Sales teams hate CRMs because most CRMs are data jails: 10+ hours a week of manual entry for the privilege of slow pages. We built the opposite — a revenue engine where background AI agents enrich contacts, score intent, and draft outreach, turning a 20-minute manual workflow into a 30-second review.
Before: the data-entry tax
Reps logged every interaction by hand, waited seconds for pages to load, and missed buying signals buried in noise. The tool served management reporting, not selling.
The constraint
Anything slower than ~100ms breaks a rep’s flow, and anything requiring manual logging gets skipped. So the architecture had two non-negotiables: zero-latency UI and zero-manual-entry data.
The build (riskiest first)
- Optimistic “headless” UI — interface updates instantly regardless of network; interactions land under 50ms.
- Background AI agents — multi-model enrichment (OpenAI) scores deal health from email sentiment and drafts outreach automatically.
- Deal Room — one surface with only actionable tasks, replacing menu-diving.
The first approach to scoring — single-model, synchronous — was too slow and blocked the UI. Moving enrichment to background jobs with async review was the fix, and it shaped the whole product: nothing AI does blocks a human.
Result
- 95% reduction in manual data entry
- 20-minute review → 30 seconds
- 20+ hours saved per rep, per week
- $2M+ pipeline managed by a lean team
Lessons
- Internal tools should be boring (Next.js, Postgres) everywhere except the one sharp edge (here: AI enrichment).
- Latency is a feature for tools people live in — 50ms vs 3 seconds changes behavior, not just experience.
- Human review of AI output isn’t overhead; it’s the product. The 30-second review is the workflow.
Built with the 10-Day MVP Scoping Playbook, decided via Build vs Buy, scoped from the MVP Scope Worksheet.