by Richard Lewis, Director of Analysis
In my line of work, I find that churn models are a very common occurrence. From an agency perspective, in the traditional world of marketing analysis, there is no doubt that churn models are seen as a commodity. There is no end to the line of analytical consultancies pitching to develop churn models for almost every type of industry and sector, often without a true description, or even understanding, of how the client organisation is going to realise value from this basic analytical deliverable.
I have sometimes the thought that if Model Citizens were to offer churn modelling as a primary proposition in our toolbox, then we could win a significant amount of business.
It’s an appealing thought. Most of the marketers who have retention KPIs (Key Performance Indicators) associated with their objectives view the existence of a churn model as a key building block of their strategy.
Agencies have typically responded to this perceived need and most have added the proposition to their own box of tricks without further thought. Each agency promises a ‘better’ model to their prospective clients than their competitors (based on an often unfounded self-belief). The abundance of these un-varied propositions serves to both confuse the marketplace and reinforces the necessity to the marketer of having one.
So, should Model Citizens, as a marketing analysis consultancy, join the ‘me too’ bandwagon of churn modellers? In all the confusion, we might win the incremental business, we might not. But what we most certainly would not win is the trusted relationships that we currently have the luxury of enjoying with our clients. Why is this?
The truth of the matter is, quite literally, a churn model is no good!
For Model Citizens, any issue with churn modelling is not related to the technical, data, mathematical or even statistical skills required. Data mining and predictive analytics is what we do! This issue is one of proven business benefit or, more specifically, a lack of it.
In order to understand why, it is necessary to standardise a definition of a churn model. A churn model delivers, at a customer level, a score that reflects the customer’s likelihood to churn, ideally within a fixed time period in the future. This score is developed by taking customers who have previously churned in the past, and comparing them to customers who could have churned in the same period in the past but didn’t. The identified differences between the two groups are overlaid on existing and currently live customers. In this way, groups of currently live customers can be split into two groups. Those who look more like the churned group are deemed to have high likelihood to churn in the next time period. Those who look like the customers that did not churn are deemed to have a low likelihood to churn within the next time period.
Outwardly this seems eminently sensible and a bit of god send to the marketer tasked with reducing churn. An ability to identify sub sets of the customer base that are more likely to churn than others – what more could they ask for? Trouble is that the marketer will soon find out that this insight will prove fruitless.
The simple reason for this is that, no matter how accurately the model separates out future churners from future non churners, the model does not tell the marketer what, if anything, to do about it. Every analyst who’s worth their salt knows that in an analytical project it is essential to identify the business objectives before identifying the data driven solution. Every sales person in an analytical agency will say that their analysts do this too. But every time a churn model is identified as the solution is a classic case of ignoring or even missing out altogether this crucial business understanding phase.
To get to the bottom of this, let’s explore the real business objectives. What the marketer is interested in achieving is a reduction in churn. The plan is to achieve this through the targeting (hence need for a predictive model) of a retention proposition to specific customer groups for which the promotion demonstrates the highest profitable reduction in churn.
Note that the actual churn rate is not mentioned, neither is churn likelihood or propensity, or any language typically deployed in the selling of a Churn Model. The only customer measure of interest, which is stated in the business objective, is the change in the level of churn from churn without promotion to the new churn with promotion.
Underlying churn rate is of no importance whatsoever. What is importance is the difference that can be made to churn regardless of whether it is achieved from customer segments with high or even low churn likelihood.
Looking at the business objective, the real analytical response and solution would be to model the customers with a high likelihood to respond positively to the retentions proposition. That is to say, identify the differences between those that would have churned, but as a direct result of the retentions proposition now won’t. True it doesn’t sound as ‘whizzy’ or even as ‘simple’ as a churn model, but it is the correct solution in response to the posed business objective.
So why aren’t marketers being offered the correct solution to the problem. The answer is simple:
· It requires a new type of modelling called Uplift Modelling that isn’t widely known by analysts. This means fewer agencies are offering the solution and visibility of the technique in client side marketing roles is suppressed.
· A churn model is a simple concept for un-analytical salesmen to sell. More complex analytical solutions such as Uplift Modelling require analytical people to explain and sell.
· Often marketers, unaware of the real alternative specifically ask for a churn model.
In my own view, not knowing the technicalities of Uplift Modelling is no excuse for miss-selling churn models as a full retentions solution. Uplift modelling is relatively new – but it is also a tested and proven method. It most certainly is a better use of precious resources in tough times.
Ask yourself a question. Is the churn model being considered because it is being pushed on you by external forces, or is it truly a direct response to a clearly stated business need that you have imparted?
If the former, or if you want to know more about Uplift Modelling or Model Citizens Retentions solutions and what they could actually achieve for your business, please email churn@model-citizens.biz