About a month ago I wrote about why so many products stop at tracking: a record cannot be wrong the way a recommendation can, and somebody has to be willing to own that in public (here).
Here is the part I left out. Building the deciding part is not the finish line. You can build it, ship it, and still watch people redo your answer by hand.
What I saw
Some time ago I spent a few hours inside a product that is not a tracker at all. It builds models. You describe how your business works, and it shows you the connections — boxes, arrows, one number feeding into the next. It is well made. It works.
I never got to the moment where it clicked. Part of that is me: I am not a finance specialist and not the target user. But I think something else was going on too, and I think it would happen to the target user as well.
I could see the structure and I could not see the formulas. The picture told me that A affects B. It did not show me how. So there was nothing for me to check.
Here is what I think a specialist does with that. They build the model in the tool. Then they rebuild it in a spreadsheet, to see whether the number is right. And then they do it again the next four or five times. Not because the tool was wrong — because nothing ever gave them a reason to stop checking.
That is the part that bothers me most, because it works against the whole point of the product. The promise is that the user stops working in spreadsheets. If every answer sends them back to one to verify it, the product has not replaced the spreadsheet. It has added a step in front of it.
Your real competitor is not another app
If you build a financial planning tool, your real competitor is not another app. It is the spreadsheet. And I think people trust a spreadsheet for one reason: they can see every step. They know where the number came from, because they built the path themselves.
That is a high bar for anything that wants to decide for them. A spreadsheet never tells you that you are wrong, so it never has to earn that right. A tool that recommends does. I think it has to keep its reasoning visible, and the reasoning has to still make sense months later.
And the cost is bigger than one user. The whole point of a model is to show it to somebody else — a partner, a board, an investor. If the person who built it cannot explain where the number came from, the person being shown it will not believe it either. The doubt does not stop with one user. It travels.
What I would try
I have been thinking about this as a product person, not a finance one 🙊. If the problem is that there is nothing to check, the job is to give people something to check — before they go looking for it somewhere else.
I worked these through with AI as a thinking partner and kept the four I could argue for. Where I am not convinced, I have said so.
1. Show the arithmetic where the number is. Not the formula in a help page — the actual sum, next to the result: 1,200 new customers × $40 × 6 months = $288,000. A formula tells you the shape. The numbers tell you it is right. A spreadsheet gives you both at once, and that is most of why people trust it.
The objection to my own idea: it makes the screen noisy. Somebody building a model does not want arithmetic under every box. So it probably has to be something you switch on, and switched off by default.
Which leads somewhere else. Maybe the answer is not one view with a toggle, but two views for two people. The builder is assembling. The audience is interrogating — why this number, what happens if you are wrong, what did you assume. Most tools design only for the builder, and then ask them to present with the same screen.
2. Let people walk backwards. Click any number and see what it was made from, one step at a time, until you reach something a human typed. “Where did this come from” should be answered inside the product, not from somebody’s memory three months later.
This is also a place to beat a spreadsheet rather than copy it — tracing a number through forty tabs is horrible, and people accept it only because there is no alternative.
3. Keep the reason next to the assumption. Every model has numbers somebody chose. 3% churn. $40 average price. Eighteen months of runway. The number gets stored; the reason almost never does.
“3% because that is our last six months” and “3% because an investor asked us to model it” are different claims, and by the board meeting nobody remembers which one it was. A product that keeps the reason attached to the number holds something a spreadsheet keeps losing. I think that is where the defensible part is. The calculation is not — AI can produce a one-off model in an afternoon.
4. The uncomfortable one: let them export it to a spreadsheet with the formulas live. This sounds like surrender. I am not sure it is. Rebuilding the model by hand costs the user an evening, and quietly teaches them they can work without the tool. A button that does the same in ten seconds ends that habit. You lose a feeling of lock-in, and you keep the thing that actually matters — being the place the model lives.
This is the one I am least sure about, and I would like to see it argued.
What I am not sure about
All four are guesses. I have not tested any of them, I am not a finance specialist, and I am not the target user — when I looked at that model I checked the maths with AI rather than from my own knowledge, which tells you where my limit is.
I made a much smaller version of the same call in my own training app — the part that picks exercises and weights runs on rules I wrote myself rather than a model, because I wanted to be able to say exactly why it suggested what it suggested. That deserves its own post, and I will write it once the app is out.
What I keep thinking about
Under all four options there is one question — will the person believe it — and it is better answered early, while the product is still being built.
So probably something like: will they trust the answer the first time they see it? If not, what would have to be visible before they would? And if they are going to check it anyway — and I think most people will, at least at the start — where do they go to check, and could that happen inside the product instead of somewhere else?
I do not think checking is a bad sign by itself. Someone who checks is someone still deciding whether to rely on you. The version that costs you is when there is nowhere to check except a spreadsheet, because then the habit forms outside your product, and habits that form outside do not come back on their own.
That is what impressed me most in those few hours. The tool worked, and I still did not believe it, and it took me a while to see that those two things were unrelated.
Leave a Reply