I used to live next to a house with a beautiful front garden. But one thing always puzzled me. Whenever they put the sprinkler out, on each rotation only half of the water would actually hit the lawn. The other half of the time, the water fell only on their gravel driveway. Every summer I would wonder why they’d waste all that water when – at best – only half of it was doing the job.
You don’t need to be an analyst to know that a set-up where you definitely waste at least 50% of your resources is a bad idea. And yet, I see the same thing happening over and over again in customer retention marketing campaigns. Brands don’t spend enough time figuring out which of their customers are the proverbial lawn or gravel, and their campaigns end up showering the whole lot with expensive – and ineffective – communications. And, just like my neighbour could have adjusted the rotation on his sprinkler, there’s an analytical technique that can give marketers the insight they need to target their retention campaigns with pinpoint precision. It’s called uplift modelling, and if you aren’t using it yet, you should be.
Uplift modelling is an analytical technique that uses customer data and response data from previous campaigns to identify likely future campaign responders. When applied successfully, it allows marketers to segment audiences based on their propensity to respond to the planned campaign itself, sifting out the effects of other activity such as TV advertising or other marketing activity. Uplift modelling’s popularity comes from its ability to help marketers optimise the impact of their campaigns by identifying the true responders – those who will only do the thing you want them to do if they receive the campaign.
In this article, we’ll discuss the four main reasons why customer retention marketing campaigns are a particularly strong use case for uplift modelling. We’ll go a step further, in fact. Experience tells us that if you’re not using uplift modelling when planning retention campaigns, you’re operating in the dark. And that means:
Uplift modelling allows you to segment your customers into one of four categories based on how likely they are to attrite with or without marketing communications. The result of uplift modelling when planning retention campaigns is a quadrant that looks like this:
Seeing your customers like this - segmented via propensity to attrite with or without retention marketing - delivers four significant advantages.
Uplift modelling starts by analysing response data from previous retention campaigns to work out what actually happened. What you’ll invariably see is that there were some customers that your activity had no effect on. They either left anyway, or stayed anyway, as if your campaign never happened. In uplift modelling, we call these groups the Sure Things and the Lost Causes.
If you’d known from the outset that some customers wouldn’t have been influenced one way or the other, would you have gone to the expense of sending the campaign to them in the first place?
Of course, when working with a new model it’s unwise to put blind faith in its accuracy. Statistical models can never be 100% accurate, and so the prudent approach is to test the new model on a subset of customers, prove it works, and then roll-out the new approach once you’ve built sufficient confidence.
In summary - save the cash! Uplift modelling allows you to identify the customer groups that are unlikely to be influenced by your marketing, allowing you to exclude them from the campaign.
Even worse than having no effect at all, there are some customers that leave because of your retention campaigns. This group is affectionately known in uplift modelling circles as the Sleeping Dogs. As counter-intuitive as it might seem, uplift modelling in retention campaigns sometimes reveals that some customers will take your campaign as a nudge to start shopping around. They were not planning to leave, but your campaign has prompted them to review the products and services they have with you and begin looking at competitors.
When retention campaigns don’t use uplift modelling, it is possible that the campaign results in a ‘net negative’ effect. While you’re actively persuading some customers to stay, you may also have encouraged a greater number of others to leave. Clearly, this gains even more significance if you’re currently segmenting customers based on value alone – ‘high value Sleeping Dogs’ are the potential landmines that retention marketers must always be conscious of. Get it wrong and, like an episode of Netflix’s Dark, your carefully-crafted marketing message will actually create the negative outcome you’re trying to prevent. Your marketing will prompt them to quit.
In summary – keep your customers! Uplift modelling allows you to identify your Sleeping Dogs and exclude them from the campaign right up front. Failing to do this is one of the biggest mistakes marketers make when planning retention campaigns.
As we’ve discussed, uplift modelling works by analysing past retention campaigns and building propensity models to predict the effectiveness of future retention campaigns for different audience segments. Like all propensity models, uplift models can be tricky to build and will never be 100% accurate. But with every campaign you run, provided you’ve set up appropriate control cells, you will gather new results data that your data science team can feed back into the uplift model. This additional data allows your analytics team to validate and refine the model, allowing you to make it increasingly accurate over time.
In summary – the model gets better with age! Used intelligently as part of a sophisticated customer retention programme, uplift modelling can lead to a virtuous circle of optimisation and improvement where you continually learn more about your customers and their likely response to retention campaigns. This level of analysis will also help you refine messaging by providing evidence of the success factors that lead to desirable customer behaviours.
As we’ve seen, when it comes to retention marketing, not all customers are equally good targets for campaigns. Some won’t respond to your marketing at all, some will actively leave because of it. The one remaining group from our quadrant is where marketers should focus the lion’s share of their effort. The Persuadables segment represents customers that have a high probability of attriting without retention marketing, and that can also be influenced to stay with the appropriate marketing campaign.
Ask yourself, if you could identify this at-risk group, how would you treat them differently? Would you offer greater incentives to stay? Would you invest more in marketing to them? Would you further segment them by value to identify your high value and low value Persuadables, and direct even more effort towards retaining the former group? All of these things become questions that marketing teams can debate (and lead to practical approaches that you can test) once you engage the analytics team to develop an uplift model.
In summary – make sure you only water the lawn! Identifying the customers that are both most likely to leave and that are most likely to be convinced to stay sets up a natural bullseye for marketing teams – your most precious area of lawn. The area most in need of watering. Knowing who these customers are helps inform budget allocation and opens the door to testing new approaches. Done well, you'll have a successful testing programme that sees you hold onto an increasing percentage of the customers that you have the best chance of retaining.
Our new data science report, The Ambitious Marketer's Guide to Customer Retention Analytics, discusses three practical ways marketers can use analytics to retain more customers. Download your free copy now.Get the Free Report