Context and objectives

For several years, a leading Belgian telecom operator had been dealing with a structural customer retention problem. With a churn rate above 10%, the estimated annual impact ran into the hundreds of millions of euros.

Retention in telecom is complicated by a mix of factors:

• Some are external, like the rise of promotional bundles combining mobile, internet, and TV across competing operators.

• Others are internal, from core service quality and pricing competitiveness to customer support efficiency in resolving issues.

On top of this, siloed data and inconsistent data quality made it difficult to pinpoint what was actually driving churn.

The operator turned to Agilytic to:

• Identify churn drivers

• Target at-risk customers

• Build a reliable churn prediction capability

Approach

1. Defining the problem scope

We ran a workshop with the operator's teams to define the relevant KPIs, organized around three sources of churn:

• Administrative issues such as contract errors, untimely pricing formula changes, and price increases

• Invoicing problems including billing errors, overconsumption, and charges not covered by the fixed price

• Technical factors like network performance degradation and service interruptions

This step aligned all stakeholders on which signals were worth tracking and what "churn" meant across departments.

2. Building a unified data foundation

We consolidated customer data from the CRM with billing and network data into a single, clean data warehouse. Given the fragmented state of the operator's existing data infrastructure, this unified foundation was a prerequisite for any meaningful analysis.

3. Developing and deploying the churn prediction model

With clean data in place, we built a retention model to:

• Identify the key churn drivers

• Score each customer's churn propensity

The model could forecast churn up to three months before it occurred, giving the operator enough lead time to plan and execute retention campaigns.

Each customer received a score along with two supporting arguments explaining their churn probability. We pushed these scores directly into the operator's CRM so front-line support staff had all relevant context during both inbound and outbound calls.

4. Refinement and handover

To keep predictions current, we developed a nightly automated update that delivered fresh scores via CSV directly into the CRM.

The project concluded with a comprehensive handover, including training sessions and detailed documentation for the operator's teams.

Results

The project delivered impact on several fronts:

• A clearer understanding of the root causes of customer churn

• Actionable customer segments, each assigned personalized next-best retention initiatives

• A decrease in annual customer churn, equivalent to roughly 1% of the total customer portfolio

The operator now runs churn prediction as part of its daily operations, with retention teams using fresh scores for every customer interaction.