Analytics is a sharp knife, but without a strategic blueprint you might hurt someone

A business problem framing should lead to an analytics problem framing and then you create this killer analytics solution powered by sophisticated machine learning algorithms and everyone is happy correct?

Well… maybe or maybe not.

Let me illustrate with some examples from banking.

1. The maybe part: Fraud Detection: In the world of fraud detection the business problem framing is very focused and narrow, i.e. can you improve detection of fraud when it occurs (or ideally prior to it) and reduce the fraud loss that occurs as a consequence. Very simple business objective. The analytics team tries to build the best fraud classifier using one or multiple machine learning algorithms, including ensemble of models (ex: Random Forest, Artificial Neural Network, Gradient Boosted Models etc.) so as to improve detection and ultimately reduce the fraud $’s detected and net loss post recovery. Easily interpretable business problem and clear path to analytical problem framing.
2. The maybe not part: Enhance Customer Relationship: If a head of retail banking approached you with “Enhance Customer Relationship” as the business objective, now you start to scratch your head. Well how do we decompose this business problem framing into set of sub-objectives that have tangible and focused business outcome desired? So for illustrative purposes, the decomposition might be as follows:

The Conclusion: So what looks like an obvious business objectives, requires 4 levels of drill down to break down the business problem framing into sub-objectives before we can start addressing it.

Now here’s the interesting part: Some of these sub-objectives requires strong introspection on business strategy before we let the analytics guys loose on the problem. For example, you might not want to drive all Millennials to digital channels en-mass. You might want to drive a segment of them to the branch for financial planning support based on income profile, job profile, home ownership etc. The bank needs to do their strategic blueprint homework before they build out a set of analytical objectives for the data science team. Just converting all Millennial interaction to digital might be sub-optimal (through it might seem like the right strategy at a 100,000 feet level).

Similarly, you might want to have a selective business strategy by region, by customer segment, by lifecycle stage of customer for which product to curate to which category of customers before you let the data science team start running their curation models.

Hopefully I have been able to illustrate the point of the need to combine business strategy with sophisticated analytics and technology enablement to really move the dial on the key business objectives. The analytics and strategy teams need to work hand-in-hand. They need to develop hypothesis on business strategy and “test-learn” the success of such strategies post the analytics enabled execution, and keep an open mind to change if the outcome is not as expected. Analytics business leaders can support their stakeholders and clients in building or refining their strategic roadmap prior to embarking on the analytical engagement. I would love to hear pro or con arguments on this topic from the analytics community as well as examples from other industries.