Satisfied but Churning: How a Telecoms Operator Used Market Research and Strategy Advisory to Fix a Contact Centre Analytics Programme Whose Sentiment Scores Were Not Predicting Customer Departure

Executive Snapshot

Client

Mobile Telecoms Operator, South Africa

Situation/Challenge

The client had deployed a contact centre analytics platform that scored customer sentiment across all voice and digital service interactions and flagged customers with declining sentiment scores for proactive retention outreach. After eighteen months, the customer success team had run the programme diligently but churn among sentiment-flagged customers was not materially lower than among unflagged customers, and post-churn analysis showed that a substantial share of churned customers had received no negative sentiment flag at all in the three months before they left.

Objective

Commission a market research report benchmarking contact centre analytics churn prediction effectiveness at comparable mobile operators, then engage strategy advisory to diagnose why the sentiment model was missing churners and identify what additional or alternative signal would improve its predictive accuracy.

Constancy Researchers Solution

Market Research Reports combined with Strategy & Growth Advisory, a Contact Centre Analytics Market Report benchmarking churn signal effectiveness at comparable mobile operators, followed by strategy advisory diagnosing the predictive gap and building a revised early warning framework.

Impact

Market research confirmed that sentiment-only contact centre analytics models were documented as having limited churn predictive power for customers who churned silently, without escalating through service contact at all. Strategy advisory identified that integrating usage behaviour signals alongside sentiment scores produced meaningfully better churn prediction, especially for the silent churner segment that the sentiment model was missing entirely.

Client Outcome

The client integrated usage analytics into its churn early warning framework alongside sentiment scoring, and early intervention coverage of eventual churners improved substantially within two quarters.

The Situation / Challenge

Contact centre sentiment analytics captures a real signal about customer dissatisfaction, but it captures only the dissatisfaction of customers who express their frustration through service contact. A customer who is quietly unhappy, who has already decided to leave but has not raised a complaint, who has simply stopped engaging with a brand they have mentally already departed from, generates no negative sentiment signal in a contact centre analytics model because they generate no contact centre interaction at all.

The client’s customer success team had been using the sentiment-flagged list as its primary churn intervention target and running a diligent outreach programme. The intervention itself was well-executed.

Without understanding the structural reason the sentiment model was missing a significant portion of eventual churners, the customer success team would continue diligently executing interventions against a list that the actual churn pattern was showing was materially incomplete.

Key Challenges

  • No market benchmarking of sentiment analytics churn prediction effectiveness at comparable mobile operators to understand how typical the client’s prediction gap was.
  • No analysis of what proportion of total churn occurred among customers generating no negative sentiment signal in the contact centre system.
  • A customer success team running a well-designed intervention programme against a signal structurally missing a significant share of actual churners.
  • Post-churn analysis showing a substantial share of churned customers had no prior negative sentiment flag, without a clear explanation of why.
  • No framework for identifying what additional data sources or signals would capture the churning customers the sentiment model was missing.
  • Leadership pressure to improve early warning coverage before the next planning cycle committed to projections built on the existing model.

Contact centre sentiment analytics identifies customers who are unhappy enough to contact the service operation. It does not identify customers who are unhappy enough to leave without saying anything. For telecoms operators where silent churn is a significant proportion of total departure, a churn early warning system built exclusively on contact centre sentiment is structurally incomplete regardless of how accurately it scores the interactions it can see.

Constancy Researchers Solution

Constancy Researchers delivered a market research report establishing how comparable mobile operators had addressed the silent churner gap in contact centre analytics churn prediction, then applied strategy advisory to redesign the client’s early warning framework to capture the churning customers the sentiment model was missing.

Contact Centre Analytics Market Report: Churn Signal Benchmarking
  • Delivered a market research report reviewing contact centre analytics churn prediction effectiveness at comparable mobile telecoms operators.
  • Confirmed that sentiment-only models at comparable operators captured between forty and sixty percent of eventual churners, with the remainder falling into the silent churner category that generated no contact centre interaction and therefore no sentiment signal before departure.
Silent Churner Segment Profiling Analytics
  • Analysed second-purchase rates and twelve-month revenue per customer by acquisition channel.
  • Found paid social channel cohorts showing second-purchase rates significantly below both the historical baseline and same-period cohorts from other channels, concentrating the CLV underperformance on a specific acquisition channel source rather than a general market dynamic.
Usage Behaviour Signal Assessment
  • Assessed the client’s available usage behaviour data, including data consumption trend, voice usage pattern, roaming activity, and digital channel engagement.
  • Confirmed that the client’s existing data infrastructure captured sufficient usage behaviour signals to build a usage-based early warning component that could run alongside the sentiment model, and that combining usage signals with sentiment scoring would cover both the contact-centre-visible and silent churner populations.
Integrated Early Warning Framework Design
  • Designed an integrated churn early warning framework combining the existing sentiment scoring with a usage behaviour decline indicator.
  • Defined the specific usage behaviour thresholds that would constitute a silent churner flag, calibrated against the usage decline patterns identified in the churned customer profiling workstream.
Intervention Segmentation & Commercial Strategy
  • Delivered a revised intervention segmentation distinguishing sentiment-flagged customers from usage-flagged silent churner candidates.

The engagement gave the customer success team a structurally complete early warning framework rather than a technically well-executed programme running against a list that was missing a significant share of its actual target population.

Impact

  • Market benchmarking confirmed sentiment-only models at comparable operators captured between forty and sixty percent of eventual churners.
  • The silent churner profiling confirmed the client’s silent churner share was consistent with the market benchmark.
  • Silent churners showed a characteristic usage behaviour decline pattern in the two months before departure.
  • The client’s existing data infrastructure was confirmed as sufficient to build a usage-based early warning component.
  • An integrated framework combining sentiment scoring with usage decline indicators was designed and implemented.
  • Specific usage behaviour thresholds were calibrated against the churned customer profiling findings.
  • Differentiated outreach messaging was designed for sentiment-flagged and usage-flagged intervention segments.
  • Early intervention coverage of eventual churners improved substantially within two quarters of the integrated framework going live.

Client Outcome

Coverage Improvement

Early intervention coverage of eventual churners improved substantially within two quarters of deploying the integrated sentiment and usage early warning framework.

Silent Churner Captured

Usage behaviour decline indicators began flagging the silent churner segment that the sentiment-only model had been structurally unable to reach.

Root Cause Identified

The structural incompleteness of a sentiment-only early warning system for a market with significant silent churn was confirmed as the primary cause of the prediction gap.

Usage Signals Activated

Existing usage data infrastructure was leveraged to add a usage-based early warning component without requiring new data collection or platform investment.

Intervention Differentiation

Separate outreach strategies were deployed for sentiment-flagged and usage-flagged customers, reflecting their different dissatisfaction profiles.

Benchmark Context

Market research confirmed the prediction gap was consistent with documented sentiment-only model limitations, replacing speculation about programme execution quality.

Methodology Strengthened

The revised churn projection methodology for the next subscriber growth planning cycle incorporated the more complete early warning signal.

Team Confidence

The customer success team gained clarity that their intervention programme had been well-executed against a structurally incomplete signal, rather than poorly executed against an adequate one.

Market Positioning

The operator was repositioned as a churn analytics investor that tests the structural completeness of its early warning signal rather than assuming sentiment coverage is sufficient.

Case Studies

Learn how our success stories power data-driven growth across industries

Speak with an Analyst

    Download TOC