The workflow before#
MasterControl sells quality and compliance software to life sciences companies, and many of its renewals are worth $100K or more a year. The customer success team judged account health with rules of thumb. The signals that mattered were all there: product usage, support tickets, contract and payment history, training completion. They just sat in five separate systems, so building a picture of any one account meant pulling and stitching data by hand. In practice, warning signs showed up close to renewal. By then the team had a few weeks to change a customer's mind instead of a quarter to fix what was going wrong.
What we built#
In a 12-week proof of concept, we built a churn model that gives every account one risk score and says why. It reads three years of customer history across usage, support, contracts, payments, firmographics and training, and predicts which accounts are likely to churn in the next 90 days. Customer success gets a ranked list with reasons, and a playbook for each kind of risk.
One score per account
The model combines usage trends, ticket patterns, contract and payment history, firmographics and training completion into a single score. It holds 85% accuracy on held-out data, and the top 20% of flagged accounts churn at three times the base rate. We set the window at 90 days because that's long enough for an intervention to work before renewal.
Reasons a CSM can act on
Each flagged account shows the factors behind its score, such as falling engagement or rising ticket volume. We built the explanations in from the start because a customer success manager won't call a customer on a number they can't explain. The reason tells them what the conversation should be about.
Playbooks matched to risk drivers
We paired the model with intervention playbooks for the main drivers: falling engagement, rising ticket volume, low adoption of advanced features, contract reductions and declining training completion. A flag arrives with a suggested next move, so the team spends its time on the call, not on figuring out what to say.
How we delivered it
We ran the work as a 12-week proof of concept. The first weeks went to joining the five data sources and agreeing on what churn meant. Model training and validation came next, then the dashboard and playbooks with the customer success team, so the output matched how they already reviewed accounts.
Before and after#
| Measure | Before | After |
|---|---|---|
| At-risk account identification | Rules of thumb, near renewal | Model score, 90 days ahead |
| Churn rate in flagged accounts | No ranking | Top 20% flagged churn at 3x the base rate |
| Account health review | Manual pulls from 5 systems | One scored dashboard with reasons |
| Response to a flag | Decided case by case | Playbook matched to the risk driver |
What's next#
Next is moving the scores and alerts into Salesforce, so account managers see risk inside the tool they already work in, and retraining the model on what happened after each intervention. We estimate that protecting even a few points of churn on $200M of recurring revenue is worth about $2M a year.
