Predictive Satisfaction Score

How a telecom provider forecasted satisfaction in real time—before customers were ever surveyed

Scroll

Challenge

Too Late to Act

A major telecom with millions of subscribers was investing heavily in NPS and CSAT programs, but results arrived too late to act. By the time survey data was processed, at-risk customers had often already churned or escalated. The CX team needed a way to anticipate satisfaction—and target retention and care efforts—before feedback was even collected.

By the time satisfaction survey results came in, the damage was already done. Customers had dropped calls, visited stores, churned—or posted about it. Weeks later, a score would show up. But it was a postmortem, not a signal.

CX, Delayed

Customer experience data was reactive. Surveys captured how people felt—but long after they acted. Care teams couldn’t prioritize. Marketing couldn’t personalize. And operations couldn’t see emerging risks.

What If You Could Predict It?

What if you could anticipate satisfaction based on behavior—before feedback was even submitted? What if you could intervene early—before the next bill, visit, or cancellation?

Approach

So We Modeled the Score

We partnered with the telco’s CX, data science, and network teams to build a predictive model—one that estimated satisfaction scores using operational signals, usage patterns, and digital behaviors in near real time.

Solution

We Trained on 750K Customers

We used two years of NPS and CSAT data—linked to service usage, care interactions, call logs, network incidents, and mobile app activity. A supervised ML model was trained and validated on over 750,000 customers, predicting satisfaction with 86% accuracy at the segment level.

Methodology

We followed a structured pipeline: (1) unify survey and operational data at customer level with consistent identifiers and time windows; (2) engineer features from usage, care, network, and app behavior; (3) train supervised models (e.g. gradient boosting) to predict NPS/CSAT segment, with cross-validation and holdout testing; (4) deploy weekly score updates and integrate outputs into CRM and care dashboards; (5) measure lift from proactive interventions vs. control segments.

Data sources

NPS and CSAT survey responses (two years; 750K+ linked customers). Service and usage data: plan type, tenure, data usage, call/SMS patterns. Care interactions: contact frequency, channel, resolution codes. Network and incident data: outages, latency, coverage. Mobile app activity: sessions, feature use, support requests. All joined at customer level with appropriate lag windows for prediction.

We Scored the Entire Base

Every active customer was assigned a predicted satisfaction score—updated weekly and mapped to key drivers. The client could now target low-predicted-sat customers with retention offers, support calls, or proactive fixes—before dissatisfaction escalated.

“Before, we were reacting. Now we’re in front of it. We can see risk coming, not just explain it after the fact.”

— VP, Customer Experience

We Integrated It Everywhere

The score was embedded into CRM systems, care dashboards, and marketing lists—used daily by agents, analysts, and journey teams. Visuals included segment-level score distributions, trend charts of predicted vs. actual satisfaction over time, driver importance plots, and care queues ranked by predicted risk. It became the new backbone of CX decisions across the enterprise.

Implementation timeline

Weeks 1–4: Data discovery, linkage, and feature design. Weeks 5–10: Model development, validation (86% accuracy target), and back-testing. Weeks 11–14: Integration with CRM and care tools; pilot with one care team. Weeks 15–18: Rollout to six teams and weekly refresh cadence. Steady state within five months of kickoff.

Take the next step

Explore how Intellimark can help you anticipate customer sentiment—and act before it’s too late.

Outcome

Satisfaction became a leading indicator, not a lagging one:

Metrics / Results

86% model accuracy across key segments

750K customers modeled with weekly score updates

14% churn reduction among at-risk groups reached proactively

6 teams using the score daily across care, marketing, and analytics

Fin.

Knowing how customers feel is good. Knowing before they tell you? That changes the game.