AI AGENTS, SERVICE / CHURN

A Churn Agent that spots signals before the trouble starts and gets your service team into action in time.

This AI Churn Agent runs always-on in your CRM and monitors what customers do and do not do. Too long without contact, declining usage, support tickets with escalation language, contract renewal approaching without a conversation. The agent scores risk, routes to the right owner and triggers the save playbook. A human steps in at the moment it matters, not only after the cancellation arrives.

Schedule a strategy call See all AI Agents

What you get

01
Churn signal model Scoring on behaviour, contact frequency, sentiment and contract status, weighted to your customer base
02
Live Churn Agent in CRM Always-on in HubSpot, with owner routing and save playbook per risk level
03
Retention dashboard Monthly view of risk portfolio, save rate and lead time between signal and action

Most churn projects start with a nice model and end in a report reviewed once a quarter. By the time a customer turns red, the save window has usually already closed. We do it the other way around. First the signals that genuinely matter in your customer base, then an agent that monitors them daily, then a save playbook and owner routing, and only then do you go live. Lead time gained and save rate as KPIs, not model accuracy on paper.

What the agent does

Scores daily churn risk per customer on behavioural data, contact frequency, ticket sentiment and contract status
Detects early signals: too long without contact, declining usage, falling NPS, support tickets with escalation language, contract renewal gap
Routes risk to the right owner in HubSpot (CS lead, account owner or service lead) with context and suggested next step
Triggers the save playbook per risk level: warm check-in, executive sponsor call, retention offer or escalation
Logs every save action as an activity on the Company record with outcome, so the playbook learns
Monthly cycle for model tuning and playbook updates based on false positives and missed churns

What the agent does not do

Send retention offers or amend contracts autonomously, those remain human decisions
Contact the customer directly, all save actions run via a human with the agent as briefing
Take final responsibility for retention, your CS or service lead remains owner of the portfolio

Under the hood

Which signals, stack and playbooks run under the Churn Agent

A Churn Agent is only as good as the signals it sees and the action that follows. Four things to know before you go live: which signals, which triage, which save actions and what we measure.

Detection

Which signals does the agent monitor

Default four signal groups: contact frequency (too long without a meaningful touch with the owner), usage (declining product or service activity where relevant), sentiment (support tickets and NPS with escalation or frustration language) and contract (renewal approaching without a conversation, or a recent price change). Before we start we do a signal audit on your customer base: which signals have historically predicted churn, how to weight them, where the data gaps are. An agent running on loose signals produces noise, so this is not an optional step.

Triage

How the signal reaches the right person

Risk score is translated into three tiers: green (monitor), orange (warm check-in needed) and red (save action now). Per tier the agent routes to the right owner in HubSpot, with conversation context and suggested next step. The owner receives a task with a deadline and the save playbook attached. No email flood, one prioritised list per day. The service lead sees the portfolio overview in a dashboard.

Save actions

Which playbook per risk level

Four default playbooks: warm check-in (orange, owner call with agenda), executive sponsor call (red and strategic account), retention offer (red with budget mandate, human decision) and escalation path to service lead (red with a blocker at Addmark or supplier). The agent proposes the playbook, a human decides and executes. Outcome is logged back so the agent learns which save action works in which situation.

KPIs

Which outcomes do you measure

Four standard KPIs: save rate (what percentage of red-tier customers are saved within 90 days), lead time gained (how many days earlier the agent spots a risk compared to before), false positive rate (how often the agent flags a healthy customer) and gross retention lift (difference in retention versus the baseline). Monthly review with the service lead to adjust model weights and playbooks based on what works and what does not.

Case · B2B service, [Carel: add: industry and size, e.g. ~120 customer accounts]

From quarterly churn report to daily save actions with lead time that makes a difference

A service organisation had a tidy churn report every quarter. By the time a customer appeared in red, the save window had usually already closed and it ended in a farewell conversation. We first reviewed the historical churn and examined which signals had actually predicted it in this customer base. Then a Churn Agent on HubSpot with daily scoring, owner routing and four save playbooks.

After one quarter the service team spots risk signals an average of [Carel: add: X] weeks earlier than before. The save rate on red-tier customers has risen from [Carel: add: Y%] to [Carel: add: Z%]. More importantly, the conversations that do follow have substance, because the owner has context rather than a sense of goodbye.

Carsten leads the engagement, with Kim on service process and Carel on architecture

On a Churn Agent engagement you work with Carsten as owner of the signal model and data integration. Kim joins for save playbooks and service process, Carel for agent architecture and scope.

Carsten Huiskamp

Service Lead / Churn Agent owner

Owner of the engagement. Builds the signal model, handles the data integration between HubSpot and the agent, and maintains the scoring after go-live. Stays involved in monthly reviews until save rate and lead time are stable.
Kim Janssen

Service process design

Works on save playbooks, owner routing and the collaboration between agent and service team. Ensures adoption lands on your side and that it does not stay at a dashboard.
Carel Schrier

RevOps Lead / Agent architect

Writes scope and architecture, connects the Churn Agent to the broader RevOps setup. Reviews after each month whether the signal model and playbooks need adjustment.

Frequently asked questions

What customers ask about the Churn Agent.

Questions every service team asks before putting AI close to their retention work.

What triggers an escalation to red?

Default three types of triggers. Combination signal (for example: no meaningful contact in 60 days, plus a ticket with escalation language, plus renewal within 90 days). Sentiment trigger (NPS moving from promoter to detractor, or explicit cancellation language in a ticket). Contract trigger (renewal date approaching without a scheduled conversation). Per signal type your service lead sets the threshold. We start with conservative thresholds to prevent alert fatigue and adjust in the monthly review.

How early does the agent detect a churn risk?

Depends on the signals in your customer base. In practice we see lead time of [Carel: add: avg. X weeks earlier than before]. That sounds small, but it is exactly the difference between a save conversation that has substance and a farewell conversation. Before we start we do a historical analysis: which signals preceded the churns you saw last year, and how much time would that have gained.

What if the risk score is wrong?

Two safeguards. First: the agent proposes the save playbook, a human decides and executes. A false positive costs a brief check-in, not an escalation to the customer. Second: the outcome of every save action is logged back (was it justified, customer saved, customer left anyway). In the monthly review we examine false positives and missed churns and adjust weights. Not foolproof, but a continuous improvement loop.

Does this work without HubSpot?

The default stack is HubSpot, because that is where most customer context already lives. Outside HubSpot is also possible, provided there is a system of record where contact history, ticket data and contract information come together. We have done Salesforce, a custom CRM is possible but requires more integration work. Before we start we check what data is available and which signals are feasible from it.

What about data privacy and the EU AI Act?

The Churn Agent processes customer data that already exists in your CRM and does not send data to the customer directly. In most configurations this falls under limited risk under the AI Act, provided you are internally transparent about the scoring model and human-in-the-loop is the default for save actions. We classify per use case, document the data flow and the scoring logic, so your compliance team can review it.

How much does a Churn Agent implementation cost?

Scope depends on data maturity and number of accounts. A typical engagement is 8 to 12 weeks to go-live, with a one-time investment for signal audit, model build, integration and playbook design, plus a monthly retainer for model tuning and review. We give an honest scope after the strategy call, not before we know where you stand. A Maturity scan beforehand helps to sharpen that conversation.

Ready for lead time?

Ready to see churn signals before the trouble starts?

Schedule a strategy call. We look at your current retention work, customer base and HubSpot setup and give an honest scope. Not a sales call.

Schedule a strategy call Take the Maturity scan first