Work / Architecture and technical direction
Wevolv3 Outreach
An outbound engine that finds crypto projects ready to buy and starts the conversation.
1 / 2
Context
Wevolv3 didn't need more names on a list. It needed crypto projects that were actually shopping for growth help, reached while the need was still fresh.
The challenge
The engine kept going quiet on its own. The queue of approved leads drained in about a day and nobody could say why. It looked like a volume problem. It wasn't.
My role
Architecture, diagnosis and technical direction, 2026. Built with AI agents working under my direction.
Approach
- Measure every stage of the funnel before changing anything. Optimizing without that is guessing, and the numbers showed the loss was nowhere near where everyone assumed.
- Make every reply leave a durable way to reach that person, so the conversation survives even if the account that started it doesn't.
- Don't trust any number the system reports as healthy until someone has checked it.
Execution
I rebuilt how the engine finds prospects, so its searches stopped coming back empty and stayed stable from one cycle to the next. I audited spend and cut about 9,800 duplicated paid requests a day. And I put a gate in front of every release: nothing ships unless it passes 818 automated checks.
- Empty searches from 11 of 15 to zero.
- Profiles found per cycle from 143 to 428.
- Approved leads per cycle from 9 to 42, against a cap of 50. The queue stopped draining.
- 183 replies from prospects, 37 of them positive, and a conversion rate of 3 to 5%.
- About 9,800 duplicated paid requests a day removed.
The most expensive failures were the ones the system reported as healthy. The admin panel showed zero activity on the three sources producing 69% of that week's leads, and the release process once reported success on a release that shipped nothing. That's now the first thing I check.
The how stays for the conversation. Let's talk.