Why Churn Prediction ain't Churn Prevention

By Gope Walker

Posted on September 14th 2021

As a self-confessed stats geek and CEO of an analytical consultancy, I should probably be the first person to try to sell you the benefits of churn prediction models.

Just so we’re all clear, the churn prediction I’m talking about isn’t the likelihood of milk turning into butter - that will perhaps be in a future blog. The churn prediction I’m talking about is the one about customers leaving your business. The Internet describes it as, ‘ churn prediction modelling techniques attempt to understand the precise customer behaviours and attributes which signal the risk and timing of customer churn.'

Basically, it’s about predicting customers leaving before they’ve left. The theory is that if you can identify them before they leave, you can act on that information, tell them how much you love them, and then they’ll stay. If you think you’ve heard of a similar concept before, you have - it’s a central theme in Steven Spielberg’s 2002 film, Minority Report (based on the 1956 short story " The Minority Report" by Philip K. Dick)

The film stars Mr runs-around-a-lot Tom Cruise as a police officer where PreCrime, a specialized police department, apprehends criminals based on foreknowledge provided by three psychics called " precogs".

I’ll not spoil it for you, but things didn’t go according to plan as sometimes predictions can go wrong.

From a statistical point of view, predictive modelling can be split into 4 different sections, which are laid out in the chart below with Minority Report as an example:

Let's look at each of these individually and explain what they mean…

This makes Tom Cruise and statistical modellers happy. Tom gets to lock away bad guys before they get to commit a crime and statistical modellers get to prove that their models are accurate by correctly predicting people who are about to leave. Tom heads to the pub with his friends at PreCrime and the data modeller goes home and plays with their Lego Millennium Falcon.

Happy bunny Mr C

Ambivalent TC

This is the picture you get when you type ‘Tom Cruise ambivalent’ in Google, which describes how he and statistical modellers feel about ‘True Positives’. In Minority Report, the precogs do not care about innocent people being innocent. In the same ilk, statistical modellers or retention managers don’t care about existing customers who are staying (even if the Marketing department do).

Tom is quite mournful with Type I Errors. This is where he finds a person guilty who is actually innocent. There have been a lot of famous Type I Errors throughout history, so I’ll just leave this here… List of exonerated death row inmates

From a Statistical Modeller's point of view, these are customers who are incorrectly identified as leaving. This is an unfortunate truth of model building - that no model will ever be perfect. For a company, this means you may be wasting valuable resources on trying to retain customers which were anyway going to stay, which can be costly. This could be improved through creating a better model or gathering more data on your customers.

Pensive Tommy

Angry Thomas

Tom isn’t a happy bunny, statistical modeller isn’t a happy bunny and the retention manager DEFINITELY isn’t a happy bunny. Tom knows a guilty person has got away, the statistical modeller’s model didn’t predict someone leaving and the retention manager’s Customer Base has gone down as they’ve lost a customer. Unhappy times all round.

Obviously, Type I and Type II errors are bad and one solution to reduce them is through small adjustments to the model, or gathering more data on the customers. However, even with a 100% accurate churn prediction model, that isn’t going to keep customers who were intending to leave. Ultimately, the overall business goal is churn prevention (i.e. preventing customers from leaving), not just creating a great statistical model after all.

Let’s talk churn prevention

If you have a super-duper Churn Prediction model that will correctly predict if a customer will leave or not, that is only half the battle. If your Statistical Modeller creates something that lets you know a customer is definitely going to leave in the next 28 days, what do you do?

  1. Call them and beg them to stay with a new contract at a better rate?
  2. Give them 3 months free or half price to lock them in for a bit longer?
  3. Give them free vouchers from Just Eat or Amazon?
  4. Send them some flowers asking them that you’ll change and try harder from now on?

I could go on, but can you see the point? There are lots of different churn prevention tools, but which is the right one, and more specifically, how am I supposed to know?

Gamble and Huff who wrote a song in the 1950s called ‘If you don’t know me by now' If You Don't Know Me By Now - Harold Melvin. You may think it was written about a relationship between a man and a women, but I choose to believe it was about a disgruntled customer and a service contract provider.

Ultimately, if you are worried about your customers leaving and you don’t know what to do about it, it’s a bit too late I’m afraid. Successful businesses aim to know their customers from day one, and nurture that relationship from the start. That relationship should be treated like any other relationship, whether it’s with a partner, friend, pet or plant, i.e. learn about what they like and what they don’t like and, whatever you do, don’t take them for granted.

The biggest predictor I’ve consistently seen in churn prediction model building is users' answers to the question ‘Why are you joining us?’. From my 22 years of working with Churn Prediction models, this is THE Golden Question to ask during the onboarding process. If you know why people are joining, you’ll be able to determine why they may want to leave in the future and what you can do about it.

Then, after a customer has joined, learning as much as possible about them will help you shape their relationship with you. If you do everything right, they won’t want to leave and hence Churn Prediction isn’t required. The big caveat here is ‘do everything right’, this is a difficult task but understanding your customers through usage segmentation, demographic profiling and relevant & timely CRM Customer relationship management is key.

Ultimately, if you are worried about your customers leaving and you don’t know what to do about it, it may be too late for your existing customers - but never too late to start for your new customers.

However, to have effective churn prevention you need to have robust and accurate data at the heart of your business….and you know who can help you create this, don’t you? (Hint: It’s not Tom Cruise!)