offer a meal deal you can't refuse

seeing like a supermarket

I am beginning to get stuck in to “The Ordinal Society” by Kieran Healy and Marion Fourcade.  I’ll try and write some proper thoughts when I’ve finished it, but at about 20% in according to my Kindle, I am already a big fan, and am having Thoughts about its general subject matter.

The big theme of the book[1] is the way in which the collection and use of personal data shape society; there’s a feedback loop between what we choose, the options we’re offered and the kinds of systems which get the chance to offer us options.  As a detailed description of how seemingly-neutral algorithmic decision making systems are embedded in much less neutral, much less obviously algorithmic structures of power, it’s the sort of thing that was always going to be catnip to me, and so far it’s living up to the promise.

One thing it got me thinking about, though, is one of the great examples of Big Data that people always used to go on about back in 20-dickety-teen, before the latest AI revolution but when Data Science was the thing of the future.  As I wrote in “The Cynic’s Guide to Fintech”

The real evangelists of Big Data thinking in the fintech industry seem to think that providing the actual financial services is a tiresome encumbrance, a loss leader that you have to provide in order to get access to the true mother lode of value — a load of Big Data. This is, to be frank, incomprehensible. Transactions data just isn’t that valuable, and in order to get lots of it, you need to have a very big operation indeed.

Look at it this way — the single most developed transaction data analytics business in the world is Dunnhumby[2]. They have more than a decade of experience, and they work with the Tesco Clubcard dataset, which accounts for a material proportion of all the supermarket transactions in the UK over the last ten years. And they have managed to segment Tesco’s customer base into a grand total of … eight economically meaningful groupings. Along with that, they have come up with insights like “people who buy carrots also often buy cucumbers”. Dunnhumby is currently being shopped around for sale at a valuation of about £2bn, and this is most likely the biggest and best such company that has ever existed. Which is not to say that it’s not valuable — the thing is worth a couple of billion dollars, after all — but it’s clearly incremental to Tesco rather than transformational. It’s also pretty clear that things like this make sense as potentially helpful adjuncts to a profitable main business, not as goals in themselves.

That pretty much sums up my view of the value of transactions data as a marketing resource or as something that you can actually do anything with; since I’m currently reading a very good book on the subject, I might change my mind quite soon.  But specifically with respect to loyalty card data[3] I realise that I was hugely underselling the point by mentioning “potentially helpful adjuncts to a profitable main business”.

The purpose of a Tesco ClubCard is not primarily about the data.  The purpose of the ClubCard is that it’s a card which makes you more likely to go to Tesco’s than to Sainsbury’s.  Its heritage is not from the Amex and Visa systems; it’s from the Co-Op dividend card. 

In other words, data collection is more or less incidental to the real purpose of the system; price discrimination. (There’s actually a Competition & Markets Authority investigation going on into this at the moment, as they ponder where the line is drawn in terms of “give us this data and we’ll give you useful discounts on products you like”, versus “the advertised price is only available for members of our loyalty scheme”, versus “we don’t really need to do price competition any more as most of our customers are locked in via card membership”).

To my mind, this is a distinction worth being aware of in a number of fields, and it will remain at the back of my mind while reading the rest of the book.  Sometimes, gathering data is something people do because they want to influence your behaviour.  But a lot of the rest of the time, people are more directly trying to control you, and gathering the data is just the way that they check you’re mostly doing what you’re told.

---

[1] I reserve the right to change my mind about this as and when other themes are introduced at later points.

[2] Arguably true at the time, may or may not be now, don’t think it’s worth checking as it doesn’t affect the actual point.

[3] I have checked in the index and Fourcade & Healy don’t directly make this point about storecards, although I think they make other points that might be equivalent.