Sales Intelligence for SaaS

A sales intelligence platform that found which accounts were worth a conversation and who to reach inside them. Took it from a three-day prototype to production, and owned it end to end.

The problem

• The existing pipeline was built for service companies and didn't serve SaaS teams, across a base of 2,800+ tracked accounts.
• Identified signals by hand, a custom prompt per signal, a semi automated flow.
• It worked, but it did not scale. I was the bottleneck.

What I built

• Turned the whole flow into an agent, with the dev team.
• Define a use case; the agent read it and picked the signals it needed.
• Pulled the data via Apify and our own database, then generated the opportunity.
• The judgment I applied by hand became something the agent did on its own.

What good meant

The hard part wasn't wiring the agent up, it was deciding what "good" meant. I didn't want the lowest latency or the most opportunities, I wanted the ones worth acting on. So I built a small eval set I'd judged by hand, ran the agent's output against it, watched where it over-flagged, and tuned the prompts, signal thresholds, and model choice until precision held. It was a deliberate tradeoff: accuracy over speed and volume, because a fast answer that wastes a salesperson's time on a bad lead is worse than a slightly slower one that doesn't.