It’s the oldest retention question of all; were they pushed, or did they jump?

Was it failing operational performance that pushed the customer away or was it an irresistible competitive offer? Perhaps it was both or perhaps, something altogether different. The answer to ‘pushed?’ demands an entirely different organizational response than the answer to ‘jumped?’

More fundamentally, the timing of the investigation into what drove defection is also largely dependent on the answer to the question, were they pushed, or did they jump?

They were Pushed

As is always the case, proximity to the event is a hallmark of good research design. In some instances, such as power utilities, the event, for example bill shock, that commenced the path to defection, might be the best part of a year from the actual departure. In practice, the initial issue that disappointed a customer, sensitises them to any subsequent issues. For example, in aviation, a customer who experienced a late flight departure is significantly more likely to complain about the inflight entertainment system. The root cause was the late departure however, research conducted after the customer has defected is just as likely to attribute the defection to the inflight entertainment system.

This is well illustrated by the following exhibit. The likelihood of an Applebee’s guest experiencing any issue at all was 21.9%. However, the likelihood that such guests would experience a second problem was 52.2%, and a third problem, even higher at 62.6%.

Once sensitized, for those individuals, problems become ubiquitous and with it, the noise surrounding the actual root cause is exponentially amplified.

When it comes to operational performance and customer experience, interviewing the customer after they have defected, and hoping to get a strong predictive model of the root cause of the defection pathway, has many impediments to success. Often organizations make the rudimentary misstep of conducting research with folks who have already defected. Those respondents inadvertently mix up their shopping behaviour with their defection behaviour. The resultant findings often disproportionately attribute price as a cause of defection.

The Data Scientist’s Paradigm

A strong predictive model of defection consideration can be built using survey data and a well-designed code frame of hypothesized defection related drivers. That is, applying the well-established approach of developing a hypothesis and testing that hypothesis using small data analytics such as logistic or multiple regression. But it could be better. Given that a client sample of customers is used for the survey, and allowing for privacy regulations, the unique identification number of the respondent should be used to append organizational data. Typically, we have found several important explanatory variables amongst the organizational data that were previously overlooked.

Developing an explanatory model for defection where the dependent variable is ‘seriously considering defection’ is how these dormant and previously unidentified variables can be found.

In a financial services study, we found that of the 12 explanatory variables identified, five were from the survey data and seven were from the organizational data.

Big data in the form of organizational data has an important, complementary role to small data analytics. It is fashionable to start with big data analytics however, we have found this alone does not produce the best explanatory models. It may be helpful to pause for a moment and to return to some first principles; survey data is collected with a specific purpose in mind – to solve a specific issue – in this case, defection whereas, big data is often generated as a consequence of a digital transaction. Post data generation, an independently identified purpose is introduced in the hope that it might address the previously unseen purpose.

I strongly caution folks seeking to identify drivers of defection against being swept up in big data analytics terms such as semi and unsupervised algorithms, machine learning and artificial intelligence, and to remain cognisant of the sometimes-misplaced authority with which these terms are used.

When it comes to operational performance, the most important factors to commence with in identifying the drivers of defection and halting the defection behaviour are:

1. the best dependent variable;
2. a comprehensive list of hypothesized drivers;
3. survey data collection undertaken as near as possible to the service failure;
4. organizational data appended to the survey data and,
5. conduct the analytics applying multivariate analysis

They Jumped

Those pesky competitors sometimes initiate customer defection. Also, an event in the customer’s life can bring about a supplier review. Returning to the power utilities example, moving house is a common example that commences with a supplier review and ends in defection. The defection does not directly relate to customer experience. There is simply a better perceived offer elsewhere.

In the instance where customers are drawn away by a competitive offer, the sample of respondents is ideally taken from recent defectors. Researching these respondents also requires good design work. Often, the most commonly stated reason for jumping is price. The explanation for this is two-fold:

1. The proximity of the interview to the shopping experience in which price is a primary consideration; and
2. The frailty of the respondents’ stated versus inferred response. Inferred responses (uncovered through econometric modelling) are far more likely to get to the underlying behavioral motivations.

When it comes to customers jumping, is core to uncover the hierarchy of value-related drivers of choice – both price and non-price, and to discover, exactly on what basis a competitor was considered better value for money. To do this, the organization needs to consider both price and non-price attributes and, particularly when it comes to price, what part price reputation plays. The dependent variable is no longer ‘seriously considering defection’ but instead, ‘value for money.’

All in All

In instances where the root cause of defection is suspected to be operational failure, defection research should be undertaken with existing customers and the dependent variable should be ‘seriously considering defecting’. Invariably, we have found that it is operational failure that is the major cause of a customer seriously considering defecting. And, in many instances, from a random sample of existing customers, we have found that more than one in five customers have recently seriously considered defecting.

From an organizational perspective, when it comes to operational failure, the best chance of cutting off defection, is to identify exactly what sets the customers off on that defection path. Critically, proximity to that initial event is a primary research design consideration, as is classifying a defector as someone who is just, seriously considering moving on.

In other instances, when a customer was attracted by a better offer, value for money and not operational performance should be at the core of the investigation.

Ken Roberts | Executive Chairman | Forethought