The optimism of #Purpose & science of #design in forging #digital customer experiences! Opinions my own!

The banking industry regularly mounts campaigns to improve customer value by offering new products to existing customers.

This problem is compounded because of increased capability to send multiple campaigns through several distribution channels over multiple time periods. The combination of alternatives creates a complicated array of possible marketing actions.

Determining the optimal combination of Client x Channel x Offer x Time x …

Optimization inputs

First we need to build scores by product or by product group. The score value depends on several parameters with a function f(x) but the comparison between scores depends on factors such as product type or product development.

Obviously having a score only makes sense after the customer has been in our books for several months – and the number of months may differ by product type. For a card line burn program, we might have 2 product type propensities: a) that the customer takes up a credit line extension and b) corresponding probability of defaulting on payments.

Next we need to assess the score distribution across the customer base for optimization. If for example, 80% of the customer records have only one score allocated, we will effectively only optimize the 20% that are multi-scored. We may need to take a step back and develop a “critical mass” of score attributes per customer record – an effort that need not be daunting if we use predictive modelling tools, or settle for look alike models and approximates.

A third step consists in introducing rules for prioritizing which will allow for the introduction of the company’s strategic priorities:

– priority of a campaign e.g. regulatory and mandatory contacts from the Bank must take precedence over sales campaigns (lifecycle dependencies should also be taken into account)

– priority of scores per client, e.g. revenue vs. response propensities

Beware introducing too many (restrictive) prioritization rules; we may end up eroding the value of mathematical optimization, rendering the problem a simple rank-ordering exercise.

Channel preferences and agnosticism

Different channels serve different purposes – depending on the interaction type required, inbound or outbound. The volume of web interactions for example, should be several magnitudes larger than that of a call center’s daily capacity. For closure of larger ticket items, we may choose to go with a staffed channel, and perhaps a costlier one such as a relationship manager.

We must discern and be cognizant of the different qualities each channel offers e.g. almost without maximal limit for the web channel… but with important maximum constraints for call centers, and the same for relationship managers.

Channel preferences are important optimization inputs. Some customers may display distinct interaction footprints across the various channels; some customers really prefer to deal with us over a single channel. The use of channel hurdle rates e.g. usage behaviour by transaction type and rationale may allow for the evaluation of its suitability as an optimization parameter.

The multi-channel customer with a large set of potential treatments (suites of offers) must be treated differently because they represent the strongest arbitration possibilities – where the effect of optimization is most pronounced. These rare gems of a customer, they have multiple score attributes across several products and channels. For example, propensity X for a take-up of a credit card (..or loan, insurance, bond etc) over a response Y if delivered over a telesales call (..or SMS, eDM, Internet Banking etc).