Yes, we use AI. No, you can't get this from a prompt.
The same models you use are part of our infrastructure. What they produce on their own is a draft nobody checked. The difference between that draft and finished intelligence is two years of system building, days of machine time per engagement, and a person who signs their name to it.
What a prompt gives you
Ask a frontier model about your competitor and you get an answer in thirty seconds. It is fluent, confident, and shaped like intelligence. Here is what it is actually made of.
It answers from what it already believes, plus a quick search if it bothers. It does not know the date it looked. It cannot show you the page. It will give you a number it never saw, because a plausible number is what the sentence needed. It cannot tell the difference between a claim your competitor made about itself and a fact someone checked. And when it is wrong, there is no one to call.
In one diagnostic of a public company, the first-pass figures for all five competitors' profitability were wrong. Two had the sign flipped. Each one cited a real investor document. That draft looked finished. It would have gone to a chief executive. Our verification layer caught all five before a person ever saw them.
That is not a story about a bad model. The models are extraordinary. It is a story about what a model does when nothing is checking it.
What we add
A prompt is one step. An engagement is a system, and most of the system exists to stop the model from doing what models do.
A method decides what gets gathered. Not whatever the model happens to know, but the sources, frameworks and questions that fit the decision you are making. Public and subscribed data, your intake if you choose to share one, and the gaps between them.
Every claim gets a capture. The answer, the prompt that produced it, the date, the source it leaned on, and a hash so it cannot be quietly changed later. That is what the Evidence Locker is. A skeptical reader can click any sentence and land on the pixels.
Every number gets checked against a primary source. Not a summary of a summary. The filing, the page, the document. Where a claim cannot be checked, it is labelled as reasoning, not fact. Where the evidence is thin, the deliverable says so.
Eighty-six checks exist to stop us overclaiming. They look for false precision, superlatives with no number behind them, theatrical assertions, forecasts dressed as facts, and claims that only the company itself has ever made. Then other models are invited, in several rounds, to find anything we overstated, and the corrections are applied rather than argued.
Freshness has rules. A price older than seven days is captured again. A page we say exists is checked again after fourteen. A leadership claim expires after ninety. A prompt has no idea how old its answer is.
A person signs off. Someone who knows your name reads it before it ships, and is the one you call afterwards.
What it costs us
We tell you this because the price only makes sense if you know what is behind it.
Two years went into building the system before the first engagement shipped: the methods, the checks, the review rounds, the capture infrastructure. Each engagement then takes days of machine time, not seconds, because gathering and verifying is slow by design. And every one of them takes hours of human judgment, which is the part we will never automate away. That is why this is priced like work and not like a subscription, and why it is still a fraction of what a strategy firm would charge for less.
What we will not claim
We cannot see inside any model, and neither can anyone else. Answers vary by session, account, region and day, which is exactly why we capture what a buyer would plausibly have seen, on a date, with the prompt recorded. We do not promise outcomes. Anyone who promises you an answer engine will say something specific about you on a timetable is selling something nobody can deliver.
The short version
AI is our infrastructure, the way imaging is a surgeon's. Faster and more precise. The judgment is ours. If you want the draft, the prompt is free and you already have it. If you want the finished intelligence, with the proof behind every line, that is what we make.