the pricing model survivable? | Unit economics under inference cost, and the coverage of real enterprise workflows, say more about durable margin than a headline ARR number does. |
| Impact, Ethics & Safety | What is the tail risk? | The category that rarely moves a term sheet and often moves a headline. It is where a position that is cheap to fix pre-term-sheet becomes expensive post-close. |
The scoring scale, the source tiers and the factor weighting are set out on the methodology page.
Where it fits in a deal process
- 1
Before the first partner meeting
Pull the published score for the product and its cohort. A product that is unevaluated is not a red flag; a product that scores well below peers on the dimension the pitch leans on is a question to put to the founder in person.
- 2
During technical diligence
Use the evaluation as the baseline and spend your own technical hours on what it does not cover: the roadmap, the team, the code you are actually buying. The checklist on our due diligence guide is built for exactly this split.
- 3
In the investment memo
Cite the factor-level evidence, not the headline number. Every score traces to a source with a tier attached, so a memo can say why a risk is rated the way it is rather than asserting it.
- 4
Post-close, as a monitoring signal
Scores are re-run as the framework and the product change. Movement on Security or Reliability between evaluations is an early indicator that is independent of anything the company reports to you.
What this is not
- It is not investment advice. We evaluate products. We do not value companies, model returns, or take a position on whether a round is priced correctly.
- It is not a substitute for technical diligence. An evaluation establishes a comparable baseline so your own technical hours go to the things only you can assess.
- It does not see inside the company. Burn, pipeline, churn, the cap table and the team are outside the scope of a product evaluation entirely.
- A score is a point in time. Every report carries the date, product version and access tier it was run against, because all three change.
Start with the checklist
The due diligence checklist is the investor-facing companion to this page: what to verify, in what order, and what counts as evidence. If you are weighing a specific product, you can ask us to evaluate it.