I built a regret check for the software deal you are about to sign

Somewhere in your inbox is a renewal letter with a number on it that made you say something out loud. That letter is how most software shopping actually starts. Not calmly, not with a spreadsheet, but angry, with a vendor demo booked for Thursday and a salesperson who has run this exact negotiation a thousand times sitting across from someone who has run it twice. Roughly 60 percent of businesses made a software purchase they regret in the past 18 months, by Capterra's count, and the regret is severe enough that the survey defines it as seeking major remedial action, like ripping the system out. That is not a statistic about bad products. It is a statistic about good products signed on the vendor's terms.

So yesterday's build is a checker. The Software Regret Check asks you ten multiple-choice questions about the deal you are about to sign, and it hands back three things: a verdict in words, a score you can see into, and a specific thing to negotiate for every answer that added risk. It runs entirely in your browser. There is no server behind it, nothing you pick is sent anywhere, and nobody sees your answers, including me. It takes about two minutes, which is roughly the length of the hold music on the call where you try to cancel the last thing you signed.

What the ten questions actually are

The questions are not clever. That is on purpose. Every one of them is a place where a real business, in a public forum, described losing real money.

How long does the contract run, because the loudest regret threads in our research corpus are people trying to escape multi-year agreements they signed in year one enthusiasm. How is it priced, because per-seat pricing quietly turns every new hire into a software tax, and operators resent it in exactly those words. Is the renewal price capped in writing, because the renewal letter that starts the whole cycle exists precisely when it is not. Can you export your data yourself, in a format you could actually use, because the answer to that question is the difference between a subscription and a hostage situation. Who sets it up, because implementation is where the money goes to die. One nonprofit in our research described being two years into a Salesforce rollout, still not set up properly, three thousand dollars sunk. Capterra's data says the same thing at scale: businesses that regret a purchase disproportionately blame the implementation, with 31 percent citing vendors who missed the deadlines they promised.

Then the softer ones, which are the ones that actually predict adoption failure. Did you try it on your own work, or did you watch a demo of someone else's. How much of what you are paying for will you actually use, because paying enterprise prices for a 60 percent fit is the single most repeated complaint shape in the corpus. Who wanted this tool, because a purchase that management picked or a vendor pitched has a very different survival rate than one the daily users asked for. We wrote about that failure mode in July: forty seats bought, three people logging in. What does cancelling look like, because auto-renewal into a new term is a trap with a calendar attached. And does it replace anything, because a tool that does not retire another tool is not a solution, it is an addition, and additions are how a 20-person company ends up with 40 subscriptions.

Each answer is worth zero, one, two, or three points. Twenty-five points are possible. Zero to five reads Sign it. Six to twelve reads Negotiate before you sign. Thirteen or more reads Walk away, or rewrite the deal.

The design decision that mattered: you can see into the score

Every scoring tool I have ever used online has the same tell. It asks its questions, disappears into itself, and comes back with a number and a color, and you have no idea why. That opacity is not an accident. Most of those quizzes are lead magnets for whoever built them, and the score exists to funnel you toward a conversation, so decomposing it would be giving away the meeting.

This one decomposes completely, because the decomposition is the product. Under the verdict sits a row for every question. Each row says Clear, Caution, or Flag, in words, shows exactly how many points that answer contributed, and if it contributed anything, tells you what to do about it. Multi-year contract? Ask for a twelve-month term at the multi-year price, or an exit clause at month twelve. No renewal cap? Ask for one in the order form, five percent or CPI, and know that vendors sign these all the time for anyone who asks. No self-serve data export? Ask for a full export in a documented format, demonstrated, before you sign, not described, demonstrated. The score is not the answer. The list of asks is the answer. The score just tells you how long the list is.

The tool also shows its own calibration, which most instruments online refuse to do. The zone boundaries are printed on the scale, and if your total lands within a point of one, the verdict says so out loud, in words: you are near the line. A deal that scores six is not meaningfully worse than a deal that scores five, and a tool that pretends otherwise is lying with typography.

There was one more state to design, and it is the one most builders skip: the half-answered state. Most people will not finish all ten questions before looking at the reading, so the reading has to be honest about being partial. Answer three questions and the verdict is labeled provisional, with the range it could still land in spelled out. Three answers in, the tool might say the deal reads fine so far but could still land anywhere from Negotiate to Walk away. A provisional number that looks final is how people make decisions on a third of the evidence.

What did not go smoothly

Two things, and I am writing them down because this series is a build log, not a brochure.

First, the ask lines forced a judgment call the design brief had not made. Some answers add a single point of caution without there being any sensible negotiation attached. An annual contract is mildly less flexible than monthly, but "ask them to make it monthly" is not advice, it is noise. So three of the possible answers add points without offering an ask, and I do not love it. A row that says Caution and then goes silent is a small dead end, and the next build in a flagging shape has a standing instruction in the project ledger: no flag without a next step, or no points at all.

Second, and more honestly: the weights are mine. Multi-year is three points and annual is one because I read several hundred operator complaints and that is my judgment of the relative burn risk, not because anyone measured it. The tool shows you where the scoring lines sit, but it does not tell you where they came from, and a tool that presents editorial judgment with the visual confidence of an instrument is doing something I want to be caught doing. So consider yourself told here: the point table is an opinion with a user interface. I think it is a good opinion, sourced from a lot of public pain. It is still an opinion.

What it cannot do

It scores the terms of the deal. It cannot score the product. A tool with a monthly contract, a written renewal cap, clean data export, and a real trial can still be bad software that your team hates by March, and the Regret Check will have waved it through, because everything it can see was genuinely fine. The inverse matters too: the best product in your category might sit behind a rigid multi-year order form, and the right answer might still be to sign it with your eyes open. The verdict is about risk, not fate.

It also does not know the size of the commitment. A thirty-dollar-a-month tool and a thirty-thousand-a-year platform get scored on the same 25-point scale, and they obviously should not feel the same. Adding a deal-size question would fix that at the cost of making a deliberately numberless tool ask for a number, and I chose numberless for launch. That trade might get revisited.

And it cannot read your contract. It asks what you believe the terms are. If you do not know whether the renewal price is capped, the tool scores that as risk, on the theory that not knowing is the risk, but it cannot go find out for you. The finding out is your job, and the whole point of the checklist is that you now know which clauses are worth an hour of finding out before the signature and not after.

Why a checker, and why free

This is the third build in the daily series, and the shelf logic is simple. The first was a calculator that could tell you to stay with the tool you have. The second was a ranker that could tell you a task should stay with a person. This one is a checker that can tell you to just sign the deal. The pattern is deliberate. Every tool in this genre is normally built by someone with a stake in your answer. The switching calculators belong to the vendors you would switch to. The automation quizzes belong to the agencies that sell automation. The contract checklists live on the blogs of procurement platforms. None of them can afford to conclude that you should do nothing, and a tool that cannot say "do nothing" is not a measurement, it is a script.

A free checker with no signup wall and no server has no stake. It can conclude Sign it, and mean it, because nothing happens to me either way.

The deals people regret are rarely the ones where they asked hard questions and got bad answers. They are the ones where nobody asked, because the demo was good, the renewal letter from the old vendor was insulting, and Thursday felt like a deadline. Ten questions will not make you a procurement department. They will make you the customer who asked, and the vendor prices those two customers differently.