Forecast Financial Modelling – Forecast

Case study

Covid-19 Behavioural Machine Learning Model

Media, Telecommunications

The final model provided key finance decision makers with up to date sports churn propensities for immediate use across a number of potential workstreams, ranging from provisioning as a result of lost revenue, mitigation of risk, and defining appropriate communication strategies.

 Client

Our client is a leading telecommunications and technology company, offering a full range of communications services and competing in all telecommunications markets within the United Kingdom.

Purpose 

The client required a detailed model to identify customers at risk of pausing or cancelling their sports product holding as a result of COVID-19. The client has over 3.5m active sports customers on their book and were looking for a model to predict the potential decay of this base. Forecast delivered a propensity model for the entire customer base within a tight deadline.

 Key deliverables and functionality of the model included:

  • Propensity scores split by the three target outcomes–Do Nothing, Cancel, or Pause their subscriptions.
  • Strategic, regularly maintained and updated data tables allowing for the timely identification of new pause/cancel customers.
  • Analysis of the predictive ability of the modelling suite with evaluation metrics back tested using a short period of actuals.
  • The ability to create and investigate customer cohorts for each of the three target outcomes.
  • Automation of the end-to-end process to identify active sports customers who have not yet paused or cancelled, predict the risk profile of these customers, and update strategic tables with this risk profile.
  • The use of strategic platforms and tools (specifically cloud computing and modelling) to ensure reusability and ongoing consumption.

Approach

Forecast constructed a data pipeline feeding strategic data from Google Big Query onto the GCP AI Platform for further exploratory analysis. Forecast then built and fine tuned a Machine Learning approach that took in the most identifying features to predict propensities on a customer level for the target classes. We delivered a full modelling solution that can be projected on the entire customer base, and re-run frequently to assess the changing circumstances. Forecast delivered this project on a very strict deadline of 3 daysand continues to provide support to the client’s decision-making team.

Outcome

The final model provided key Finance decision makers with up to date sports churn propensities for immediate use across a number ofpotential workstreams, ranging from provisioning as a result of lost revenue, mitigation of risk, and defining appropriate communication strategies. These decision makers helped the client navigate the 

More like this

Case study

Energy, Retail

Demand & Margin Forecast

Case study

Telecommunications

Demand Forecast Model

More case studies