An independent read on the AI work someone is selling you.

Before you sign

The Read.

Two scopes, depending on whether you have a specific proposal in hand or are still deciding what is worth doing at all.

• + A written memo, plus a call to walk through it
• + Priced and scoped before the work starts, so the fee is known
• + Sometimes the answer is do not sign this, and if it is, I am not bidding on the replacement

Scope note: the legal terms are your counsel's job, and the contract language is something a good AI model will already walk you through. What I assess is the technical design, the data assumptions underneath it, and whether the price matches the work those imply.

Ask for a read

After you sign

Owner's Technical Representative.

The review does not have to stop at the signature. On a funded pilot, the only people with the expertise to judge the work are often the same people being paid to deliver it.

Who is reading it

Someone who has built and priced these systems.

I am a Ph.D. physicist who spent eight years inside Oracle's AI Center of Excellence, taking AI and machine learning engagements from customer discovery through technical design to production for enterprise clients in manufacturing, oil and gas, finance, public sector, real estate, and retail. Before that, four years building machine learning systems on large data platforms. The reason I can tell you what a proposal should cost is that I have scoped, priced, and delivered the same kind of work.

See selected work and publications, or the full capability list.

Questions

What buyers usually ask first.

• How do I know whether an AI proposal is realistic?

  Most of the answer is in the data assumptions rather than the architecture. A proposal that describes an impressive system while staying vague about what data exists, where it lives, and what condition it is in is describing a hope. I read for the join between what is promised and what the data can actually support, then for whether the timeline and the price match the work that implies.

• Can't I just have an AI review the proposal?

  For the contract terms, yes, and you probably should. A good model will work carefully through data rights, retention, sub-processors, SLAs, liability caps, renewal mechanics, and lock-in, and that is real value at no cost. What it cannot do is tell you whether the technical approach will work on your data. It cannot see your data, talk to the people who maintain it, or price the engineering against what you actually have. That is the read I do, so the two are complementary rather than competing.

• What should an AI pilot cost?

  There is no benchmark price worth quoting, because scope varies more than any published range would suggest. What is useful is a cost line built from your specifics: your data volumes, the model and infrastructure the design actually requires, and the engineering time the work implies at current rates. That number is what you hold a vendor's quote against. Producing it is most of what the opportunity assessment does.

• What should I ask a vendor before signing?

  The questions that matter are usually about what happens when things do not go well: what specifically constitutes success, who decides whether it was met, what happens to the engagement if it is not, and who owns and operates the result afterward. A read gives you those questions in the specific terms of your proposal rather than as a generic checklist.

• Nobody in-house can judge this. What are our options?

  You can hire the expertise permanently, ask the vendor to grade its own homework, or bring in someone independent for the specific decision. The third is what this is. It exists because the first is slow and expensive for a single decision, and the second is not really an option.

• What is an owner's technical representative?

  The term comes from construction, where an owner hires their own engineer to represent their interests against the contractor's. Same idea here. I define success criteria and acceptance tests before work starts, review independently at each checkpoint, and give you a call on whether to continue. I do not staff the build: your team or the vendor implements, and I hold the standard.

• Who pays you, and do you take referral fees?

  You pay me, and nobody else does. No vendor relationships, no referral fees, no reseller margin, no partner status with anyone whose product might come up. That is the entire basis of the service, so it is not something I trade away for a particular engagement.

• What if you tell us not to sign?

  Then that is the deliverable, and it is often the most valuable one. And if that is the answer, I am not bidding on the replacement. A reviewer who might win the work has a reason to find fault, which would make the review worth less to you than it cost.

• How quickly can you turn a review around?

  A proposal or SOW review is five business days from when you send the documents. If you are up against a signing deadline, say so when you write and I will tell you honestly whether it fits.