Bayesian Networks for Customer Experience

In the world of Customer Experience, we can all agree that focusing on the customer is the key to future business growth. However, as the channels consumers use to talk about the brand or company proliferate, it can be difficult to decide what to do. Social media, surveys, call center logs, and online reviews are just a handful of feedback options. Not only do you need to make sure you have your eyes and ears open in all of these areas, you also need ways to measure outcomes via business relevant KPIs. Given these challenges, ENGINE wants to touch upon one important aspect of a CX program: understanding which touch-points drive your customer satisfaction.

Paula Sprowl, ENGINE’s SVP of Data Analytics, wrote a few weeks back about your company’s ‘analytical horsepower’. In that piece, she affirmed that analytic horsepower is not just about the techniques you bring to bear, but the expertise of the team to ensure your business question is answered. This knowledge is important when selecting approaches to driver analyses. Customer Experience work can be especially daunting, as many times, respondents experience different touch-points creating lots of missing information.

Missing data can be the bane of any data scientist. It makes the use of multivariate techniques near impossible, especially when as few as 10% of your customers have experienced certain touch-points. Simply imputing data is not the answer as it can provide a skewed perspective on what is important. A client touchpoint, like a customer service call, can have a strong relationship with outcomes for that individual, but may reach a small number of customers.

Many times, when data is missing, we fall back on the old standby of correlation. Correlation is a measure of the linear relationship between two variables. You may have heard the adage ‘correlation not causation?’ Because correlation is a one-to-one measure, you can look at the relationship these touch-points have with key KPIs (e.g. Overall Satisfaction, Net Promoter Score, etc.) The drawback is when you want to understand the relative importance of all touch-points in relation to one another.

If you do leverage correlation, you must take into account the incidence of the touchpoint. We will often create a 3D quad plot or bubble chart where we show importance on the y-axis, performance on the x-axis and bubble size as the incidence of that touchpoint. This still requires a bit of mental gymnastics to decide on where you should focus efforts. This is where Bayesian Networks come in.

Bayesian Networks offer flexibility in the types of data that can be leveraged in multivariate models. It can handle a range of data types that can be hard to accommodate in traditional modeling techniques. The probabilistic nature of Bayesian Networks is even suited for modeling non-linear relationships between attributes within the model.

Bayesian Networks are even capable of handling touch-points that the customer did not experience. By taking advantage of the underlying probabilistic nature of a Bayesian Network, a flag is placed on attributes that we’re not experienced by a respondent. During the modeling process, these respondents are held out when determining the relationships found between attributes within the network. This allows the model to determine how that specific touchpoint plays into customer experience without being weighed down by spurious relationships of those who did not experience that touchpoint.

This dynamic ability to determine the relationships between attributes among those who experienced certain touchpoints allows Bayesian Networks to retain the ability to create multivariate models, even when customers don’t experience every touchpoint. This overcomes the problem of missing data and gives us a more direct understanding of customer touchpoints. We also have the ability to create simulators that can evaluate changes in performance of a touch-point, and what happens if that touch-point reaches more customers.

Bayesian Networks can be a powerful tool to help you identify key customer touch-points that move your business forward. ENGINE has the expertise to build these models and simulate future outcomes. We also have the expertise to bring the right solution to the table for your most pressing business questions. In the end, a tool is only as good as the people leveraging it.

 Written by Kyle Swan, Senior Research Director, and Tyler Dugan, Analyst, at ENGINE Insights.