Diggin' Deeper, Vol. 2: identifying hidden risks with AI

Artificial Intelligence Machine Learning Natural Language Processing NLP Churn Prediction Churn Prevention

by Johannes Humbert – LinkedIn | Twitter

Satisfied customers are valuable customers. Better are enthusiastic customers who don't even think about leaving. Churn prevention based on artificial intelligence can excite - customers, marketing and sales.

There is a rule in marketing and sales that it costs up to eight times as much to bring back a disappointed customer and four times as much to keep a customer as it costs to win a new one. The reasons why a customer decides to leave are extremely varied and depend on many factors, which also vary from industry to industry. The reasons are also just as varied depending on the phase in the Customer Lifetime Circle. It is therefore no wonder that churn prediction and churn prevention play an important role in the marketing and sales mix.

Once understand why?

Before introducing CRM measures to retain customers, it is better to first understand exactly why an existing customer can become a potential soon-to-be ex-customer. Sounds logical, doesn't it? After all, you want to counteract this in a targeted manner and not just randomly make offers to all customers who are in your portfolio, according to the motto "We'll also pick up the dissatisfied customers with this. Of course, there are also situations where you don't have to think too much about why customers are suddenly running away. Be it a shitstorm in social media, faulty or defective products in test evaluations, scandals in PR, failed pricing policy or competitor offers or or or ... you don't need artificial intelligence for that, really.

Under the radar, on the radar

For the many other reasons for churn that are not so obvious, there are AI solutions to understand them. They dig deep beneath the surface. With transparent algorithms, the reasons for churn are comprehensibly determined from customer data - and continuously optimised. In other words, it is not a one-time process, but an ongoing analysis that is constantly refined. For this purpose, historical and current customer data are used and causal relationships are automatically interpreted - naturally in compliance with the German Data Protection Act (DSGVO).

Reasons made visible, recommendations made

Once it is clear what the possible reasons for churn are, targeted CRM measures can be introduced. In other words, CRM (here: Customer Relationship Management) in its best form, not indiscriminately pouring water on everything - because that hides another danger, more on that in a moment. In the spirit of CRM, these measures are then evaluated and the results validated. In this way, the AI model is constantly learning and can continuously improve further recommendations for action. One can say: the longer this repetitive process is carried out, the deeper the AI model digs into the matter and unearths important insights.

Targeting some and not others or all ...

Wake-up churn - this is what happens when you take a watering can approach. You wake up customers that you should not have woken up. "Oh, that's right, I have this one insurance policy that is actually completely superfluous ..." or "Now that I think about it, maybe I should look around for a new provider ...". An AI model can recognise such customers and thus exclude them from CRM measures. In other words: Yes, it is better not to wake sleeping dogs.

But there is more to it in depth

An AI model learns from the CRM data, so far so good. Churn prediction is, as I said, only the beginning. Churn prevention is the next step. This includes things like causal inference and uplift models. But that would go into too much depth here. Let's just say that A/B testing, for example, digs further into the data and extracts even more knowledge from the ground. And the more you know, the better you can act - instead of just reacting. And the results then look pretty good - for marketing, sales and of course the customers themselves.

Previous Post Next Post