Share article
The best customer experience is the one the customer never even notices. No complaint, no call to support, no frowning at an email. Just a feeling that everything simply works. So how do companies move from firefighting to spotting the smoke before the first spark appears?
A few years ago I was sitting at a conference when a speaker said something that has stuck with me ever since: “Traditional customer service is essentially a form of apology. You mess something up, the customer calls, you apologise and try to put it right.” It was provocative, but he had a point. Most of what we call customer experience today is really a sophisticated system for reacting to things that have already gone wrong.
I’ll admit it took me a long time to question this. I took it for granted that CX was about reacting well. Quickly, with empathy, competently. But when you stop to think about it, there’s a quiet resignation buried in there: we’re working hard to handle the moment when the customer is already angry, disappointed or confused. It’s a bit like being a brilliant firefighter and congratulating yourself on how many people you’ve pulled out of a burning house. Instead of asking why the house was on fire in the first place.
And that is exactly where predictive CX comes in: anticipating customer problems, needs and behaviour before they happen. This isn’t a fashionable trend or yet another buzzword. It’s a fundamental shift in what it actually means to look after a customer.
From reacting to anticipating: a shift happening right now
When we talk about predictive analytics in CX, in plain English we’re talking about using data, statistical models and machine learning to work out what a customer will do, need or feel before it happens. Instead of asking “what just happened?” we ask “what’s about to happen?”.
The difference between a reactive and a proactive approach is enormous in practice. Forrester research from 2023 found that companies which proactively reach out to customers before a problem arises achieve 20 to 30 per cent higher retention than those who wait for the customer to call. Salesforce’s State of Service study (6th edition, 2024) found that 61 per cent of customers expect proactive communication from companies, but only 33 per cent of companies actually do it systematically. That’s a huge gap. And it’s precisely in that gap that the winners of customer experience are being decided right now.
What fascinates me most about this is the psychological dimension of the whole shift. When a company solves a problem before the customer even notices it, nothing dramatic happens. There’s no wow moment. It’s more of a quiet, understated sense that you’re being looked after. And as research in behavioural economics shows, that very feeling is a far stronger driver of loyalty than any compensation offered after an incident. Roger Dooley puts it neatly in his book Friction (2019): the most valuable customer experience is the one with zero friction, not the one with a great solution to friction.
What predictive CX can actually do
Let’s get concrete, because I find abstract talk about “the power of data” just as boring as you probably do. Predictive analytics in CX today works mainly in four areas, and each has its own character.
Churn prediction is probably the most widespread application. Based on dozens or even hundreds of variables (usage frequency, behavioural changes, contacts with support, survey results, sentiment in messages), the model estimates the probability that a customer will leave in the coming weeks or months. McKinsey’s report The Next Frontier of Customer Engagement (2023) states that companies using advanced predictive churn models can reduce churn by 10 to 15 per cent compared with traditional segmentation approaches.
But here comes the part I love. A prediction on its own is useless. Having a list of at-risk customers is about as helpful as having a weather forecast and still leaving the house without an umbrella. What matters is what you do with the information. And this is where many companies fall down, because the prediction lives inside the analytics team’s silo and never translates into concrete action.
Predicting the next need (next best action, next best offer) is another widely discussed area. Amazon has been doing it for years, Netflix too, and now it’s reaching banks, telecoms and B2B companies. The idea is that based on the behaviour of similar customers, the system can work out what a specific customer will need at a given moment (and just as importantly, what they won’t need). Gartner’s Customer Experience Trends 2024 reports that by 2026, organisations with mature next best action systems will generate 25 per cent higher customer lifetime value than companies relying on traditional segmentation.
Predictive service interventions are the most interesting area for me, because this is where the idea of “solving a problem before the customer feels it” really comes alive. A telecoms operator notices that a customer is starting to lose calls in a certain area and sends a pre-emptive text explaining what’s happening and when it will be fixed. A carmaker spots in its vehicle data that a component is heading for failure and contacts the customer with a service offer before anything breaks. Tesla does this at scale, and Rolls-Royce does it with aircraft engines, just for a very different audience.
