Take the question
Your plan or move, or your product, exactly as you put it. No forms, no taxonomy to learn — a sentence is enough.
Read it back
We restate what we understood and name anything you left open, so a gap is visible rather than quietly filled in.
For a plan, every part is sorted by how new it is: what already exists, what's an upgrade to something you offer, what's new to the market, and what's new to the world. That sorting decides how the rest of the run is evidenced — a product that exists can be measured, and one that doesn't has to be reasoned about from things that do.
Establish the ground truth
Before we ask anyone anything, we establish what's actually true — from documents, not from a model's memory. Every fact keeps its source link and how confident we are in it.
For Liquid GTM, it's the live market around the move — who else is doing it, what they charge, how it landed — established by web search and kept with its sources. Any figure that can't be traced back to one of them is flagged in the briefing.
For an audit, it's your own pages: the real rate, the fee, the terms. The benchmark every answer is graded against is a document we can show you.
Ask the assistants
We put the questions your customers actually ask — phrased the way a person types them, not the way an analyst writes them — to the live, web-connected assistants. Several times each. Every answer is kept.
A Liquid GTM run asks about the demand your move depends on — does anyone want this, what would they pay, who already does it well. An audit asks about your product against a versioned set of intents: the rate, the catches, is it worth it, how to open one, how it compares, what you'd recommend.
Every answer we work from came from a real assistant answering a real question.
Weigh every projectionLiquid GTM
There is no verified truth to grade a projection against — the thing hasn't launched. So instead of a score, every projected response carries what it rests on: a live demand signal, a web-verified comparable, or reasoning.
And it carries a falsifier — the specific thing that would prove it wrong. A read that cannot be wrong isn't a read.
Grade every answerAudit
A separate model, running with no creative latitude, marks each answer against the verified truth: right, partly right, different, or it never gave a figure at all.
That last one matters more than it sounds. An assistant that simply omits your rate when a customer asks isn't wrong — it's absent, which is often the more expensive outcome.
Cap the confidence to the evidenceLiquid GTM
Confidence on a projection isn't the model's opinion of itself. It's a ceiling computed from how much evidence the run actually gathered — applied afterwards, in code, and it can only ever lower a number.
Thin evidence, low ceiling. No exceptions, and nothing reaches certainty, because nothing about an unlaunched product is certain.
Score it — with the uncertainty attachedAudit
Accuracy is never a bare number. Every score carries a 95% confidence interval and the number of answers behind it.
Ask three questions and the interval is wide, and we show it wide. Ask more and it tightens. That's the only honest way to buy confidence, and it means you always know how much weight a number will hold.
Put it in context
A number alone doesn't tell you what to do. Both engines establish where you stand against a real competitor set, sourced from the web rather than assumed.
An audit adds two things a score can't give you. Whether the assistants recommend you when a customer asks the category question without naming you — the question that decides whether they ever reach you at all. And how it compares to your last run, so you can see what moved, with a change only called a change when the intervals don't overlap.
Report it
Written for whoever needs to read it — board, executive, or analyst. Same evidence underneath, three altitudes on top. Every figure traces back to something you can open.
Seal the recordAudit
Everything behind the run — the questions, the models, the answers, the grading, the arithmetic — is fingerprinted and written to our ledger as it happens, before you see the result.
Anyone you hand the audit to can check it themselves — no account, no login, no need to take our word for anything.
An audit you can't check is an opinion with a logo on it. Every measured number traces back to the answer it came from, and any audit can be verified by someone who has never met us.
The math driving the results
The sampling frame, the confidence intervals and why they're Wilson intervals, how the grader is calibrated against human judgement, and the ceilings that stop a number claiming more than its evidence supports.
Read the math →