All case studies
Analytics

Predictive Satisfaction

Satisfaction scores told the team how customers felt only after the survey closed, often too late to intervene. We built a model that forecasts satisfaction from behavioral signals, moving the warning forward by about six weeks.

−31% churn among flagged accounts
Predictive Satisfaction

At a glance

Client

Subscription services provider

Sector

Analytics

Engagement

Predictive modeling

Timeline

14 weeks

The challenge

Knowing too late to act

Satisfaction was measured by a quarterly relationship survey with a response rate under a third, so the picture was both late and partial. By the time a score dipped, the accounts behind it had often already cut usage or opened a cancellation conversation. Customer success was reacting to churn instead of getting ahead of it, with no way to prioritize which of hundreds of accounts to call first.

Our process

What we did

  1. 01

    Outcome definition

    Worked with success and finance to define what “at risk” means in revenue terms, not just a low score.

  2. 02

    Signal discovery

    Analyzed two years of product usage, support tickets, and billing events to find the behaviors that precede a satisfaction drop.

  3. 03

    Model build

    Trained and cross-validated a model that scores every account’s satisfaction risk continuously, not quarterly.

  4. 04

    Early-warning feed

    Pushed ranked at-risk accounts into the CSM workflow, each with the reasons behind its score.

  5. 05

    Closed-loop validation

    Fed intervention outcomes back into the model so its accuracy improves over time.

The results

0.89

model AUC on held-out accounts

82%

of churned accounts flagged before they left

6 wks

earlier warning than the quarterly survey

−31%

churn among flagged accounts after rollout

1,400+

accounts scored continuously, every day

Could your numbers tell a story like this?

Book a Consultation