We use cookies to personalize content, to provide social media features and to analyze our traffic. We also share information about your use of our site with our social media, advertising and analytics partners. For information on how to change your cookie settings, please see our Privacy policy. Otherwise, if you agree to our use of cookies, please continue to use our website.

Engagement to Prediction – Detecting Attrition with Time Series

Customer engagement is a word often used but one that I typically think is used in a very loose context. Similar to words like segmentation and optimization, engagement can mean a lot of different things. Merkle’s view of engagement (which I agree with, not only because I work there, but because it makes sense) is that engagement is really about non-monetary interactions with a company. Things like web browsing, open emails, window shopping, participating in an event, or even going onto your phone and checking the balance of your bank account. On the contrary, things like making a purchase or opening an account are monetary interactions or basically where money is exchanged.

We’ve done a lot of work around the idea of customer engagement to monetize engagement activity – so how much is a non-monetary transaction worth? This is achieved by using predictive modeling to assign weights to engagement activities that lead to future customer value (or incremental LTV). The analytics behind this enables us to objectively rank order things like browsing various types of content online versus playing a game on a mobile app. Customer engagement scores are extremely powerful, especially in business models where the financial transaction is not as clean – so think about media companies like ESPN and MTV and how they would assess the value of each customer.

Well, I think we’ve dawned upon the next evolution of customer engagement analytics – behavioral trigger detection. Since we can now estimate how much each engagement activity is per customer, the next logical step is to use forecasting or time series based models to predict the expected baseline of each customer engagement value BEFORE it happens. By building this baseline with predictive analytics, we have effectively created a benchmark for pattern defection. So when a customer deviates from this baseline, either a positive or more likely a negative trigger is created. The most important pattern here is a deviation of declining customer engagement away from the forecast which is a critical leading indictor to attrition or churn. 

The applications of this type of customer engagement scores coupled with time-series predications have implications across all sorts of different industries – the most logical (in my opinion) being CPG/OEM, media/content companies and financial services and banks. Let’s put it into the hands of marketers by telling them what’s going to happen and hopefully they have the creative ingenuity to do something effective to change that outcome. Whoa…this is kind of like messing with the time-space continuum…

I’m not an IBMer, but that doesn’t mean I can’t also help build a smarter planet.

Join the Discussion