opsmarshal
Workflow 11 — RevOps Infrastructure

Churn / Expansion Sentinel

Fuses product telemetry, support sentiment, invoice behavior, and stakeholder changes into account health that means something — and triggers renewal-risk or expansion playbooks with CSM approval, months before the renewal date makes it urgent.

At a glance

The Churn / Expansion Sentinel combines product telemetry, support sentiment, invoice behavior and stakeholder changes into an account health score, and triggers renewal-risk or expansion playbooks with CSM approval well ahead of the renewal date. It is one of fifteen workflows in the Enterprise GTM Agent Stack, an Opsmarshal series on GTM stack architecture.

Owned by
CS Ops
Writes back to
CS platform, CRM renewal objects, Audit log
Held to
≥ 90 days — Risk detection lead time
Gate
Permission → Evidence → Approval → Audit
Writes to: CS platform + renewal objectsApproval: CSM on every playHorizon: 2 quarters pre-renewal
Business Pain

Why this exists

The status quo

Health scores today are a traffic light nobody trusts: green accounts churn, red accounts renew, and CSMs learn the score is theater. Real risk hides in trajectories — seat stagnation, champion exit, support tone shift.

What this workflow changes

Health becomes an explained number: every score shows its drivers and their direction. Risk plays trigger with enough lead time to matter, expansion signals route with the same rigor, and the CSM stays the owner of every touch.

Architecture

Reference architecture

Signal Sources
Product telemetryusage depth, seat trends, feature adoption
Support + sentimentvolume, severity, tone trajectory
Billing + stakeholdersinvoice behavior; champion moves via WF-05
Agent Layer
Health Modelper-segment weights, versioned, recalibrated on real renewals
Risk/Expansion Detectortrajectory-based, not snapshot-based
Play Selectorrisk and expansion playbooks from the CS-owned library
Governance
Gate: permission → evidence → approval → auditevery play requires CSM approval · manual overrides expire and carry reasons · overridden scores are reported alongside the raw score
Systems of Record
CS platformhealth + drivers + trend
CRM renewal objectsrisk-adjusted forecast inputs
Audit logplays, approvals, overrides
DecisionChoiceRationale
Explainability over accuracyA slightly worse model CSMs trust beats a black box they ignoreAdoption is the metric. Driver-level explanations make the score arguable — and arguable means used.
Segment-specific weightingEnterprise health ≠ SMB healthLogin frequency predicts SMB churn and means nothing at 5,000 seats. Weights are per-segment config, versioned, recalibrated against actual renewal outcomes.
Live Demo

Run a scenario through the gate

Three curated scenarios — one that passes, one that flags for a human, one that the gate blocks. Watch the verdict, the evidence rows, and the audit log.

Account Health Sentinel

Governance audit log

[ready] agent idle — pick a scenario
Operator Control Panel

Someone has to run this thing

Agents don't remove operators — they change what operators watch. Console spec: who owns it, what they check, what pages them, and how they pull the plug.

Churn / Expansion Sentinel — Ops Console

Owner: CS Ops

Daily monitors

  • Telemetry pipeline health — a silent source corrupts every score downstream
  • Triggered-play queue awaiting CSM action

Weekly reviews

  • Score-vs-outcome calibration on closed renewals
  • Play acceptance by CSM — low acceptance means the model or the plays lost trust

Alert thresholds → who gets paged

  • Health data source silent 24h → CS Ops
  • Commit-quarter account drops 2 health bands → CSM + manager, same day

Manual controls

  • Manual health override — expiring, reason required, reported next to the raw score
  • Play library editor — CS-owned
  • Kill switch — scoring continues in shadow; play triggers halt
Eval Metrics

What this workflow is held to

≥ 90 days
Risk detection lead time
before renewal — late detection is decoration
Calibrated
Predicted vs. actual churn by band
published quarterly, the model's public exam
→ 0
Unflagged churn
churned accounts with no prior flag — the shame metric
Failure Modes

How it breaks, and what catches it

Green-account churn

The model misses relationship-driven exits that telemetry can't see. Mitigation: Champion-tracker signals (WF-05) feed health directly; relationship risk is a first-class driver, not a footnote.

Alert fatigue

Every dip triggers a play until CSMs mute the sentinel entirely. Mitigation: Play budgets per CSM per week; sub-threshold changes batch into digests instead of interrupts.

Override gaming

Manual overrides hide risk to protect the renewal forecast. Mitigation: Overrides expire, require reasons, and are reported alongside the un-overridden score — hiding risk becomes visible work.

Opsmarshal builds workflows like this one inside client stacks — see GTM stack architecture and AI enablement and agentic workflows, or book a call to walk through your own.

opsmarshal

Revenue Operations and AI Agentic Workflows for B2B companies in transformation. We build the machine. You run the company.

Services

Company

© 2026 Opsmarshal. All rights reserved. Privacy  ·  Terms