UNCOMMON AND UNSEEN CHALLENGES POSED BY THE CURRENT ECONOMIC ENVIRONMENT REQUIRE FAST, TARGETED, AND DATA-DRIVEN INTERVENTIONS TO THE BUSINESS EXECUTION IN THE FRONTLINE. MARGIN, CONVERSION, AND VOLUME ARE ALL DRIVEN BY PRICE. SO HOW CAN ORGANIZATIONS RESOLVE MULTIPLE ISSUES WITH ONE INSTRUMENT?
Dozens of studies (if not personal experience) can confirm that acquiring a new customer is far more expensive (5 – 25 times) and time-consuming than retaining an existing one. In the economic climate of today, it is wise, if not critical, to work extra hard on customer retention or prevention of churn.
Unsurprisingly, Data Science/AI with its problem-solving and predictive qualities once again, comes to the rescue. While CHURN PREDICTION is one of the older applications of AI/Analytics, it is still relevant today. Companies in various industries are employing different tactics to make their churn prevention programs more advanced and sophisticated.
While Churn is typically defined as the number/percentage of customers that end their relationship within a given period, we also look at volume lost at a customer (resulting in a share of wallet decline) as churn. Since it is rare for B2B companies to lose a customer in its entirety all of a sudden, assessing churn at a customer-product level proves more insightful and actionable.
HOW DOES THIS WORK?
STEP 1: UNDERSTANDING DRIVERS OF CUSTOMER CHURN & DECLINING SoW
As a starting point, look back in time to understand why customers churned or reduced their volumes. Historical price and volume trends can indicate if price hikes were the cause of churn. However, customers and volumes could also have churned because of non-price factors such as poor complaint handling, recurring supply disruptions, insufficient sales contact, regulatory reasons and so on. To find the impact of non-price factors on customer churn, data from customer surveys (eg; NPS), customer service /support data (eg: tickets, resolution, SLA adherence ), supply delivery reports, CRM visit/phone reports/logs need to be used. Depending on the data source and type, different techniques such as PCA, boosting models, association rule learning can be used to find correlations between churn and its attributes.
At a chemical manufacturer, we identified that customers that had logged more than 2 complaints that were not resolved within 14 to 17 days, were likely to take their business elsewhere with the next 2 contract terms.
STEP 2: BUILDING A MODEL TO PREDICT CHURN LIKELIHOOD PER CUSTOMER-PRODUCT COMBINATION
Once key attributes/drivers of Churn/SoW Decline are identified, a model is then built to predict which customer-product combinations are at risk of churn. A predictive churn model can be built using simulation models or SVM or neural networks that gives a churn probability per customer-product combination and provides the likeliest drivers for churn.
3/4th of the organizations we have worked with, stop at this point. They make these churn prediction insights available to sales account managers who are then left to their own devices to plan an appropriate response to this churn risk. More often than not, the response of sales is a price concession or discount campaign. And unsurprisingly, it works – results show that more than 45% of high churn risk clients can be retained with price concessions. But the pertinent question still is – Was the price campaign worth it? Was the customer worth retaining?
STEP 3: ASSESS RETENTION LIKELYHOOD OF CUSTOMERS
To take things to the next level, customers that are flagged as medium – high risk should be assessed to see if they can be retained when targeted with a campaign. For customers that are likely to churn regardless of a campaign or those are likely to be further triggered to churn if targeted – it is best to not take any action. For those that are likely to be persuaded to stay & hold their volume, it is worthwhile to offer a price concession. The determination of who should be left alone v/s who can be persuaded is based on a technique called uplift modelling.
The next logical question, for the customers that can be persuaded, is: What should the extent of the price concession or campaign be – or what price can they be retained at?
STEP 4: WILLINGNESS – TO – PAY OR RETENTION PRICING
Determining the best price to retain a customer – is done using willingness to pay pricing. It is a discount factor computed based on historical price-volume analysis of a customer with his/her peers as well as determinants such as historical negotiation behaviour, price aggressiveness, price sensitivity etc. Providing sales with concrete recommendation on who to target and how much of a discount/concession needs to be given improves the chances of retention from ~45% to ~70% and reduces the effort & resources spent on the non-persuadables and excessive discounting.
With pressure on commercial teams on the rise, we are seeing an an uptick in the number of companies using advanced AI and ML techniques to enrich their churn prevention initiatives actionable pricing insights for sales.