2 min read

Predictive churn algorithm helps plastics manufacturer prioritise CRM engagement

Profile

A major multinational plastics packaging manufacturer serving customers in a range of sectors including pharmaceuticals, food packaging, labels, and others.

 

Situation

Our client wanted to create a process for managing churn by anticipating risk at the customer level and help its sales team pre-empt likely customer churn. The business had a diverse portfolio of customers, each with different data sources and behavioural patterns whose behaviour they needed to pre-empt and manage. The business needed a model driven by “common sense” that was credible, easily explainable to the sales team, led to clearly demonstrated results. The tool needed to integrate with SAP and Salesforce and an automated churn dashboard in PowerBI.

 

Action

We hypothesised all the potential drivers and indicators of churn within the business and then applied statistical techniques to analyse customer-level order histories, behavioural patterns, and periodicities to develop a ‘consistency index’ of order regularity that was used to anticipate future lower customer order volume.

We found that churn in the business was non-contractual and non-trivial to anticipate, therefore statistical techniques were deployed to quantify the level of risk associated with particular types of order behaviour. Based on this analysis, we built a statistical model that aimed to detect significant changes to order volume on an ongoing basis. Sometimes many of these changes were almost imperceptibly small nevertheless significantly correlated with a high probability of future churn.

We then weighted risk levels based on the customers’ behavioural signals, such as recent service experiences. We consolidated the multi-variate model to calculate risk scores for each individual customer based on recent performance.

Whilst the business was already fully aware of risks associated with the largest customers, our model focussed on the mid-tier of customers who otherwise would have been underserved due to limited time available to the sales team.

Once we developed the model, we worked with the client through a second sprint of work to “productionise” the codebase (fully automate within their existing systems), connect it to their business warehouse (SAP) and automatically feed the outputs to their existing CRM environment (Salesforce) and BI tools (PowerBI) to be used by sales teams across all regions.

 

Tools and techniques used in this work

  • Bespoke Python scripts and libraries

  • Data Engineering to build data pipelines and productionise code

  • Advanced SQL

  • Statistical Modelling

  • Automated Power BI Dashboards (including advanced permissions)

  • Salesforce CRM integration

 

 

Impact

The model was able to effectively identify customers who were at higher risk of churn based on their recent behaviour. This visibility provided additional support to the sales team in prioritising customer outreach and understanding risk, allowing them to streamline their workload and manage churn.

Without this model the sales team would have had capacity to engage with “mid-tier” customers likely to churn. As a result, the sales team now have a real-time understanding of customer behaviour and propensity to churn (including indicators of likely churn in order to inform their outreach strategy).

 

So what?

QuantSpark’s creative approach to behavioural modelling was able to provide predictive support to a business with limited prior understanding of customer churn behaviour. Working collaboratively with senior management and the sales teams, we were able to build trust and confidence in analytics and thereby build momentum for digital transformation across the entire organisation. This new approach to analysing sales and customer outreach allowed the sales teams to use data to optimise their approach and improve retention.