In part 3 of this series, we built an early-warning system: a health score that fires an alert the moment an account starts drifting toward risk — or toward expansion. That system works. Customer success teams in North America, Canada, and Australia are increasingly good at generating alerts.
The problem is what happens next. An alert tells you something changed. It doesn't tell your CSM, your account manager, or your renewals owner what to do about it. Without a playbook, alerts pile up in a queue, get triaged inconsistently, and lose their value within weeks.
This is the retention analytics playbook: the operating layer that sits between "the score moved" and "revenue was protected or grown."
Why alerts alone don't move retention
Most B2B SaaS teams in the US, Canada, and Australia already have the data. Usage drop-off, support ticket spikes, low adoption of key features — all of it is visible in a modern customer health score. What's missing is a standard response:
- No default owner — an alert fires but sits unassigned
- No service-level agreement (SLA) — "soon" is not a timeline
- No differentiated response — a churn-risk alert and an expansion-ready alert get the same generic check-in email
A retention analytics playbook fixes all three by pairing every signal with a named play, a named owner, and a deadline.
The three plays every playbook needs
Rather than one generic "customer health workflow," high-performing teams run three distinct plays:
1. The Expansion Play — triggered when usage, seat count, or feature adoption crosses a growth threshold. Goal: get a scoped upsell or cross-sell conversation on the calendar within 5 business days.
2. The Save Play — triggered when health score drops below a risk threshold or churn-risk indicators cluster together. Goal: root-cause the decline and propose a remediation plan within 48 hours.
3. The Renewal Play — triggered 90 days ahead of contract end for any account with a stable or declining score. Goal: lock the renewal conversation before it becomes reactive.
Current Quarter Health Signals
Turning signals into a playbook table
The simplest way to operationalize this is a single reference table your whole CS org can point to. Keep it short enough that people actually use it.
| Signal | Play | Owner | SLA |
|---|---|---|---|
| Health score +15 pts in 30 days | Expansion Play | Account Manager | 5 business days |
| Health score drops below 50 | Save Play | CSM + CS Lead | 48 hours |
| 3+ unresolved support tickets | Save Play (escalated) | CSM + Support Lead | 24 hours |
| Feature adoption stalls 2 cycles | Expansion Play (soft) | CSM | 10 business days |
| 90 days to renewal, score stable/declining | Renewal Play | Account Manager | Immediate |
Account Distribution by Play
Segmented from the same health score feeding the alert system in Part 3.
Worked example: a 60-account cohort
Take a mid-market book of 60 accounts. Before the playbook was in place, health score alerts existed but had no standard response — the cohort's net revenue retention (NRR) sat at 94%.
After introducing the three plays with clear owners and SLAs over two quarters:
- 9 accounts triggered the Expansion Play, and 6 closed an upsell — adding roughly 11 points of expansion revenue
- 14 accounts triggered the Save Play; 10 were successfully retained instead of churning, protecting the equivalent of 6 points of revenue that would otherwise have been lost
- Combined effect: NRR moved from 94% to approximately 109% — without adding a single new logo
The score didn't change the outcome. The playbook did — the score was the trigger, the SLA was what made teams act on it.
Common mistakes to avoid
- Treating the health score as a report, not a trigger. If nobody's job changes when the score moves, the score is decorative.
- One playbook for every account tier. An enterprise save play and an SMB save play should not look the same — scale the response to the account's value.
- No feedback loop. Track which plays actually save or expand revenue, and retire the ones that don't. A playbook that's never revisited quietly stops matching reality.
- Missing ownership at handoff points. The moment with the highest failure rate is the handoff between the alert firing and a human picking it up — assign an owner in the same system that generates the alert, not a separate spreadsheet.
Alerts get attention. Playbooks get results. If part 3 gave your team the early-warning system, this is the operating manual that turns those warnings into retained and expanded revenue.
Coming next in this series
Up next: how do you know if your health score is actually good? Part 5 breaks down the benchmarks.
