Good morning. Three customers requiring attention today. Customer A (Retail Corp): 6 weeks of commercial silence + renewal deadline in 28 days. Customer B (MedTech): 62% drop in primary-feature adoption + 2 tickets open for 9 days. Customer C (Logistics South): last renewal decided 11 days late, recent sales email unanswered.
AGENT · CHURN WATCH
Customer churn signals arrive before the cancellation notice.
Churn Watch reads three sources every day: the CRM (contact frequency, quality of recent conversations), the customer success system (product adoption, open support tickets), and the renewals system (upcoming deadlines, last customer action). It recognises patterns consistent with churn risk and brings them to the sales team's attention.
Churn Watch at work.
Confirmed. Escalating Customer A to the senior rep now. Opening a check-in call for Customer B on Friday.
Decisions recorded in the CRM. Tasks assigned to the relevant owners. Next reads scheduled for tomorrow.
Why it exists.
Cancellation almost always arrives after something has already happened. The customer stopped responding to the account manager's emails two months earlier, opened three unresolved support tickets, stopped using the feature that led them to sign. By the time the cancellation letter arrives, the relationship has been over for weeks.
How it works each day.
Churn Watch works upstream. It reads daily the three sources that describe the state of the customer relationship: the CRM (contact frequency, quality of the last interaction, email sentiment if available), the customer success system (product usage frequency, features used, anomalous events), the renewal system (upcoming deadlines, the customer's last action on previous renewals). It compares the customer's state with the baseline of comparable customers, identifies significant deviations, brings to the attention of the commercial team the customers showing patterns compatible with churn risk.
The decision stays with the team.
The decision stays with the account manager or the customer success lead. The agent does not write to the customer, does not suspend contracts, does not change the CRM state without approval. It sets the table for attention.
Three business functions that change the way they manage the customer portfolio.
Customer success lead
Sees in a structured way the customers who require attention before the deadline, not once the deadline has passed. The renewal conversation changes nature: it stops being an emergency, it becomes a planned follow-up.
CustomerSignalRisk
Idritech SpA Silent 12 days · 3 emails unanswered high
Nexar Group Proposal expired · no extension med
Orbis Srl Meeting cancelled 2× med
Proxa Corp Active contact · reply < 24h low
Head of sales
Sees the size of the risk across the entire customer portfolio, and can allocate the team's time to the cases with greater weight. Intervention decisions (check-in calls, escalation to a senior account manager, CEO intervention on key customers) arrive based on structured patterns, not on intuition.
RB RevOps Brief
Idritech SpA silent 8d · proposal expiring
Nexar Group proposal sent · no reply
Aludra Srl signature expected this week
Account manager
Does not chase all customers uniformly. Their weekly attention focuses on the few cases that truly need their intervention, while all other relationships proceed without interference.
87 Tone too formal — try a direct opener
71 Long message: summary in the first 2 lines
94 Excellent — consistent with the customer history
A daily read that transforms the way renewals are managed.
Churn Watch reads three sources every day.
For the customer success lead of a regulated B2B company, reading the morning Slack channel is the first step of the day. Churn Watch has completed its overnight cycle: it read the CRM interactions of the last two months across all 340 active customers, product adoption, open tickets, renewal deadlines in the next 90 days.
Eight customers, each with the risk pattern in two lines.
Today's list has eight customers. For each one the lead sees the risk pattern in two lines: Customer A, 6 weeks of commercial silence + renewal deadline in 28 days + last commercial contact four months ago; Customer B, 60% drop in primary-feature adoption + support tickets open for 12 days; Customer C, last renewal decided two weeks late and recent sales emails unanswered.
The lead decides. The agent records and creates tasks.
The lead decides case by case. Customer A: escalate to the senior rep. Customer B: customer success specialist check-in call within the week. Customer C: monitor for two weeks and reassess. Decisions go into the CRM as assigned tasks. The event stays in the runtime audit registry, queryable by the customer success lead with a standard SQL client.
Declarative rules from the customer's customer success team.
Churn Watch's risk rules are declarative. The customer's customer success or sales operations team defines in a readable format what "risk pattern" means for their case (e.g. commercial silence for N weeks on a customer with plan X, adoption drop on feature Y beyond the configured threshold, support tickets open for more than N days). The rules live in the customer's repository, versioned, validated at agent startup.
- Language
- TypeScript (Node.js)
- LLM model
- customer's choice: Anthropic, OpenAI, Mistral, open source models hosted internally, AWS Bedrock for a private model
- Built-in controls used
- pii-detector, tool-rate-limit, message-length-limit
- Native delivery channels
- Slack, Telegram, WhatsApp, OpenAI-compatible HTTP
- Scheduling
- configurable per instance (typical late-morning daily read)
- CRM integrations
- HubSpot native (contacts, deals, tickets, meetings, tasks, notes, company, email)
- Other CRMs
- Salesforce, Microsoft Dynamics, industry-specific systems: integration delivered during the project by the Exelab team
- Customer success integration
- dedicated adapter delivered during the project by the Exelab team (system varies per customer)
- Memory
- persistent per instance, pgvector + PostgreSQL FTS on customer history
- Registry
- immutable, queryable with a standard SQL client
Frequently asked questions about the agent.
A dashboard shows a score. Churn Watch delivers a short list of customers — with the specific risk pattern for each one — to the sales or customer success team's work channel, next to the CRM record, before the deadline. The difference is in the flow: the dashboard is something you go look at; the agent arrives at the start of the day.
The CRM is the core system, because it tracks commercial interactions. The customer success or product-adoption system is the second, describing what the customer does with the product. The renewals system (or the CRM module dedicated to renewals) is the third. Polyant's native tools cover HubSpot. For other systems (Salesforce, Microsoft Dynamics, sector-specific management tools, proprietary customer success platforms), the integration is delivered during the project by the Exelab team.
Every flag decision is inspectable, and the outcome of each rule is tracked separately. When a customer is flagged for a pattern the customer success lead considers irrelevant to the specific case, the team updates the declarative rule, tests it in the development environment, and promotes it to production. Rule improvement stays within the customer's team — it does not pass through an external vendor.
The value of Churn Watch is measured on the percentage of flagged customers who, after the team's intervention, remain active customers at the next renewal. The pre-agent baseline (churn percentage without structured flagging) is established during discovery. The first weeks of operation compare the state of flagged customers against the baseline. Churn Watch is one of the agents the customer can choose to build for their sales or customer success team.
From a 30-minute conversation to the squad in production.
A 30-45 minute conversation to understand how Churn Watch would be configured for the customer's case. Which systems (CRM, customer success, renewals), which risk patterns, which read frequency.