In this series, we’ve looked a lot into predictive analytics and the goal of obtaining a sneak preview of the future. In this installment, I’d like to do the opposite and look back in time. Once a company has climbed the added value ladder of Data → Predictions → Decisions → Actions, making better decisions in all customer channels is not the only benefit. A centralized (customer) decision hub will not just improve the quality and consistency of decisions, it will also record every single one of them. This doesn’t only impact the decision itself, but also the data used to make the decision; the rules and predictive models that represent the actual decision strategy; and last but not least, the business outcome that resulted from that decision (not a trivial process in itself, and a worthy topic for a future discussion). This detailed record of all previous decisions has major business benefits, as we will see.
Recording all this data adds up. Companies that climb this added value ladder will routinely make multiple decisions during every interaction. With interactions in all sort of channels – from online, mobile, and social, to more traditional ones like branch, call center, mail, email, text, etc. – we’re easily looking at many hundreds of millions of recorded decisions, which pretty soon adds up to Big Data.
The value of interaction data
So, is it worth keeping all that detailed history? You bet it is - and for a multitude of reasons. The obvious one, that I will largely ignore here, is for the purpose of (retroactive) reporting. It’s crucial, for instance, to understand what customers bought what and in which channel. What are the demographics of the most valuable customers? This kind of business intelligence (BI) yields insight where previously intuition may have reigned – and evidence always beats assumptions.
But nowadays, keeping a very detailed decision history has other benefits as well. First, as a basis for learning, adaptive – or self-learning – predictive models need detailed data to calibrate their understanding of customer behavior. Modern predictive analytics technology will sieve through big data unsupervised, only reporting on the quality of its conclusions. If the models become reliable enough (once the discovered patterns in customer behavior prove stable), these models can start contributing to better decisions. This process of automated, continuous learning requires detailed recording of all decisions and the result of those decisions.
Big data rules
The third benefit is one that needs to be addressed with a little more detail. Recording not just every single decision, but also the rules and models that were used to make it, creates the opportunity for simulation. This is not the high-level spreadsheet variant but a full rewind of all the detailed customer interactions followed by a fast-forward action of using a modified version of the customer strategy. The result is a detailed “what-if” analysis: what if we had prioritized this higher/lower, priced this higher/lower, accepted a higher/lower risk? Having recorded the full detail of all customer interactions we can use the exact same data, but then apply a modified version of the rules and predictive models (i.e. a revised customer strategy) and look at the delta. This leads to a more empirical basis for prospective changes to the customer strategy.
There are a few things to consider when implementing a capability like this. Data storage, by the way, is not really one of them. Sure, billions of interactions may have to be captured, but it’s offline storage which is cheap, and it’s only the data that is actually used to make decisions. Therefore, data storage requirements go hand-in-hand with better informed (i.e. higher quality) decisions which means the additional returns will easily pay for more, cheap disk space (on-premise or in the cloud).
To be able to simulate, in detail, a customer journey over time and across channels, you can’t glue together a herd of siloed customer strategies. It’s quite alright to have separate channel systems, but the customer decisioning needs to be centralized. The “how” of the customer interaction (i.e. what it looks like and how it’s executed) can be left to a specialized channel application of choice, but the decision as to “what” to talk about needs to be carefully orchestrated. Simulation is only one consideration here – a centralized decision hub also enforces consistency, leverages cross-channel learnings, and in general supports a customer-centric strategy.
A federated, loosely coupled customer decision capability may sound flexible but makes it nearly impossible to reliably replay the customer experience and simulate the effect of changes. To learn and adapt is critical in a dynamic environment, but some changes may not have enough precedents by which to learn. In that case, it’s better to conduct a trial and error approach safely in the past than make the change in a production system and hope for the best. As we all know, hope is not a strategy.
Image Credit: Jessica Tam (Wish) [CC-BY-2.0 (http://creativecommons.org/licenses/by/2.0)],via Wikimedia Commons