Better bra sizing.

It's an idea I've been kicking around for a few days. The technical and marketing-based obstacles towards getting it to work have turned me off pursuing it, but I figured it was worth sharing.

I work with operational databases for a luxury fashion retailer, and bra sizing as it currently exists (back size + cup size) makes absolutely no sense. I will sometimes ask female friends to explain how the size given for a garment can possibly be of any use in determining comfort and fit. Their answer: it doesn't.

Their actual answer tends to be a rant about inconsistency between product ranges and how contemporary bra sizing is next to useless. A couple have been both eloquent and insightful. A few times a year I'll have an idea I get excited about turning into some sort of web-based service, and in spite of its silly-sounding nature, this one is easily the one that's had the most philanthropic weight behind it.

The idea: a website containing a comprehensive list of commercially available bras. Users sign up, locate bras they own (or have tried on) and rate them along various measures of comfort/fit/support, etc. The service then locates clusters of users with similar preferences to them (exact method of analysis still up for debate, but a few likely candidates stand out), and suggests specific sizes and ranges that would meet their needs.

There are three sides to this. The first is users getting the service described above. The second is the option to license out the size/fit data to interested third parties, such as manufacturers and retailers, which would probably be the most sizeable revenue stream. The third is the possibility of using the data to produce a better sizing scheme that more accurately tackles the real-world problem.

I see two main problems with the idea. The first is encouraging user uptake (convincing women to spend time inputting details about their underwear into a website). The second, which is related, is giving them incentive to do so without the recommendation algorithm in place. I have no idea if k-NN or spectral partitioning or probabilistic classifiers or regression analysis will be any good at all in carving up the data appropriately, and I won't know until I get a sizeable set of data to develop against. There'd need to be an existing service provision for the users to encourage them to sign up and provide the data before the interesting work even begins. An existing comprehensive list of commercially available bras complete with a flat non-super-stats-enhanced rating system might be enough to get the ball rolling.

I should reiterate that I'm unlikely to pursue this idea. While I have a background in web dev, data analysis and technical fashion retail, I'm far from an expert in any of them. Still, if anyone wants to convince me otherwise, give me more reasons why it's a bad idea, or steal it outright, please go ahead.