Executive summary

Host Organisation : BCG X Client : PowerCo

1. Setting the business situation and background

The Energy Market has experienced a lot change in recent years and there are more options than ever for customers to choose from. A major gas and electricity suplier called powerCO that supplies gas and electricity utility to small and medium sized entreprizes is concerned about their customers leaving for better offers from other energy providers. This business scenario has become a big issue for powerCO . They then reached out to BCG X to help them diagnose the reason why their custormers are churning.

2. Problem After we the data scientist team investigated the business situation we believed that the main concern of our client is that ;PowerCO is experiencing churning and wants to know what are the key reasons behind ?

3. Hypothesis and data As a data scientist team after looking at the business probem through the lense of a data scientist we came to an hypothesis like Customer churning may be caused by customer’s price sensitivity and many other related factors

In order to test whether churn is driven by customer’s price sensitivity we modeled churn probabilities of customers. In order to build our model the Data scientist team required the following data from PowerCO. a)Customer historical data which includes characteristics of each client, for example industry, historical electricity and gas consumption, date joined as customer etc … b)churn data which should indicate if a customer has churned c) historical price data which includes the price the company charges for both electricity and gas at granualr time intervals.

4. Our findings After careful Exploration of the data we found some insigths that we believe may be of interest for PowerCO stakeholders

This graph above is simply showing us the churning status of the company We found that neary 10% of PowerCO’s clients has churned.

This gragh above shows us the churning rates across the type of contract of a customer meaning whether or not they have signed for gas service at PowerCO. What we found is that nearly 10% of non gas customers churned compared to 8% of gas customers. The key take away here is that on average a non gas customer is 2% more likely to churn than a gas customer.

This chart above shows how antiquity of a customer could infuence their decision of leaving. We see that clients with only 1 year of age do not churn whereas after 2 years of being PowerCO’s client roughly 70% of them stayed and this retention rate went back up all the to the 9th year and declined afterwards.

Note on Key metrics In order to make sure that our model perform well and inform business decision for our client the data scientist team managed to build some key metrics to increase the predictive power of the model. This step involved a lot of engeneering skills as well as a deep understanding of the needs of our cient. The metrics that we though are relevant and that improved the predictive power of the model are the following :

1. Difference between off prices in december and preceding January

2. The average prices changes across individuals periods

3. Maximum price change between months and time periods

4. Cient tenure( Which captures the number of month a company has been a client for PowerCO

5. Consumption ,net margin and others.

This table above shows the tenures and the churning rates that correspond to each. For instance 12% of companies with 4 months tenure churned against 8% of companies with 5 months tenure. The key insight from the table is that highest churning rates between ordered tenure is observed from companies with 4 months to 5 month tenure . This suggest that keeping a client till after 4 months is a huge milestone for PowerCO compared to keeping them for longer term. Note on metrics importance : After evaluating our metrics we found that the metrics that drive the prediction power of our model ,interestingly are not price metrics but rather consumption history , net margin and others.

Model summary Within the data as we saw earlier 10% of companies churned . Out of 3286 non churning companies our model accurately predicted 3282. Within the churning companies our model accurately predicted 18 out 366 churning companies . This means that our model is much better at predicting non churning companies than it can predict churning ones.

Recommendations PowerCO stakeholders should then focus more on client RETENTION rather than on client churn because of the predictive power of our model regarding client retention rates.