The whole idea of personalization, one-to-one marketing, is the ultimate double-edged sword. And both sides of that blade are very, very sharp.

On one hand, we definitely want the ability to offer different products, services, messages, and so on, to different customers. And we need to build out those capabilities and understand the specific message we want to offer to which customer, through which channel, and at what time.

But the other side of personalization is that too many companies are going too far. And they’re starting to slice the customer base too thinly. They’re really moving—or at least, aspiring to move—towards one-to-one marketing. And that’s going too far. Why? Because it’s basically impossible.

Each individual customer is so unpredictable and random. So no matter how good the data we have, no matter how good our models are or the math behind them, the very best we can do is make approximate statements. Or better yet, to make precise statements about groups of customers.

Now, these segments can be small and very carefully defined. But it’s very hard to build your business one customer at a time. We’ll never achieve one to one, no matter how good the data gets. No matter how good the technology gets. Because customers will always be too random and too unpredictable to count on what any one individual is going to do.

The real art and science is to figure out what are the appropriate groups of customers? And how do we appeal to them? What’s the right message at the right time for them as a whole, instead of one to one?

So what should the savvy marketer do?

Segment—but not the way that a lot of companies do it today. Yes, you should embrace the data, but rather than coming up with completely different segmentation schemes every time you think about a new customer direction or new product or service, build upon previous segmentations that you’ve done.

Take learnings from earlier segmentation exercises and create a core infrastructure that makes different segmentations more comparable to each other. This way, you make each one just an incremental change, instead of having to build everything from scratch.