When to ignore what your data is telling you

How much latitude should you give front line customer service representatives (CSRs) when implementing customer relationship management strategies? In particular, when should they ignore action recommendations that have been developed from detailed customer knowledge and extensive data analysis?...

How much latitude should you give front line customer service representatives (CSRs) when implementing customer relationship management strategies? In particular, when should they ignore action recommendations that have been developed from detailed customer knowledge and extensive data analysis?

Many organizations are spending substantial sums to consolidate customer data, turn that data into marketing intelligence and then design products and services that meet the varied needs of different customer groups.

An important aspect of that effort is a series of data mining exercises that predict which products each customer would most likely be interested in or which ones they might be showing signs of cancelling. Armed with the resultant probability scores, marketing programs can be directed toward customers who represent the best potential.

In many cases, though, this is not just about centralized direct mail or telemarketing programs. Banks, retailers and telcos are working hard to ensure that the same customer intelligence is available at all possible points of contact so that customer service reps can take the same informed action in the branch, retail outlet or call centre.

How this knowledge is presented to the front line – and the role that CSRs take in implementing it – is one of the make-or-break issues of realizing a CRM strategy.

One common practice is to avoid providing CSRs with a lot of data that they must interpret themselves. Rather than providing cross-sell scores across a range of products, for example, an organization might populate a bank branch computer terminal or call centre screen with the top three products that a given customer is predicted to be interested in, along with supporting scripts outlining features and benefits. CSRs can then focus on cross-sell or retention efforts instead of getting bogged down wading through screen after screen of ‘raw’ data.

This approach has a lot going for it, not least of which are the inherent operational efficiencies and the ensured consistency of the message. But don’t forget: the models that yield product propensity scores are only as good as the raw information they’re built from, and not everything that is relevant to a customer’s purchase decision is necessarily captured in your data warehouse.

Very often, the front line rep knows something about your customer that the data warehouse doesn’t. A bank teller may know, for example, that a given individual is not only a retail customer but also owns a small business (a fact often not captured in the warehouse because of the difficulties of matching consumer and business databases). Or they may know that the customer in question is influential in their community and therefore represents potential referral value well beyond their personal sales potential.

In these circumstances, the front line CSR needs to be able to override the recommendations presented to them because they know something that the system doesn’t. Models are just tools, and any organization that employs them needs to know when to follow their recommendation and when not to. While the system’s recommendations should be followed in the majority of cases, there must be enough flexibility to allow for changes when conditions warrant. The challenge is to design the rules and procedures for overriding the system so that chaos doesn’t ensue.

The first rule is to ensure that the model is only overridden on the basis of new and relevant information. That’s easier said than done. Do you want each CSR to know every field in the warehouse so that they can recognize when data is ‘new’? And what new data is relevant?

One solution is to define the categories of information that are legitimate reasons for override. One example would be the common practice of assigning VIP status to politicians, actors, journalists and other public figures – a task that sometimes requires input from the CSR. Another example would be an area where you know there is a data gap, such as the small business customer mentioned above. Yet another category might be socio-demographic data such as the birth of a new child or a job promotion. In each case, the objective is to clarify for the CSR what types of new data they are allowed to use to justify overrides.

Second, it’s important to record that the system’s recommendation was indeed overruled as well as the reason for the override, preferably in machine-readable format. Only by knowing which customers were not treated as the system suggested can you ensure a clean reading of the database and determine whether or not the overrides are effective. And since it’s very difficult to work with free-format input, a series of override codes should be developed that can be fed back to the data warehouse for further analysis.

Third, there may be degrees of latitude – areas where the CSR can act on his or her own and others where a supervisor’s approval is required before overrides can be made. This is quite common when granting or managing credit products but makes equal sense on the marketing side.

It is also important to audit override decisions. One of the objectives of systematizing CSR actions is to ensure that the customers who get the best deal are the ones that are most valuable – not just the ones who yell the loudest. Many organizations employ model recommendations precisely because they want to ensure that price breaks are being given on a consistent and appropriate basis. Introducing some flexibility back into the process resurrects the danger that overrides will be abused, a risk that can be kept in check by systematic audits.

Finally, an added complication is that all of this must conform to privacy legislation. Adding a new piece of data to the database in order to generate an override may well require the customer’s consent, not to mention the need to comply with the other aspects of Canada’s privacy law. Proper scripting will be required to ensure that your privacy procedures are being followed at all times.

The power of data mining is well documented. But no tool, no matter how valuable, should be used blindly. Providing your front line staff with tools that enhance efficiency, while at the same time granting some latitude in their use, will often prove to be the optimal approach.

Colin Tener is president of Tener Solutions Group, a customer relationship management consultancy based in Toronto. He can be reached at (416) 585-2900 or by e-mail at tenerc@tenersolutions.com.