As I often emphasise to our clients:
We can build the fanciest, sexiest models in the world using the shiniest, cleanest data set imaginable, but that’s all completely meaningless until some real change happens at the end of the process that means the customer does something different, preferably in a positive way.
So often we find the business puts its own barriers in the way of making sure this success can happen.
One example, not so long ago, involved a credit card company facing typically high churn customer issues.
The models they had been using for years weren’t cutting it and were too simplistic to provide any degree of prediction or probability about a potentially churning customer's likely behaviour or reasons for churn.
So much so that by the time their models had identified a churn target, the customer had usually already left.
They needed a better way to identify earlier if a customer looked like they were at risk, as well as to connect the dots as to the likely reasons why – e.g., had the client just slapped on some new fees onto the product without telling the customer, which happened more often than it didn’t.
We stepped in and built some smart, easy-to-use AI models that could flag early signs of behaviour change that might indicate an elevated risk of churn, as well as connecting the experiential triggers, such as communications, fee increases, billing issues, complaints, etc., that were relevant to each customer
Essentially, this new solution increased the runway for the business by up to six months, providing ample time to initiate a series of remedial customer journeys that we designed for the client.
In testing, these models smashed the old approach and promised a transformational opportunity for the business to not only retain customers but, in doing so, dramatically course-correct on the huge acquisition spend it was laying out to keep filling the bucket with new customers, only to have them leave so soon.
It was during implementation where we hit a snag no one was expecting. One senior director of the business, who had been there forever, was really nervous about using the models.
In their words, “it’s better we let sleeping dogs lie”. By that, they meant they’d rather not reach out to the customer or do anything to alert the customer that something might be wrong as it might encourage them to leave.
Even though the data, which was rigorously tested over and over, made a clear case that each customer was at risk, and the customers that weren’t being given the remedial treatment carried on leaving as expected.
So what happened?
Absolutely nothing in truth.
The models were kept running, the business was alerted weekly to which customers were at risk and the same treatment as before was applied.
This meant that basically, a small team of call agents reached out to the highest risk customers (highest also meant closest to leaving) to check if everything was okay.
Surprise – the customers had by and large already made the decision to leave, just as before.
You can only imagine everyone’s frustration, but it made for some interesting reflections on lessons learned and in particular about not making any assumptions about how you can expect to make change.
This director had years of experience in the sector, so in many ways almost certainly had some pretty strong reasons for their concerns, no matter how irrational it felt to all of us on the ground.
Lessons Learned:
Senior Management Buy-In: It’s crucial for top management to grasp the basics of the data’s implications. They don’t need to dive deep into analytics, but they must understand how and why a model works.
Actionable Change: Just having a data model isn't enough. Companies need to act on the insights, tackling underlying problems like pricing, fees, or customer experience to make any real difference. These need to be planned upfront alongside the design of the data solution or models.
Our experience highlighted a simple truth: data and models are only as good as the actions they prompt. Identifying problems without changing the approach is a recipe for failure.
Real success comes when everyone, especially those dealing directly with customers, changes their behaviour based on data-driven insights.
Follow for more