Pick a category with proven revenue, copy the core, and pick one dimension to be clearly better at (Postiz added AI-agent support to a commodity scheduler). He chose X growth tools because Tweet Hunter had shown demand. Once demand is proven, the remaining risk is execution, which you control.
Romàn ran an agency selling lead packages manually before writing software; it validated demand, funded him, and those same clients became Gojiberry's first users at launch.
When it applies: B2B ideas where the outcome can be delivered by hand
He argues that asking if someone likes an idea is pointless unless you ask if they'll pay; waitlist signups and liked tweets don't count. Learned after five products that earned nothing.
Following levelsio: put up a landing page and a buy button for the idea. If people pay, it's validated; if not, you've avoided months of building the wrong thing. He suggests combining it with copying a validated idea and making one part better.