Customer Segmentation at an Energy Company

Turning transaction data into customer insights

Practice Areas

  • Data Science
  • AI

Business Impacts

  • Six-fold increase in targeting model effectiveness
  • 40% increase in conversions


  • High cost sales channel
  • Low conversion rates
  • Better target potential customers to reduce acquisition costs


  • Python
  • Tensorflow


Our client, a retail gas and electricity provider, operates a call center to facilitate customer requests and to offer complementary services, such as equipment maintenance. To boost sales of these services, the company conducted telesales campaigns.

While a logical channel to reach customers, offering these services through the contact center resulted in lower than expected conversion rates. The client reached out to Starschema for help segmenting its customer base to improve.

Introducing a segment view driven by advanced machine learning, our client was able to improve conversions and lower customer acquisition costs.


Call centers are an effective but costly engagement channel compared to online alternatives, such as email and e-commerce sites. In our client’s case, government regulations restricted the times and number of calls representatives can make, further increasing the opportunity cost of each unsuccessful conversion attempt.

Effectively targeting consumers that are more likely to purchase complementary services can lower customer acquisition costs and increase value added sales volumes. The sales and marketing leaders at our client found that customer acquisition costs were high in relation to lifetime customer value, especially for their value-added service offerings.

Our client used a simple model to segment its customer base for telesales campaigns, but their success rate was only 15%, which was not profitable. They reached out to Starschema to help turn their existing data into customer insights that enabled them to focus on the right customers and contact them with the right offer.


Our data science team worked with the client to identify what existing data they could provide that would be useful in developing a more sophisticated segmentation model. Our client was able to provide our team with:

  • Customer demographics including age, gender, location and mean monthly consumption
  • Customer transaction data consisting of payment and billing characteristics, transaction history and contact-center history (e-mail, phone, personal communications)
  • Past campaign results showing whether the client has been approached for previous campaigns, and the outcome

During the data preparation phase our team tested the data to extract the most meaningful variables for customers. These variables included total consumption, time since online registration, number of interactions with customer support, and more. Armed with the relevant data, our team built a sophisticated model incorporating tree-based methods, specifically a random forest classification algorithm to determine which factors drove conversion success.

The model segmented customers who responded positively to calls from those that responded negatively. Importantly, not only did the model have a very strong predictive potential but was also capable of explaining why an individual customer would be allocated to a particular likelihood of conversion category.


With the segmentation model designed, our data scientists validated it in simulations based on prior campaign data. The results produced a valuable insight that our client had two customer segments, in other words a bi-modal distribution, who would likely purchase complementary services when approached by the contact center.

The new segmentation model yielded a six-fold increase in the number of customers purchasing complementary services during simulations. When the sales and marketing team at the customer implemented the model, it helped the client increase its campaign success rate by 40%, significantly reducing customer acquisition costs and increasing profitability.

Five AI Trends in Healthcare to Watch in 2022

AI is transforming healthcare, unlocking unprecedented opportunities for enabling easier discovery of deeper insights that drive innovation – but the available technologies can very greatly in their maturity and domain-specific applicability. This white paper introduces five proven, future-resilient solutions to challenges that healthcare providers face today.

Understanding Topic Modeling and Planning Its Implementation

Topic modeling enables the analysis of text-based data to leverage insights that are difficult to extract and understand to help you optimize costs, improve operations and drive innovation. Read this white paper to understand the fundamentals of topic modeling and learn how to get started implementing it.

Improving Predictive Accuracy

A Fortune 50 healthcare conglomerate wanted to improve the predictive model of a mobile application that helps those suffering from allergic rhinitis to make informed decisions about potential exposure to outdoor allergens by predicting symptom severity.

Innovative Medical R&D Insights Using Machine Learning with Gedeon Richter

Gedeon Richter, a multinational pharmaceutical and biotechnology company, leveraged Starschema's data science expertise to jointly develop an ML-based methodology to quantify the properties of the mitochondrial network within neurons to enable more effective analysis of medications for various neurological diseases.