Renewal Risk Agent for Migration Cohorts

The short version

The Renewal Risk agent catches the accounts whose operational dashboard says "on plan" but whose qualitative signals say "forming a negative opinion." It scores every migrating account weekly on three dimensions (pricing comfort, product trust, relationship health) using support tickets, Gong sales call transcripts, customer emails, and NPS comments. The high-value signal is divergence: operational fine, qualitative concerning. The agent classifies risk into five types (pricing skepticism, quality concern, competitive evaluation, internal political risk, value misalignment) and recommends a specific intervention. 90 days before renewal, not at the renewal call. By then it's too late.

The agent that catches the renewals you didn't know were at risk

The pricing migration tracker watches behavior. The renewal risk agent watches story. Both matter. They surface different problems.

Behavior tracking catches accounts where outcome volume is below committed minimums or disputes are spiking. Those are operational drifts. Story tracking catches a different problem: accounts where everything looks fine on the dashboards but the customer is quietly forming a negative opinion about the new pricing model.

These are the accounts that show up at renewal saying "we're not renewing" with no warning signs in the operational data. The CPO finds out at the renewal call. The CRO finds out at the renewal call. CS finds out at the renewal call. By then it's too late.

The renewal risk agent fixes this by watching the qualitative signals: support ticket sentiment, sales call transcripts, executive emails, product usage anomalies that don't trigger operational alerts.

What the agent does

Five jobs.

1. Ingest qualitative signals from support tickets, sales calls (Gong/Chorus transcripts), customer emails, NPS comments, and product usage logs.
2. Score each account weekly on three dimensions: pricing comfort, product trust, relationship health.
3. Detect divergence between the qualitative score and the operational dashboard. (When the operational metrics look fine but the qualitative signals say otherwise, the divergence is the early signal.)
4. Classify the risk type and recommend a specific intervention.
5. Hand the account team a 90-day-out renewal risk report every Monday.

The seven components

1. The signal ingester. Pulls qualitative data from five sources: support tickets (last 30 days), sales call transcripts (last 30 days), inbound emails to the account team, NPS comments, and product usage anomalies. Stores raw data with account ID and timestamp.

2. The three-dimension scorer. Each account gets weekly scores 0-100 on:

• Pricing comfort: are the customer's discussions about the new pricing positive, neutral, or negative?
• Product trust: do they trust the agent's outputs, or are they still expressing concerns about quality?
• Relationship health: are interactions with the account team smooth, or is friction increasing?

The scoring is an LLM call. The prompt synthesizes the qualitative signals into the three numbers.

3. The operational scorecard linker. Pulls each account's behavioral data (from the migration tracker agent): outcome volume, dispute rate, escalation rate, payment status. Compares.

4. Divergence detector. When the operational scorecard says "on plan" but the qualitative scorer says "negative," that's the high-value signal. Flag it. When both are negative, the account is already known to be at risk and is likely being worked. When both are positive, the account is healthy. The middle case (operational fine, qualitative concerning) is the agent's specialty.

5. Risk classifier. When divergence is flagged, classify the risk type:

• pricing_skepticism: customer is privately questioning whether the new pricing is worth it
• quality_concern: customer trusts product less than they're saying
• competitive_evaluation: customer is talking to competitors
• internal_political_risk: customer's champion is leaving or is losing internal support
• value_misalignment: customer using the product but getting unexpected outcomes

6. Intervention recommender. Each risk type maps to a specific intervention. Pricing skepticism: CPO sends a personal email with their gross margin math. Quality concern: schedule a senior product review with the customer's team. Competitive evaluation: CRO escalation, plus an executive sponsor visit. Internal political risk: relationship-building with the customer's executive team. Value misalignment: account team co-designs an outcome review with the customer.

7. Weekly report. Every Monday: list of accounts at risk, the risk type, the recommended intervention, the deadline (90 days out from renewal). Sent to CPO, CRO, CCO, and the account owners.

The qualitative scoring prompt

You are scoring customer ${customer_id} on three dimensions based on the last 30 days of qualitative signals.

Recent support tickets:
${support_ticket_summaries}

Recent sales call transcripts (key moments only):
${sales_call_excerpts}

Recent customer emails to the account team:
${email_summaries}

Latest NPS comment (if any):
${nps_comment}

Score each dimension 0-100, where 50 is neutral.

PRICING COMFORT: How does the customer feel about the new pricing model?
- 0-30: explicit concerns or complaints
- 31-50: cautious, asking questions, neutral
- 51-70: comfortable, no complaints
- 71-100: actively positive, advocating for the model internally

PRODUCT TRUST: How much does the customer trust the product's outputs?
- 0-30: actively skeptical, frequent escalations
- 31-50: cautious, verifies most outputs manually
- 51-70: trusts most outputs, normal escalation rate
- 71-100: high trust, low escalation, advocating internally

RELATIONSHIP HEALTH: How is the relationship with the account team?
- 0-30: friction, complaints about responsiveness, escalations to management
- 31-50: neutral, transactional
- 51-70: positive, smooth interactions
- 71-100: deep partnership, customer is a reference

Return JSON: { "pricing_comfort": ..., "product_trust": ..., "relationship_health": ..., "rationale": "..." }

The scoring is the heart of the agent. It's worth tuning the prompt with examples from your specific customer base for two weeks before relying on the output.

What this changes about renewal cycles

Without the agent, account managers do qualitative pulse-checks based on memory and instinct. Some accounts get more attention than others, often based on visibility rather than risk.

With the agent, qualitative risk is surfaced systematically and 90 days early. The account team is working the highest-divergence accounts each week, with specific interventions matched to specific risks.

The cultural shift is that "the customer seems happy" stops being trusted as a substitute for measured signals. The agent watches the signals; the account team works the interventions.

What to try this week

Pick five accounts. For each one, look at their last 14 days of support tickets, the last sales call transcript, and the last NPS comment. Score them on the three dimensions yourself, by hand.

Now compare your score to the operational dashboard for those accounts. Do any of them diverge? Is there an account where the operational metrics look fine but the qualitative signals are concerning?

If yes, that's the one to talk to this week. The agent automates this exercise, but doing it manually for five accounts will tell you whether the divergence is real for your customer base.

The full agent blueprint, including the LLM prompts and the intervention library, is at /artifacts/agent-renewal-risk-during-migration. The companion Pricing Migration Tracker that watches behavioral signals is at /blog/agent-pricing-migration-tracker.

Related

• The PM AI Agent Fleet, the 45-agent operating system this agent slots into.
• Pricing Migration Tracker Agent, the operational companion that watches behavior.
• The Cannibalization Playbook, the strategic context.