Most "AI SDR" builds automate the send. This one automates the decision before the send: fusing intent, hiring, tech-stack, and funding signals into a play selection — which sequence, which persona, which angle, and whether to reach out at all. A human approves the play; the agent handles the mechanics; every disposition writes back to CRM.
Signal-Orchestrated Outbound is a governed agent workflow that fuses intent, hiring, tech-stack and funding signals into a play recommendation — which sequence, which persona, which angle, or whether to reach out at all — that a human approves before any message sends. It is one of fifteen workflows in the Enterprise GTM Agent Stack, an Opsmarshal series on AI enablement and agentic workflows.
How the flow runs — live
Suppression is a valid exit at any stage — logged with a reason, never silent.
Teams bought AI email tools, 10x'd volume, and watched reply rates collapse while domains got flagged. The bottleneck was never sending capacity — it was knowing which account deserves which motion this week, and having proof for why.
Signals stop being a column in a spreadsheet and become the input to a routing decision. A funding event at an ICP-fit account with a fresh VP hire triggers a different play than lone website intent. Low-conviction accounts get suppressed — deliberately doing nothing is a valid, logged output.
| Decision | Choice | Rationale |
|---|---|---|
| Plays are a curated library | Agent selects; humans author | The agent never invents a motion. Marketing and sales leadership own the playbook; the agent's job is pattern-matching signals to it. This keeps message quality governable. |
| Suppression is an output | "Do nothing" is logged with a reason | The costliest outbound failure is burning a future buyer with a bad touch. Below-threshold conviction produces a logged suppression, reviewable in the same audit trail. |
| Signal conviction decays | Time-decay per signal class | A funding round is warm for 90 days; website intent for 7. Fusion scores decay on class-specific half-lives so stale signals can't trigger fresh sends. |
| First touch is human-gated | SDR approves message 1; follow-ups auto-run | The highest-risk moment is the first impression. Gating only that touch keeps throughput high while keeping a named human accountable per account. |
Toggle the signals observed for a sample account. The fusion agent scores conviction, the selector picks a play — or suppresses — and the gate rail logs the full decision.
No signals observed. Suppression logged — this account stays untouched, and that decision is auditable.
Governance audit log
Agents don't remove operators — they change what operators watch. This is the console spec: who owns it, what they check on what cadence, what pages them, and how they pull the plug.
Intent providers change taxonomies; a topic rename silently zeroes a signal class. Mitigation: per-class volume monitors alert when a signal source deviates ±40% from its 30-day baseline.
Correlated signals (funding → hiring → press coverage) triple-count one underlying event. Mitigation: fusion agent clusters signals by root event before scoring; one event contributes once.
SDRs rubber-stamp first-touch approvals at volume, hollowing out the gate. Mitigation: approval-rate telemetry — if an SDR approves >95% in under 5 seconds each, the sample rate of required reviews increases and the pattern is surfaced to the manager.
The agent keeps selecting a play whose reply rate quietly died two months ago. Mitigation: per-play performance floors — plays under threshold for 30 days are auto-benched and flagged to their human author.
Opsmarshal builds workflows like this one inside client stacks — see AI enablement and agentic workflows and revenue operations, or book a call to walk through your own.
Revenue Operations and AI Agentic Workflows for B2B companies in transformation. We build the machine. You run the company.