What technical due diligence on an AI startup costs

The quote arrives and it is a range, which is the first problem. A technical read on the same company can be quoted at four thousand dollars and at forty thousand, and neither number is wrong. They are answers to different questions.

What moves the figure is almost never the seniority of the person reading the code. It is how much of the company is in scope, and how much of the answer has to survive a lawyer reading it afterwards.

What you are actually buying#

A read, at the low end. One experienced engineer, two or three days, the codebase and a conversation with the technical founder. The output is a memo. It tells you whether what you were told matches what exists, and it names what the reader could not see.

A review, in the middle. A week or two, several people, including security, infrastructure and sometimes a commercial read on the engineering plan. The output runs to twenty or forty pages and covers things nobody at seed stage will ask about.

A full diligence, at the top. Built for private equity buying a majority stake, where the report has to stand up in a dispute later. Crosslake, for instance, says it has evaluated more than 6,000 technology investments and serves more than 500 private equity firms. That is a different product from the one a seed fund needs, and it is priced accordingly.

Why a seed cheque usually wants the cheap one#

At seed and Series A the questions worth asking are small in number and answerable quickly. Does the thing work. Did these people build it. What will the first six months of engineering actually cost. Is there anything in here that will have to be rebuilt before Series B.

A two-week review answers those on day two and then spends twelve more days documenting things that will all be rebuilt anyway. For a company with eleven employees and a codebase eighteen months old, the extra spend buys thoroughness about a thing that is about to change completely.

The exception is when the cheque is large enough that being wrong is unrecoverable, or when something in the first read came back ambiguous. Ambiguity is the signal that buys the longer engagement, and it is a much better reason than a policy.

What a cheap memo tends to leave out#

Security, properly. Forty-eight hours is enough to notice that credentials are in the repository and not enough to audit an authentication model. Any memo claiming otherwise is overstating itself. If security is the actual concern, scope it separately and pay for it separately.

The data position. Whether the training data is licensed, what the vendor terms permit, and what happens when a model provider changes its pricing. This takes longer than reading code because it means reading contracts.

Anything about the market. A technical read says whether the thing was built well. It says nothing about whether anyone wants it, and a memo that strays into that is a memo written by somebody filling pages.

The question worth asking the firm#

Ask what the memo will say about its own gaps. A good one states on the first page what it did not look at and could not verify. A bad one is uniformly confident, which is how you end up reading twenty pages of assurance about a system nobody logged into.

Ask it before you commission, because the answer tells you whether you are buying judgement or page count.

What we do, and what it costs#

Our first memo for a fund is free, one per fund, with no obligation attached. Three to six pages, inside 48 hours, covering whether the codebase matches the story, where the team carries risk it has not priced, what the first six months of engineering will cost, and the questions worth asking the founder directly. The memo states its own gaps on the first page.

It is free because a fund cannot tell a good technical read from a confident one until it has seen one, and we would rather be judged on a memo than on a sales call. After the first, pricing is per engagement and depends on how many companies and how deep.

If you want to know what the read looks like before you spend anything, the method is written up in how to assess a startup codebase in two days. If you would rather just have the memo, book the call and bring the company you are looking at.