Predicting dissatisfaction—and calling before customers reported it.
Strong scores. But limited feedback. We helped a global provider of enterprise IT infrastructure hear what customers weren’t saying.
Customer support scores were high. The problem? Our client received feedback on less than 1% of its support cases—not enough to trust the scores.
Instead of pushing more surveys, we suggested a different route—predict who was dissatisfied without asking them. And reach out.
The call customers didn’t see coming.
We built Predictive Alerts—a model that flagged dissatisfied customers before they said a word, and alerts that notified service managers. Then we wrote call scripts so they could reach out.
High scores. Declining responses.
The provider’s customers buy complex infrastructure and rely on support to keep it running.
Satisfaction scores looked great. But they didn't represent the full picture.
Every month, the provider closed 420,000 support cases and got survey responses on just 0.8% of them.
And response rates kept falling.
Sending more surveys wasn’t an option. Some customers had no contact information on file. Others risked survey fatigue. We found the answer in the provider’s existing data.
A risk score became an alert. The alert became a call.
420,000 service records a month. 3,400 survey responses.
We cleaned the data first—removed duplicates, filled in missing fields, cut outliers and matched each survey response to its service record. Three factors predicted dissatisfaction most consistently—product type, issue severity and resolution time.
We built a predictive model around those trends. Every case got a score from 0% to 100%: the odds the customer felt dissatisfied. Then we scored the customers who never answered a survey—and saw who was at risk, even if they’d never said a word.
But knowing wasn’t enough—the provider needed action. So we built alerts.
When a case’s score crossed a set threshold, the model flagged it and a feedback platform emailed the responsible service manager.
We trained managers on why the calls mattered, what to say and how to log the result. The manager's job was simple: listen, then fix whatever turned up. Each call took 20 minutes.
The provider started with a pilot—a small group of managers covering two product lines. After each call, managers logged the customer’s reaction in a short survey. The answers showed whether the calls worked. They also fed back into the model, sharpening its outputs.
Don’t take our word for it.
"The use of predictive analytics is making it easier and easier for our teams to have in-depth customer conversations. Follow-up drives engagement and allows us to develop stronger relationships."
Calls uncovered issues customers never reported. Quarterly renewal revenue rose.
Over the program period, the provider’s quarterly renewal revenue rose 9.8%.
Before the alerts, managers only called customers who said they were dissatisfied—under 300 calls in a typical month. With alerts, managers made seven times as many calls—nearly 2,000 per month. And 32% of those calls uncovered an issue that needed corrective action.
We tracked relationship scores for a year afterward. Customers who spoke with a service manager were more loyal than those with a similar bad experience who didn’t.
So we scaled.
We expanded from two product lines with alerts to 12. And a small group of service managers grew to 300+ who received alerts and made mandatory calls.
The business used to hear from unhappy customers only if they spoke up. Now it reaches them first. No survey, no complaint. Just a call the customer didn’t see coming.
Increase in monthly follow-ups
Rise in quarterly renewal revenue
Customers surprised or glad to get the call