Analytics are everywhere. And they’re great, unless you’re the one on the wrong side of the spreadsheet sort. Data doesn’t solve all problems, however – data, like photographs, can lie. It can certainly lead you astray. We all know that perfect is often the enemy of good. But sometimes, “good enough” is actually better than “perfect.” Really. That outcome confuses the high priests of analytics. (There’s a whole section in our book, The Plateau Effect, called data idolatry.)
I saw a great example of this recently when I met serial entrepreneur Jin Koh, whose new company Original Stitch manufactures custom-made shirts for men. He’s not the only one, of course. Fashion firms have been working towards popularizing made-just-for-you clothes for some time. Turns out, it’s hard to do.
The process seems obvious. Scan your body electronically somehow, then make a shirt (or pants, or a dress) that’s precisely altered to your body. You’d think that would create a perfect shirt. It does. It’s just too perfect.
Here’s my Q&A with Koh about the problem of relying too much on data when solving a problem. We’d be wise to hear him out as we decide what artificial intelligence might mean to our futures.
Me: 3D modeling for clothes — clothes made precisely for your body, designed with exact measurements — seems like the Holy Grail of fashion. But you told me it doesn’t work. Why not?
Jin Koh. It doesn’t work because body measurements are not shirt (garment) measurements. You are buying a shirt not a bodysuit. Imagine a shirt that’s cut to your exact body measurements — it’d be like a bodysuit. Shirt measurements are cut larger than your body measurements so you can wear it, move in it, and feel comfortable in it. The difference between shirt measurements and body measurements is called “the allowances.” Some people prefer loose fit (more allowances), slim fit (less allowances). 3D scanning technologies today can extract body measurements but they cannot extract your desired fit preferences. Slim fit, super slim fit, loose fit varies from brand to brand, and is very subjective from person to person. Which is why sophisticated body scanning technologies still result in 20% product reject/return rate. Which is also why professional tailors ask you to come back for second fitting just to get the allowances right. The bigger problem is: how do you build out an automated way to understand the fit preferences for each individual and send them a shirt that fits the way they want.
Q: If that doesn’t work, what does? Should we just throw out all that precision data?
The industry is missing the ability to automatically extract your desired fit preferences at scale. We don’t have to throw out precision body measurements data. We just need to add allowances data to the body measurements. That allowance data varies from people to people. At Original Stitch, we don’t invest in extracting body measurements from 3D because they has been solved and will continue to be improved in the future. Amazon’s recent acquisition of Bodylabs can extract precise body measurements from Echo Look, for instance. At Original Stitch, we invest in extracting your desired fit preferences. Our approach is “why don’t we just replicate the measurements of your favorite shirt?” We can make a new shirt to your favorite shirt measurements and when you get it you are going to love it – providing total predictability and expectation. With Original Stitch Bodygram – simply take a photo of your favorite shirt and upload it to our site. Our proprietary computer vision and AI systems will extract shirt measurements from that photo.
Q: What does your experience with precision- and data-based design have to say to other industries about Big Data?
Body scanning technologies will improve over time to extract very precise body measurements, combined with highly precise fit preference data, these data can be served as training sets for AI to improve the result for the next customer. Big data, and data science in general will be key to unlock new values for all industries. Take the apparel industry for instance, with body and fit data, brands can increase purchase conversion rate, cost of return, and develop new clothing that fits better.
Q: Do you think there are lessons in what you say for other businesses? Data is great, but it doesn’t always allow for …allowances?
1) Data is good for business. We all know that, but we still need humans to analyze data and draw conclusions on what’s best for our customers. Such a role, one can argue, would be replaced with machine learning, but there’s a reason why Google still hires human editors to curate google news, Facebook still uses human to filter fake news, and Stitchfix hires human stylists to curate styles.
2) For the broader business, I’d suggest businesses to start initiatives around data collection and hire a data science team. Data collection is good if done in a respectful and transparent way. The primary goal is to better serve the customers with more personalized content. A data science team is a bunch of PhD’s who will 1) secure your data (top priority) 2) analyze data in big cohorts (protect your privacy) 3) develop reports to help drive product development and marketing.
3) Human assisted AI is good and all but we all have seen what Microsoft AI bot has posted on Twitter (funny), and what happens when you let Alexa and Google Home talk to each other. Human assisted AI is the solution. StitchFix collects tons of data from customers but still has a human stylist make the curation based on the data collected. The user gets better curation, and the stylist makes better recommendation thanks to data. That’s how it’s going to work.
Full disclosure: I bought my own Original Stitch shirt (and paid full price, except for a coupon) after meeting Jin Koh. And I like it.