For consultants, media buyers and fractional CFOs who advise clients on what to spend.
Knowledge is having the numbers.
Cnario turns them into decisions. Twenty to thirty minutes with your client's numbers, and every figure shows the assumptions and the confidence range behind it. Whatever they sell.
Pay once. Yours on 15 September. Refundable to the day it ships.
Check our arithmetic before you read our argument.
The full audit is published. Every figure opens to the formula behind it, so check the arithmetic rather than trusting it. No email, no login.
Six in total: a two-location clinic, a course launch, an online store, a subscription, an evergreen funnel, a solo consultancy.
If you find a figure that doesn't reconcile, tell us. We'll fix it in public.
The questions that arrive at the wrong moment
Every one of these is answered inside the sample, with the formula sitting next to the answer.
“At what CAC does this stop working?”
There's a point where unit contribution hits zero. The audit prints that number.
“If we triple the budget, does the margin survive it?”
The marginal click costs more than the average one did. The stress test re-runs the economics at 2×, 3×, 5× and 10× spend and shows you where it breaks.
“Is 3:1 healthy, or does it just look healthy?”
A strong ratio can sit on top of a payback period that quietly eats your client's cash. Those two fail in different ways, so the audit reports them apart.
“Which lever should they pull first?”
The audit sizes each one. On the store in the sample, getting from 1.9:1 to 3:1 takes either a 37% cut in CAC, from €41 to €26, or 0.8 more orders per customer, from 1.4 to 2.2. The first is a conversation with the media buyer. The second is a retention programme and a quarter of work.
“Can this client afford what I'm about to recommend?”
Break-even revenue, minimum clients, minimum billable hours, and how many months of growth it takes to get there.
“How do I explain this to someone who doesn't read a P&L?”
Plain-English diagnostics beside every metric. The document is written for your client to read on their own.
What you're using now, and what it measures
Both tools everyone reaches for have been tested on this exact task. Somebody published the results.
The spreadsheet
Auditors went through a pile of real business spreadsheets. 94% had at least one error in them.
Per cell the rate is tiny, which is exactly the problem. A file can be right nearly everywhere and still hand you one wrong number at the bottom, with nothing on screen looking off.
The same review also measured how confident the builders were. Everyone thought their file was fine.
The chat model
You already check its work. Twenty minutes verifying the formulas, every time, and at the end you're holding a chat log you can't send to anyone.
Two benchmarks this year put numbers on it.
One ran ten model setups from OpenAI, Google and Anthropic over financial spreadsheets. The best of them scored 82.4%, roughly one wrong answer in six, and on the largest file the average across every model dropped to 48.6%. Its conclusion was that no model on its own is accurate enough to be used unsupervised on professional finance work.
The other built 131 finance spreadsheet tasks graded against expert criteria. The frontier models came in under 50%, worst of all at getting the formulas right.
One error in six
You can't tell which one it was, so all six get checked. That's the twenty minutes of verification you're already doing on every answer it gives you.
Arithmetic that runs the same way every time removes the checking step. That's where the researchers landed too: they concluded reliable results need the reading and the calculating handled by separate things, with the maths fixed instead of written fresh each time. That's how ours is built.
What makes an answer you can stand behind
Five months in, the numbers have moved and someone asks where the March figure came from. Four things have to be true for that to go well.
Run it again in six months and the number hasn't moved.
Nothing in it samples, estimates or drifts. 393automated tests sit behind the arithmetic, and every figure in the document opens to the formula that made it. When your client's accountant checks your work, they land where you landed.
You can run it on the discovery call, from what they told you out loud.
There's nothing to connect. You don't need access to their ad account, you don't need their accountant to export anything, and you don't have to wait on someone's IT team to approve a new tool. Six numbers the client can tell you across a table are enough to get a first answer.
That matters most at the beginning of an engagement, which is also when the decision is usually made. The data is scattered, or the question is about a channel they haven't run yet, so there is nothing to integrate with even if we wanted to.
It tells you how much of it to believe.
