I Scanned 97 Markets in 17 Days Looking for a Software Business. Here Is What Killed Most of Them

The short answer: across 17 days of automated market mining, 165 signals were scored and only 36% reached a score of 70 or higher. Almost everything that died failed one of four tests: nobody complains in public, the pain has no owner who can buy, an incumbent already ships the obvious fix, or the person feeling the pain wants to leave the industry rather than fix it. The winners were boring in the same way every time.

I run a scanner that reads public complaint sources every day, looking for software worth building. Forums, subreddits, complaint boards, trade press, court filings, vendor pricing pages, review sites. It scores what it finds and writes a report.

After 17 daily runs I had enough to look at the failures as a dataset rather than a series of disappointments. That is what this post is.

The raw numbers

The median is the interesting number. 67 sits just under the threshold, which means the typical finding is not obviously bad. It is nearly good. That is the expensive kind of idea, the one you can talk yourself into.

The four things that killed ideas

1. Nobody complains in public

The most common kill reason in the logs is tagged WEAK-no-index: the pain is real but nobody writes about it anywhere a search engine can reach.

This matters more than it sounds. If practitioners never post about a problem, you cannot research it without buying your way in, you cannot find customers by search, and you cannot write content that reaches them. The market may exist. Your ability to reach it does not.

Salon payroll, will-call return-to-stock loops, auto-text follow-ups: all real operational pain, all invisible online, all dead on arrival for a solo founder without an existing network in that trade.

2. The pain has no owner who can buy

A pain needs a person whose job it is to fix it and whose budget covers it. The scanner calls this the rail: does the operator own the data, and can they act on it?

Repeatedly a vertical had loud, documented anger and still failed, because the angry party was not the buyer. Complaints about a home warranty company come from technicians who do not choose software. Complaints about parking enforcement come from drivers, not operators. Anger without a buyer is content, not a business.

3. An incumbent already ships the obvious fix

incumbent appears twelve times in the kill notes. Not "there are competitors", which is fine and usually good. Specifically: the obvious version of your idea is already a checkbox feature in a product these people run.

The useful discipline here is to check the incumbent's actual pricing and feature pages before scoring anything, not their marketing. Several ideas that looked like whitespace collapsed in one afternoon of reading six vendors' pricing pages.

4. The person in pain wants out, not a tool

The most interesting failure mode. One appliance-servicer thread produced bigger documented dollar losses than anything else that week, and got sorted below the threshold anyway, because the practitioners were not asking for software. One put it as plainly as possible: no amount of software magic fixes what is a predatory design issue.

When your target user's honest answer is "I am going to leave this industry", you are not selling a tool. You are selling a coping mechanism for a problem they have already decided to escape.

What the survivors had in common

Every signal that cleared 70 shared three traits:

The consequence is money, and someone counts it. Not inconvenience, not wasted time in the abstract. A number on an invoice that came back wrong, and a person whose job is reconciling it.

The operator owns the data. They already have the records needed to prove the problem. Software that requires your customer to first go get data from someone else fails, because that request is the hard part and you have not removed it.

The pain is fresh. The strongest signals traced to a change in the last eighteen months. A new intermediary, a new fee structure, a policy flip. Old pain has settled into workarounds. New pain is still raw and unbudgeted.

The honest limits of this data

This is one scanner, one operator, and seventeen days. It over-samples verticals where people post in public, which is exactly the bias described in kill reason one, so treat the 36% pass rate as a property of this method rather than a law about markets.

It also scores signals, not businesses. A signal clearing 70 means "worth a validation page and ten conversations", not "build this". Several 70-plus signals will die at the next gate, and that is the gate doing its job.

What I would tell someone starting this

Score the kill reasons, not just the winners. I got more out of seventeen days of failures than out of the twenty-three that passed, because the failures are where the pattern lives. If you are hunting for something to build, write down why each idea died in a form you can count later. After thirty of them you will stop having opinions about markets and start having data.

The four tests above cost nothing to apply. Public complaints, a buyer who owns the data, an incumbent check against real pricing pages, and an honest read on whether your user wants a tool or an exit. Most ideas fail at least one, and it is much cheaper to find out on day one than after a month of building.