We sell a one-week diagnostic sprint that ends in a written verdict: build this, or don't. It is the cheapest thing we offer and the one people ask the most questions about, because “a memo” is a vague thing to buy. This is what is in it.
1. The verdict, in the first paragraph
The memo opens with the answer, not the methodology. Build, don't build, or build something adjacent to what was asked for. Then the reasoning, then the evidence.
This ordering is deliberate. The memo has to survive being forwarded to someone who was not on any of the calls, read on a phone, between meetings. A document that makes a busy reader hunt for the conclusion will be skimmed into whatever they already believed before they opened it.
The verdict also carries its own falsification: what would have to be true for it to flip. A no-build that says “this becomes viable when your product catalogue has clean category tags” is far more useful than a no-build that just says no, because it names the work that changes the answer.
2. The architecture sketch
Not a diagram of boxes with arrows. The specific system we would build: which model, which retrieval strategy, where the data lives, what runs synchronously and what does not, and which parts we would deliberately keep boring.
Every choice carries its reason and its alternative. “A managed vector store rather than self-hosted, because your team is four people and none of them want to be on call for a database” is a sentence a client can argue with. “Best-in-class vector infrastructure” is not. We write the first kind, because the sketch is meant to be handed to your own engineers for attack.
3. The risk register
Three to five things most likely to sink the project, each with a mitigation and an honest probability. Not a compliance artifact: a ranked list of what we are actually worried about after a week inside the problem.
The register is where the data reality check lands, and it is usually the most uncomfortable page. In most engagements the top risk is not the model. It is that the data is not what everyone in the room believed it was: the export is missing fields the workflow needs, the historical records changed schema three years ago, the labels were entered by whoever had time. We have written “the primary risk is that this dataset cannot support the stated accuracy target” more than once, and each time it was the sentence the client most needed in writing.
4. The fixed quote
If the verdict is build, the memo ends with the scope and price of the sprint that follows, valid for thirty days. A number, not a range, because a range is a way of not committing while appearing to.
The quote is possible precisely because the week happened. Quoting a build off a discovery call means pricing uncertainty, and every firm handles that the same way: pad the number, or absorb the overrun later in a change request. A week of real contact with the data replaces the padding with knowledge.
The uncomfortable part
A no-build verdict costs us the build. That is the obvious objection to buying a diagnostic from the firm that would do the work, and it deserves a straight answer rather than a reassurance.
Two things hold the line. The memo is written to be defensible in front of your engineers, who have no reason to be generous to us, and a diagnostic practice only survives on verdicts that age well. A firm that talks clients into builds that fail gets one round of fees and no second engagement. We would rather write the no-build, keep the relationship, and be the people you call when the next idea is better founded.
Roughly a third of ideas should not be built as first framed. If that number were zero, the sprint would be a sales instrument with a memo attached.
Why a week, and not a call
Everything above could be guessed at in a ninety-minute conversation, and plenty of firms will. The difference is the two to three days of hands-on work against your hardest examples, enough code to know and not enough to polish. Opinions formed without touching the data are worth what they cost.
Related: the week-2 demo rule covers what happens once the verdict is build, and the diagnostic sprint page carries the scope, the fee, and what we need from you during the week.