Weekly analysis 19-25 May. Fibre 1Gbps non-renewal +34% vs baseline on the southern belt (38 postcodes), 2 consecutive weeks. Possible competitor campaign in the zone. Mobile Base stable. Contracts >24 months: churn +18% vs annual. Flag for retention marketing.
AGENT · SUBSCRIPTION ANOMALY WATCH
Telco renewal anomalies surface before the churn spike.
Subscription Anomaly Watch monitors weekly the renewal and churn patterns of telco and media subscribers. It identifies anomalous clusters — churn increase in a zone, renewal collapse on a specific offer, patterns compatible with a product problem or an aggressive competitor campaign. Structured alert to the retention team for proactive action.
Subscription Anomaly Watch at work.
On the southern cluster I'll check whether there's a competitor campaign. Activating proactive outreach for the at-risk cluster. Confirmed.
Decisions recorded. Next analysis Sunday 1 June.
Why it exists.
Telco and media operators manage millions of subscribers with structured renewal cycles. Aggregate anomalies — a churn increase concentrated in a geographic zone, a renewal collapse on a recently launched offer, an anomalous pattern in a loyal segment — are often only identified after the problem has become structural.
How it works each week.
Subscription Anomaly Watch reads the renewal and churn data every week, aggregated by geographic area, by offer, and by customer segment. It compares against the configured historical baseline. It identifies statistically significant anomalies and brings them to the retention team's attention with the specific cluster, the anomaly's duration, and the possible causes.
The decision stays with retention.
The retention and marketing team decides on interventions: proactive outreach to the at-risk cluster, cause analysis for the renewal collapse, technical product check if the anomaly is spread uniformly. The agent does not launch campaigns autonomously, does not contact subscribers without approval.
The teams that manage retention in a structured way.
Retention manager
Has every week a structured view of anomalous clusters — by zone, by offer, by segment — instead of waiting for the monthly report when the damage is already consolidated. Proactive action arrives before the cancellation spike.
Week 23 Aggregated churn
South · offer X churn +14%
Center · offer Y stable
North · offer Z +2 points
Brief to the retention director
Marketing manager
Sees the correlation between churn anomalies and market factors (competitor campaigns, technical issues, seasonality). Retention campaigns trigger on data, not intuition.
Switch window · 7d 240 customers
High-value cluster A callback prio 1
Dedicated offer €8/month · 12 months
Engagement < threshold escalation to operator
Expected win-back rate: 22%
Product manager
Identifies cases where the anomaly is concentrated on a specific offer — a signal that the product has a pricing, feature, or communication problem. The data arrives before aggregate customer service complaints.
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
An anomalous churn cluster spotted before it becomes a structural loss.
Monday morning, two and a half million subscribers analysed.
For a telco operator with two and a half million subscribers, Monday morning is the weekly retention brief moment. Subscription Anomaly Watch processed the full subscriber base renewal and churn data over the weekend.
Geographic cluster, loyal segment, growing offer.
The retention manager opens the work channel and finds three signals. The first concerns the Fibre 1Gbps offer in a specific geographic zone: non-renewal rate +34% above baseline over two weeks, geographically concentrated on thirty-eight postcodes in the southern belt, possible correlation with competitor campaign. The second concerns customers with contracts longer than twenty-four months: churn +18% above annual baseline. The third is positive: an offer launched two months ago generates renewals above forecast in three regions.
Targeted outreach. Internal analysis. Marketing comms.
The retention manager decides: proactive outreach to the Fibre 1Gbps southern cluster, internal analysis on the long-contract segment, communication to marketing on the outperforming offer. Decisions stay in the runtime registry for retention action effectiveness analysis.
Declarative rules from the customer's retention and marketing team.
The rules of Subscription Anomaly Watch are declarative. The customer's retention and marketing team defines in a readable format the relevant anomaly patterns (e.g. non-renewal rate above threshold for N consecutive weeks in a zone, renewal collapse on an offer beyond the configured percentage), the analysis segmentations (by geographic zone, by offer, by customer segment, by contract duration), and the historical baselines. CRM and billing system integration is built during delivery via a dedicated adapter.
- 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
- Native delivery channels
- Slack, Telegram, WhatsApp, OpenAI-compatible HTTP
- Scheduling
- configurable per instance (typical weekly, end of weekend for Monday brief)
- Telco CRM integration
- dedicated adapter built during delivery by the Exelab team
- Billing system integration
- dedicated adapter built during delivery by the Exelab team
- Memory
- persistent per instance, pgvector + PostgreSQL FTS
- Registry
- immutable, queryable with a standard SQL client (EU electronic communications audit inspectable)
How Subscription Anomaly Watch works in detail.
The cardinal data point is the renewal and cancellation trace with date, offer, customer segment, and geographic area. The telco CRM and billing system are the primary sources. Integration with these systems is built during delivery via a dedicated adapter. The historical baseline ideally requires twelve months of prior data to calculate seasonality correctly.
Anomaly rules include temporal persistence: a churn increase over two consecutive weeks is treated differently from an isolated spike. Persistence and statistical significance thresholds are configurable. Known seasonal factors (e.g. typical end-of-year cancellation peak) can be included in the baseline to prevent false positives on expected variations.
Yes. Anomaly rules cover both negative deviations (churn above baseline) and positive ones (renewals above forecast). Positive anomalies arrive as opportunity signals: offers or zones performing better than expected, candidates for up-sell actions or campaign extension.
The typical pattern is 10-16 weeks. Discovery and relevant segmentation mapping two weeks, anomaly rule and baseline configuration three weeks, CRM and billing system integration three to four weeks, testing with real data and hand-off to the retention team two weeks. Duration depends on the complexity of the telco CRM and the variety of offers to monitor.
From a 30-minute conversation to the squad in production.
A 30-45 minute conversation to understand how Subscription Anomaly Watch would configure to the customer's case. CRM and billing system, priority segmentations, historical baselines available.