I spent $500 building something I knew was broken on day 3.
The painful part isn't the money. It's that I had the signal. I ignored it anyway.
Here's what happened with GetLaunchIntel — and the thing nobody talks about when you know the right move and still don't make it.
The idea was clean.
New businesses get registered every day in government databases. They haven't been cold-called yet. They haven't been emailed to death. Get to them first — before any competitor list picks them up.
Built the whole thing. 12 country registries with API access. Serper pulling Google data. Hunter.io for emails. AI enrichment layer on top. Fresh leads with company name, registration date, contact number, website, social pages, and a gap analysis of their existing site.
On paper — perfect.
Then on day 3, something felt off.
I pulled a sample. The data looked great. But the websites of these "newly registered" businesses were old, established, and ranking well on Google.
I told myself: maybe they started operating early and registered late.
I moved on.
That one sentence cost me $500.
Over the next 4 days I backfilled 60,000 rows. $400 in Serper API credits. 4 days of backend work.
A month later, I ran a proper diagnostic.
What Serper was actually doing: when it couldn't find the newly registered company name on Google — because it literally just got registered — it returned the closest matching business instead. Similar name. Different company. Wrong data. All of it.
Every single lead was wrong.
The tool had no value. All I actually had was a business name and a registration date. No contact. No email. Nothing usable.
Now the question I keep sitting with isn't about the money.
It's: why do we do this?
You've read about this. You know the rule: validate before you build. Test before you scale. Don't backfill 60,000 rows before you've confirmed the core data is real.
I knew this. I've written about it. I've told other people this.
And still — I rationalized. I justified. I moved forward.
There's something that happens in the middle of building.
You've already invested a few days. The idea feels real. The UI looks good. The architecture is working. And when a signal comes that questions the whole thing — your brain doesn't say "stop and investigate."
It says "find a reason this is fine."
Psychologists call this escalation of commitment. Sunk cost dressed up as confidence.
The more effort you've already put in, the more expensive the honest answer becomes — and the cheaper it is to manufacture a comfortable one.
The fix wasn't complicated.
On day 3, I should have downloaded 20 leads, dropped them in Claude, and asked one question: does this data actually match these companies?
That's a 30-minute check. It would have killed the project cleanly, cheaply, and early.
Instead I waited until week 6.
What I'm taking forward:
When something feels off — that's not doubt. That's data. Treat it like data.
The moment you find yourself explaining away a concern instead of investigating it — that's the moment to stop, not accelerate.
$500 is a cheap lesson if I actually keep it.