AI Outbound Engine with Voice + Email Orchestration
An outbound engine that orchestrates voice calls, email threads and LinkedIn touches across an ICP, with a single agent memory that learns what lands per segment.
Outbound at the client was four disconnected tools and a shared spreadsheet. Reply rates had collapsed to under 1%.
The team had bought into a Vapi dialer, a sequencer and a data provider, none of which talked to each other.
Leadership wanted pipeline this quarter, not a twelve-month platform project.
Built a single orchestration service on Inngest that owns the cadence state machine across voice, email and LinkedIn.
Introduced a shared agent memory per prospect: what they said, what they opened, what they ignored — with per-segment priors.
Rewrote the email generator to start from a proven pattern library plus the memory, not a cold prompt.
Shipped a per-AE dashboard that shows pipeline attribution and lets reps override, not fight, the agent.
$1.2M pipeline generated in the first 90 days, at a 6.4% reply rate.
Meeting-set rate 3.2x the previous quarter, holding ICP constant.
Reduced per-meeting fully-loaded cost by 58%.
AEs kept the engine after the engagement — real product-market fit inside the team.
- Orchestration design and Inngest functions
- Agent memory schema and retrieval
- Email pattern library and eval harness
- AE-facing dashboard and override UX