The insight agencies - such as Dunhumby, Tesco's in-house analyst - like to claim the value of their contributions. What they contribute is usually tweaks to the exact mix of coupons sent to a more refined mailing list. Given the costs of direct mail, that may be worth doing. Google and Facebook are basically selling their insight technology: they are offering the world's best targeted advertising. Tesco, Sainsburys and other supermarkets with loyaly cards can tie purchases to people, and banks can look at your current account transactions. Hint: if you want to hide your patterns of consumption from financial services companies, pay by cash.
As ever, I like to contrast Amazon with the The Bank. Amazon use an insight approach when they make suggestions based on "people who bought this also bought these". It works well with genres - a textbook on Galois Theory, post-rock music, romantic novels or cookbooks - and it may work with accessories for bigger-ticket items, but outside well-defined genres, where customer volumes are small, it can get bizarre. With volume comes consistency and reliability, but also blandness: large numbers of people who bought Katy Perry also bought Ke$ha. I'd never have guessed. A mathematics publisher won't need telling that people who bought Hartshorne also bought Hatcher, and Mumford's Red Book, but it's only news to outsiders. Which is exactly who it's aimed at: consumers like you and me who are browsing. I think the Amazon algorithms are fine, and the occasional bizarre suggestions just confirm I'm in a minority. What makes Amazon so effective aren't clever algorithms - though that helps. Amazon's real advantage is that it controls its data: you tell it what you bought, using the references that it supplies. Retailers do the same with bar codes.
Not so much in retail banking. Retail bankers and their IT people aren't as aware of the advantages of standards as their counterparts in telecoms. As a result, a bank can identify the method of payment, but not what it was for. Reliably. That £254.32 you give to Sainsburys each month? We don't know if it's for food, a loan repayment, savings, a credit card payment or whatever else. Banking developed in the days when retailers generally did one thing - except the Co-op and Woolworth, and everyone paid them with cash. The industry never recognised a need for standards in transaction description.
As a result, when I took over a report that claimed to tell us how many of our customers were taking loans with other lenders, I found more ambiguities and estimates than had been advertised. And I can't resolve them. Nor can a room full of analysts, because we don't have the data. Big Brother may be watching you, but his glasses are steamed up, and he doesn't know what he's looking at half the time. (Are you convinced by those grainy CCTV photos they show on the news? Not me.)
Messy data in, sloppy conclusions out. I discovered recently that people were using an earlier an even messier version of this data to calibrate pricing models. I have no shares in The Bank for reason.
Insight analysis isn't management information, which is itself a step down from financial reporting. As long as the segment my analysis tells me should be large and spread over the country doesn't turn out to consist of two hundred people in Guernsey, I'm okay. Does it matter if it's really 4.5m people instead of 2.5m or 6.5m? As long as we start small with the ability to scale up quickly? The point is to spot a genuine segment, and it may not even be that. The point is to produce a product or service that makes money and people come back to buy. Does it really matter how precise the process is? Hell, I'd be happy if I designed the product for one group of poeple who ignored it completely but it was taken up by another group to the volumes we were hoping for. The onyl issue there is that I may not be able to repeat the fluke.
But if the success is big enough, I don't need to.
No comments:
Post a Comment