Sentiment and escalation prediction is the fourth area. Modern voice and text analysis systems can tell in real time when a customer interaction is heading in the wrong direction, and step in before it derails. Cogito, one of the vendors in this space, has published data showing that real-time emotional analysis in contact centres reduces escalations by an average of 28 per cent and improves first call resolution by 15 per cent.
Where it all gets messy
I have to be honest here. If you read most articles on predictive CX, you’d think it was a miracle recipe you just plug in and your business sprouts wings. Reality is a good deal messier.
The first problem is data quality. A predictive model is only as good as the data it’s trained on. And most companies have data in a pretty sorry state. Fragmented across systems, inconsistent in its definitions, incomplete, often contaminated with errors from manual entry. Bain’s 2024 CX Tools and Trends Report finds that 67 per cent of companies attempting to deploy predictive CX models ran into serious data problems that either delayed the project by months or killed it altogether.
The second problem is organisational. A prediction alone is no use unless you have the ability to act on it. I was talking recently to a CX manager at a larger Czech company who described an absurd situation: their data team built an excellent churn prediction model, but no one in the retention team could actually use it in practice, because the call centre was running on a completely different system and the workflow wasn’t connected. The model worked, but the results just ended up in an Excel sheet that no one ever opened.
The third, and in my view the most underrated problem, is the ethical and psychological dimension. This is where it gets really interesting. When a company knows more about a customer than the customer expects, tension creeps in. The Wharton School published research in 2023 showing that proactive personalisation follows an inverted-U shape: a small dose increases satisfaction, but beyond a certain point it becomes unsettling and counterproductive. When your bank writes to say “we noticed you were looking at mortgages last week, can we help?”, some people will appreciate it and others will be horrified. Both reactions are legitimate.
The line between “helpful anticipation” and “creepy surveillance” is thin and culturally specific. What’s normal in the US is almost illegal in Germany. And what’s legal isn’t automatically wise.
What this means for me, for you, for the company
When I think about how to actually implement predictive CX, I keep coming back to three questions a company has to answer honestly before it starts buying technology.
First: do we have the infrastructure to turn prediction into action? Predicting customer churn is all very well, but if you don’t have a finely tuned retention process that can respond in time and in a relevant way, the prediction is worth nothing. Start by mapping out the whole journey from data signal to a specific intervention with the customer. If there are more than three breaks along the way, no technology will save you.
Second: do we understand what we’re predicting? Models are often black boxes, and managers use them without grasping what they’re built on. That’s dangerous. McKinsey’s State of AI 2024 warns that 40 per cent of companies that have deployed AI in customer-facing processes cannot explain how the model arrives at its outputs. That’s not a detail. That’s a strategic risk, especially in regulated industries and with the new AI Act on the horizon.
Third: how will we communicate with customers about the fact that we’re predicting them? Transparency is critical. The Edelman Trust Barometer 2024 shows that 71 per cent of customers expect companies to give a clear explanation of how they use their data. When a company fails to do that, it loses trust, no matter how accurate the prediction.
The real question
This is one of the things I genuinely love about CX. Predictive analytics isn’t about technology. It’s about philosophy. About how a company understands its relationship with the customer.
Reactive CX is transactional at heart. A problem arises, we solve it, we move on. Predictive CX is relational. It says: we know you, we follow your story, we look after you before you have to ask. That’s a completely different stance towards the customer.
Think of the difference between a friend you have to call when you need help, and a friend who reaches out first because they can see you’re not doing well. Both can be wonderful. But the second relationship runs deeper.
So the real question isn’t whether your company has predictive models. It’s: what role do you want to play in your customers’ lives? And are you willing to build the infrastructure, the culture and the ethical guardrails that role demands?
Because the best customer experience may indeed be the one the customer never notices. But to build it, you have to notice absolutely everything.