Thin inputs produce a wide range, and the document says so on its face. Every headline figure carries a band around it, and the band widens when the inputs are guesses. A model that reports one decimal place on an assumption you invented is lying politely.
We can go out of business and your copy keeps working.
The licence check fails open. If our server never answers again, the software carries on. No phone-home, no expiry, no read-only mode. The audits are self-contained files that need no login and no connection to us.
Finmark's users were told their data would become permanently inaccessible. That's the normal outcome in this category, and it's the reason this is built the way it is.
Where each one actually wins
Two of these rows go against us.
| Sheets | A chat model | Cnario | |
|---|---|---|---|
| Modelling something we don't cover | Builds anything | Will try anything | Can't do it |
| Drafting the email that goes with it | — | Better than us | Doesn't do it |
| Time to a first answer | Depends what you've already built | Fast, then twenty minutes checking it | 20–30 minutes |
| Same answer twice | If nothing broke | Not guaranteed | Always |
| Every figure traceable to a formula | If you built it that way | No | Always |
| Handing it to a client as-is | Rebuild it first | Not really | It is the output |
| Reusing it on the next client | Copy, then clean | Starts from zero | Save, rename, fork |
| Still working in ten years | If the file survives | — | Yes |
Nothing to set up. Nothing to learn.
Six numbers off their P&L: ad spend, leads, sales, average order value, margin, fixed costs. No import, no connection, no onboarding call.
If it takes you longer than half an hour to get a verdict on your first client, something has gone wrong and we want to hear about it.
There's no formula language to learn. There isn't one in the product.
Then it stops being a fresh start every time
Save the scenario under the client's name. Fork it when they come back with a different question. Build it once for the first client; the twelfth one starts from the eleventh.
Your logo on the output. That's included in the licence.
€149, once
The licence is a one-time purchase. Twelve months of updates come with it. After that it keeps working at the version you hold, permanently, whether or not you ever pay us again.
It ships 15 September. Refundable to the day it ships, and for fourteen days after that.
€149 until 15 September. €199 after. That date doesn't move.
Pre-orders aren't open yet — check back before 15 September.
Running one decision for yourself rather than for a client? €29 buys seven days of everything, counted from the first time you sign in rather than from the moment you paid.
Cloud is €69 a year, or €49 if you add it at checkout. It keeps updates coming after the first twelve months and hosts share links you can revoke. Optional. The licence works without it.
There's no free version, and here's why
A working model finishes the job. Let you run unlimited scenarios and read the verdict, and you'd already have what you came for. A “trial” would be the product with a coupon attached.
So all six finished audits are published instead. Open them, download them, take them apart. Someone else's business proves the depth and the rigour while giving away nothing, because the entire value is your client's numbers sitting in it.
What You Might Be Thinking
I can build this in Sheets.
You have. Eleven times. The real question is what the rebuild costs per engagement, and whether the eleventh one gets the same care as the first.
ChatGPT does this for free.
For drafting, it beats us, and we're not competing there. For arithmetic over structured financial data, this year's benchmarks put the best models between 48% and 82% on exactly this task. Links are further up. Use both.
Is this a wrapper?
There's no model inside it. Arithmetic, 393 automated tests, and formulas you can read in the output. Nothing you type is sent to an AI.
How long before I get something out of it?
Twenty to thirty minutes on the first client, from cold, with six numbers in front of you. There's no import step to get wrong.
Do I have to learn it?
No formula language, no modelling syntax. If you can fill in a form, you can run it. The tabs are named after the questions rather than the maths.
What happens if you disappear?
It's a file. It sits in your Drive, your Dropbox, or wherever else you keep client work, and it opens in a browser with nothing to log into and nothing to connect to. The licence check fails open, so the software carries on running too.
What if it isn't for me?
Refundable to the day it ships and for fourteen days after. No form, no questions. The terms are on the refunds page, in the same English as this page.
Refunds pageCnario is new.
There are no testimonials on this page because we don't have any yet. The six audits are the only evidence there is, and you can take them apart before you spend anything.
Ships 15 September. €149 until then, €199 after. Refundable to the day it ships.