Sunday, February 02, 2003

Congestion Cleansing

I was thinking about the Congestion Charge today - it's coming for Londoners (and those who drive in for work or fun) in just over 2 weeks. I'm not at all close to this, so any thoughts are purely speculation. Anyway, I'm guessing that for the last couple of months it's been in the equivalent of User Acceptance Testing - i.e. the cameras have been snapping pictures of number plates as they zoom (or crawl more likely) past the cameras. Once the plate is snapped, it must be looked up in a database, presumably the DVLA's records (or, if they're doing it right, a clone database that is held locally for faster performance). Most databases that I come close to have error rates of around 20% - 30%. Let's suppose that DVLA's is 20%, maybe it's as clean as the average tax database (after all, I assume that every car older than 3 years gets its address updated with every MOT? I don't know the split of cars younger than that which will have worse data, or the number of company cars that perhaps have better data). So, the plate gets looked up, the address found and a letter goes off the next day if the money hasn't been paid. Doubtless if the letter is one of the 20% that goes to the wrong address, the recipient will ignore the first one (or maybe, if we're lucky, mark it as "address not known" and return it to sender). If they do ignore it, they're going to get a shock a few days later when the debt collection mob show up. So, how do you improve on the 20% and so cleanse your database? I guess the easiest way is: Once you get the number plate and the address from the DVLA database, you do a look up on a clone of the electoral roll and see if at least the names match. I don't think there's a single electoral roll (at least not a complete one) for London, so either there have to be separate lookups to individual boroughs, plus those outside London (and, it could be a car from Scotland or anywhere I guess, to make it more complicated) - so that probably doesn't work. I think that leaves someone like Experian or Equifax, who have good records - and have the software to do fuzzy matching on various spellings of names. I have no idea if that kind of thing is allowed by the Data Protection people, but it would seem to me the only way to improve on the matching - after all, pretty much everyone who has a car probably has a credit file and so appears on the lists that these folks maintain. Richard Granger, he of NHS and highest paid civil servant fame, is one of the people who put this programme together, so I'm guessing that all this kind of thing has already been covered as he seems very smart. It will take a few weeks for it to play out, but I'm still expecting there to be a lot of wrong addresses receiving bills for cars they don't own ... and it won't be long before the Evening Standard reports the first debt collectors turning up on someone's doorstep to confiscate or clamp their car.

No comments:

Post a Comment