It started with one bad Tuesday.

Sofia, 2020. Hristo, Nikolay, and Adrian were running private-label Amazon brands out of the same small office. Different categories, same problem. The PPC tools on the market were either expensive agency retainers or rule-based scripts that broke the first time Amazon changed a report column. Either way they swallowed three or four hours a day across the team: pulling search-term reports, pivoting them in Excel, cross-checking against margin sheets, then nudging bids one keyword at a time.

Manual work eventually fails on a Tuesday. A search-term export came in with one column misaligned. By the time anyone caught it, six SKUs had spent a full day's budget on terms that had not converted in three weeks. Not a disaster, but the third time something similar had happened that quarter.

Instead of patching the sheet again, we sat down and pulled PPC apart properly. For a few months we worked through the parts most operators only skim, mapping how Amazon's ad system actually behaves underneath the dashboards. The conclusion was uncomfortable: the decisions were not as complicated as the tools made them look. They were just being made too slowly, too inconsistently, and without the product's own context plugged in.

So we built that context into an algorithm. Nikolay coded the first version. Adrian wired in the Amazon APIs. Hristo ran it on his own account first to see whether the rules behaved the way the spec said they would.

It was rough, but it was consistent. Within three months it had paid back the time spent building it. Within six it was running for a few other brands we knew personally who had asked to try it. By the end of the first year we had stopped offering it to friends because we could not onboard fast enough.

The principle has not changed since. We build for the operators who will be the ones fixing things on a Tuesday morning. Because that is still us.