Redesigning a Customer Health Score a SaaS Team Had Stopped Trusting

A North American SaaS company whose customers renew on annual cycles, where catching churn risk early is the difference between a save and a loss. Customer Success was built to protect that revenue. The health score meant to point the team at the right accounts had drifted far enough from reality that most Customer Success Managers (CSMs) had stopped acting on it. The problem was not the model. It was everything feeding it and everything built on top of it.

• Signals that predict churn already lived in the stack and never reached the score: a departing champion, a souring support ticket, product usage falling off, an implementation that had stalled

• The math did not normalize. No weighting for account size, no adjustment for customers whose usage is seasonal by fiscal year, and no link from a score change back to what caused it

• The score read ticket counts and nothing inside them, so three angry tickets and three calm ones looked identical

• Recorded customer calls, one of the richest sources of risk in the business, were never read by the score at all

• Even when the score was right, CSMs triangulated across four to six tools to work out why, then triaged by intuition. Morning prep alone ran up to 90 minutes before the real work started

• Nobody could prove the score predicted churn, so there was no basis to trust it and no way to improve it

• Ran a four-week discovery across the Customer Success organization: 15 working sessions with 13 people spanning CSMs, RevOps, the data team, support, and leadership, including live walkthroughs of the tools CSMs open every morning

• Mapped the health-score workflow as it actually runs, then sorted the breakdowns into five structural categories, from signals-in through to the validation gap

• Specified five signal pipelines to give the score what it was missing: sentiment and severity from support tickets, per-call analysis from recorded calls, implementation status, a trusted product-usage source, and fiscal-year-aware seasonality

• Redesigned the score to weight account size and renewal timing, and to tie every movement back to its driver, so a number becomes an explanation

• Designed four CSM agents against the biggest time sinks the sessions surfaced: a morning brief, an outreach draft, a one-page call prep, and a revenue-weighted watchlist, each living inside tools the team already uses

• Sequenced the whole build into four waves, each shipping visible value, on a foundation that backtests the score against accounts that actually churned before anyone is asked to rely on it

• A four-week workflow diagnosis across 15 sessions and 13 stakeholders, delivered as 8 documents: current state, future state, tooling landscape, quantification workbook, change-adoption report, agent prototypes, solution detail, and implementation plan

• 5 structural findings, each traced to specific evidence from the CSM workflow

• 5 signal pipelines specified, turning ticket text, call voice, product usage, implementation status, and fiscal-year seasonality into inputs the score can finally read

• A recalibrated score design with driver-level explainability, weighted for account size and renewal timing

• 4 CSM agents designed, mapped to the morning-triage, outreach-drafting, call-prep, and account-ranking workflows, each embedded in existing tools rather than a new app

• A four-wave delivery sequence, each wave shipping usable value instead of waiting on a single end date

• A validation method that compares a 90-day-old score against real churn outcomes, to prove the model and tune which signals earn their weight

• A persona-segmented change-adoption plan built on the reality that the team was already absorbing heavy change

• The blind spots are named and addressable. The signals that quietly predicted churn now have a defined path into the score for the first time.

• A score movement can be explained, not just observed. When a number drops, the driver comes with it, which is the line between a metric the team argues about and one it acts on.

• Trust has a mechanism. The validation layer means the score earns belief by predicting real churn, rather than asking CSMs to take it on faith. That was the finding that gated the other four.

• The CSM day has a redesign, not another dashboard. Triage, call prep, and outreach drafting were rebuilt around the tools the team already lives in, so adoption does not hinge on learning a new app.

• The client holds a sequenced plan with named owners and a quantified opportunity model, not a slide of ambition. Each wave is scoped to ship on its own.

This was a Workflow Transformation Sprint, so the benefit here is identified, not yet earned. The outcomes above are the diagnosis, the redesign, the signal and agent design, and the sequenced plan to build them.

The opportunity was built bottom-up from several independent components, each baselined against the client's own numbers and deliberately discounted to stay conservative:

• CSM hours returned by automating morning triage, call prep, and outreach drafting across the team;

• revenue protected by lifting the save rate on the at-risk account pool;

• and further revenue protected by a more accurate score that points attention at the accounts intervention can actually save.

Combined, the identified annual benefit exceeds seven figures, and runs into the millions across the life of the redesigned workflow. The client's own leadership considered that conservative.

Architech does not yet count it as a win. The changes to process, AI, data, and systems take several months to build, and several more before enough customers move through the redesigned workflow to show a trend in the KPIs that matter: retained revenue, recovered CSM time, and churn-forecast accuracy. Those numbers get reported here when they are measured in production against the baselines captured during the sprint, not before.