Customer journey optimization: A real-world example


There are so many ways in which a customer’s journey of experiences can be negatively affected, from forms on websites that are unclear or complicated, to inconsistent or non-relevant interactions over many channels. It is important that these interactions are measured and reduced to maximize customer engagement and increase customer satisfaction over the long run.

You can tackle this challenge from several directions, from A/B testing and Multi-Armed Bandit tests that optimize interactions at specific points in a journey (these are available in SAS 360), to approaches that optimize the full customer journeys over many sequential points in the journey.

Optimizing full analytically-driven customer journeys

This year I was involved in a project for a large retailer and the retailer believed there were a significant number of interactions with customers that had an impact on response rates – i.e., positive (halo) effects and negative (cannibalization) effects. These are difficult to deal with using standard optimization techniques that assume independence of contacts, and therefore full customer journey optimization was used to identify these effects and address this complexity successfully.

As always, the first step was to get the consolidated data, at the individual customer level. We were able to accomplish this because we had good, quality data for the project – customer-level demographic data, and contact history data.

The stages of customer journey optimization

The journey optimization was then carried out in three stages:

Stage 1 –Creating analytically driven customer journeys is an important advancement towards truly effective analytically driven omnichannel marketing. We used decision trees (in SAS Enterprise Miner) on the customer history data to map widely varied journeys that customers were taking. Traditionally, decision trees are used on a wider set of data, but by using just the history, the paths of significant activity that led to purchases were identified.

Stage 2 – Next, these analytically driven journey maps were used as inputs for optimization. For every journey identified by the decision trees, a predictive model was created (using SAS Enterprise Miner) to predict spending, so that for every customer, for every journey, we can predict how much they will spend.

Stage 3 – Finally, the data was optimized using SAS Marketing Optimization, and constraints were applied to establish a final set of scenarios that the retailer agreed would be appropriate to implement.

This illustrates how decision trees can be used to map customer journeys, and these journeys can then be optimized; to replace the disjointed and disconnected results of traditional optimization methods. We are also beginning to use the cutting-edge machine learning technique of deep reinforcement learning to further optimize customer journeys. These techniques will be incorporated into SAS Customer Intelligence solutions to ensure that SAS users can explore this complicated and increasingly important area of customer intelligence.


About Author

Simon Waller

Business Solutions Manager

Simon is trained in Applied Statistics and has successfully employed SAS analytical skills in actuarial, risk and many years of SAS consulting and pre-sales in the CRM and Customer Intelligence fields.

1 Comment

  1. what is the status of this deep reinforcement learning project?
    we are trying to implement same deep reinforcement learning to optimize the clv in 1-1 campaigns. we using all the contact history plus response.
    it wil be easier to get it in production, if we can do this in Sas.

Leave A Reply

Back to Top