Consumer Profiling and Credit Card Data Mining

I’ve always loved read­ing and learn­ing about data min­ing and its applic­a­tions in vari­ous fields. Because of this, Charles Duhigg’s com­pre­hens­ive look at the con­sumer pro­fil­ing prac­tices of cred­it card com­pan­ies was my favour­ite read over the week­end.

[Research­ers] emphas­ized that the biggest profits didn’t come from people who always paid off their bills but rather from less-respons­ible cli­ents who nev­er paid their entire bal­ance. […]

But giv­ing cred­it cards to ris­ki­er cus­tom­ers posed a prob­lem: How do you know which card­hold­ers will pay some­thing each month, provid­ing fat profits, and which will simply run up a huge tab and then dis­ap­pear?

[One] solu­tion was learn­ing to pre­dict how dif­fer­ent types of cus­tom­ers would behave. Card com­pan­ies began run­ning tens of thou­sands of exper­i­ments each year, test­ing the emo­tions eli­cited by vari­ous card col­ors and the appeal of dif­fer­ent envel­ope sizes, for instance, or wheth­er new immig­rants were more respons­ible than card­hold­ers born in this coun­try. By under­stand­ing cus­tom­ers’ psyches, the com­pan­ies hoped, they could tell who was a bad risk and either deny their applic­a­tion or, for those who were already card­hold­ers, start shrink­ing their avail­able cred­it and increas­ing min­im­um pay­ments to squeeze out as much cash as pos­sible before they defaul­ted.

There are some fas­cin­at­ing insights in the art­icle, and through­out I was reminded of this Marissa May­er quote (from her Charlie Rose appear­ance), taken from Super Crunch­ers—a book on num­ber ana­lys­is and data min­ing:

Cred­it-card com­pan­ies can tell wheth­er a couple is going to get divorced two years before­hand, with 98% like­li­hood.

The valid­ity of that state­ment seems slightly dubi­ous, but I love it non­ethe­less.

4 thoughts on “Consumer Profiling and Credit Card Data Mining

  1. Paul

    I know from my own data assur­ance work with the UK Inform­a­tion Com­mis­sion­er, that cred­it card com­pan­ies (and more pre­cisely the cred­it ref­er­ence agen­cies they have come to rely upon heav­ily) are not the beacons of rig­or­ous sci­entif­ic ana­lys­is they would like ordin­ary cit­izens to think.

    For a start, accord­ing to a 1992 sur­vey of mined data from one large cred­it ref­er­ence agency (“CRA”) under­taken by a pri­vacy tech­no­logy writer for her book, over 30% of the indi­vidu­als’ data was incor­rect “enough to neg­at­ively influ­ence a cred­it decision”. That’s an alarm­ingly high error rate.

    On the oth­er hand, you won’t hear that study men­tioned by the CRAs’ mar­ket­ing mater­i­al boast­ing of highly accur­ate report­ing res­ults for indi­vidu­als.

    Anoth­er aspect which is for­got­ten is that cred­it inform­a­tion is often made accur­ate *react­ively*. For example, by default one UK CRA Elect­or­al Register data is updated only once a year. So if you moved into your new address less than a year ago, chances are you’ll be turned down for cred­it even though you are a registered and tax-pay­ing voter. After being turned down for cred­it by your bank, the CRA will likely receive a dis­gruntled call from you when you ask for (and pay for) your full cred­it record. At this point they’ll update your record, and send it to you. You’ll prob­ably see that the data is there cor­rectly, and take it back to your bank. And when you get there, hey presto! the bank’s records from the CRA will now show you’re registered cor­rectly. Do you see what they did there?

    In the UK at least, this reac­tion­ary rolling of cred­it records is unac­cept­able under the Data Pro­tec­tion Act 1998, but it is quietly tol­er­ated by the Inform­a­tion Com­mis­sion­er. To do it any oth­er way would mean that the CRA would have to be pro­act­ive, and that means spend­ing money updat­ing your data and chas­ing for updates – and they really don’t want that hassle when there’s gold to be had in them thar … inac­cur­ate records.

    So next time your bank does a cred­it check on you, take the res­ults with a big pinch of salt and tell the bank that they really shouldn’t rely on the CRA’s mar­ket­ing mater­i­als to gauge the accur­acy of the data.

  2. Paul

    And one oth­er thing about Marissa Meyer’s quote there.

    “Cred­it-card com­pan­ies can tell wheth­er a couple is going to get divorced two years before­hand, with 98% like­li­hood.”

    So when, why and how did the cred­it card com­pany per­form this ground­break­ing and expens­ive research exactly? I say expens­ive because they will have needed con­trols samples for false pos­it­ives and false neg­at­ives and a very large sample in the first place to factor out any con­flat­ing or com­pound­ing effects (demo­graph­ics, age, racial back­ground, house­hold income, etc. etc.).

    Which leads me to con­clude that to give this ‘research’ any cre­dence, either Marissa Mey­er has a ques­tion­able grasp of stat­ist­ic­al meth­ods or she simply hasn’t men­tally scoped the prob­lem. Or both.

    98% of stat­ist­ics aren not made up on the spot, but the remain­ing 2% give stat­ist­ics a very bad name.

  3. Lloyd Morgan Post author

    Insight­ful com­ments about the CRAs and I thank you for that; I’ll def­in­itely be more dubi­ous over future cred­it check res­ults, no mat­ter the out­come.

    I par­tic­u­larly agree with your com­ments about the Marissa May­er quote.

    I per­son­ally like the quote as it goes some way to help­ing people real­ise the amount of per­son­al inform­a­tion (or at least pro­file-able inform­a­tion) we give away to com­pan­ies purely by pur­chas­ing goods.

    As you say, to take this stat­ist­ic at face value and/or to give the research cre­dence without doubt shows at least a “ques­tion­able grasp of stat­ist­ic­al meth­ods”. I hadn’t thought about it, but by repeat­ing this on a show such as Charlie Rose she has seem­ingly done exactly that.

  4. Pingback: On Credit Cards « Accismus

Comments are closed.