McDonald’s apparently has 515 pages of data on one loyalty member. That’s both more – and considerably less – than it sounds like.
There’s a fascinating piece in WIRED recently from Reece Rogers, who decided to ask McDonald’s for all of the data the company had collected on him.
The answer came back at 515 pages. Which is a great headline, and more than a little creepy. But having spent a fair amount of time looking at the kinds of data generated by loyalty programs, my first reaction was slightly different.
What is actually in 515 pages?
Because there are really two different things going on here. The first is the sheer amount of data exhaust that a modern loyalty program can generate. Every transaction carries a bunch of information with it: time, location, products, price, offers, points, channel, payment information and so on. Add app activity, promotions, redemptions and years of transaction history and you can generate an awfully big file without necessarily knowing anything terribly profound about the person on the other end of it.
And because this was produced in response to a privacy request, presumably some portion of those 515 pages is also structure, definitions, disclosures and other material required to explain what McDonald’s has and where it came from. So I’m not convinced that McDonald’s actually has 515 pages of insight on Reece Rogers. But it clearly has some.
And that’s where the story gets much more interesting.
McDonald’s has classified the occasions when Rogers tends to visit. It knows his most frequently purchased products and categories. It has recommendations for what he’s likely to want next. It has calculated his likelihood of attrition. It also predicts that over the next six weeks he’ll visit 2.16 times, spend an average of $13.49 per visit and spend $29.15 in total. That’s pretty sophisticated stuff.
And there’s no particular reason to doubt that McDonald’s is good at it. It has enormous transaction volume, frequent purchases, identifiable customers and a very large loyalty population. That’s a terrific environment for building predictive models.
But there’s something missing.
McDonald’s knows a tremendous amount about Reece Rogers’ relationship with McDonald’s. But that’s not the same thing as knowing a tremendous amount about Reece Rogers. Consider the $29.15.
Let’s assume the model nails it. Six weeks from now Rogers has spent exactly $29.15 at McDonald’s. Is that good? We actually have no idea. If Rogers spends $40 at quick-service restaurants during those six weeks, McDonald’s has captured nearly three quarters of his category spend. If he spends $400, they’ve captured about seven percent.
Same $29.15. Same customer. Same prediction. Completely different economic relationship.
We’ve spent decades getting better at customer data. CRM gave us a record of the relationship. Loyalty gave us identifiable transaction histories. CDPs connected previously disconnected signals. Personalization and predictive analytics gave us increasingly sophisticated ways to decide what somebody might do next. AI is now making all of that considerably more powerful.
But almost all of it is looking in the same direction – from the company out.
We know what you bought from us.
We know how frequently you visit us.
We know which of our offers you respond to.
We know which of our products you’re likely to buy next.
We can make a pretty good guess at how much you’re going to spend with us.
All useful. All increasingly sophisticated. But none of it tells us where we actually sit in the customer’s economic life. Which makes me wonder what would be on page 516.
Not another McDonald’s transaction. Not another app event. Not another propensity score – the rest of the picture. How much does Rogers spend in the category? Where else does he go? What makes him choose one restaurant over another? What percentage of his spending does McDonald’s actually capture? What other programs does he belong to? What value is sitting in those programs? Which incentives actually move his behavior and which simply reward behavior that was going to happen anyway?
That context changes the meaning of almost everything in the first 515 pages.
And I think this is where the distinction between customer data and customer economics becomes important. Most of what we call customer intelligence is still ultimately a measurement of the customer’s activity inside the enterprise. We can make that measurement increasingly granular and increasingly predictive, but we’re still describing one side of the relationship.
Customer economics asks a somewhat different question: where does this company actually sit in the customer’s economic world?
That’s harder.
It also gets us closer to questions that matter well beyond engagement metrics: share of wallet, competitive position, incremental behavior, unrealized opportunity and ultimately the economic value of the customer relationship itself.
This is also why I’m a little skeptical of some of the claims being made about AI suddenly giving companies a complete understanding of their customers. AI can do extraordinary things with the information it’s given. It can find patterns in those 515 pages that no human analyst ever would. But it can’t infer an entire economic universe from data that describes only one part of it. It’s just not the same thing as understanding the customer.
Which brings us back to the apparent absurdity of Rogers’ 515-page file. On one level, it’s a remarkable demonstration of just how much information a modern loyalty operation can accumulate about one person. On another, it’s a reminder of the boundary around all of that information. McDonald’s can apparently predict Rogers’ spending six weeks into the future to the penny. What it can’t necessarily tell us is whether $29.15 is a triumph, a failure or a rounding error.
For that, we’d need page 516.
Photo by Andrii Kordis on Unsplash
