You need a fertile imagination if you’re going to build a business out of data. That might seem an odd thing to say – surely a business that makes its decisions on data runs by the numbers alone – but data can only take you so far. At some point human judgement and intuition has to take over. That intuition leads you to gather more data, which may challenge or validate your hunch. This then leads to more questions, more data… and the interplay continues as the intuition and the data reinforce each other all the time.
This was pretty clear at the very start of Dressipi. To be truly useful, a fashion recommendation service like ours needed to be able to ‘guess’ the taste and style of its users from the data we have available. Now this is a complex and therefore lengthy task. This isn’t just because taste isn’t a rational thing, but also because it can evolve over time or suddenly change because of an external factor. For example, how you normally dress at home may be very different to how you want to dress to attend a particular event like a wedding.
So, there were two ways of approaching this. We could wait to gather more and more data knowing that we would find ‘an answer’ eventually – or shortcut the learning cycle by applying some good old fashioned human instinct and expertise. In our case we achieved this by getting a stylist who knows the fashion industry and women’s minds inside-out to review and challenge the recommendations and output that our recommendation services gave to users.
We quickly realised this could be the solution. The data we were gathering from the women who signed up to use the service was great for improving the components that we knew about, but it didn’t quickly enough identify what we didn’t know, or had not yet thought about. This was more effectively found by the instinct and creativity of a human stylist – which could then be validated and fine-tuned by a cycle of data and stylist. We were so convinced by this approach that we immediately hired a team of stylists: one to sit alongside each one of our programmers and data scientists.
Our stylists’ expertise in selecting products and putting outfits together for real women was a revelation. They were able literally to bring our data to life, matching garments and outfits to a range of women in such a way that our algorithms could ‘learn’ very quickly when it was truly appropriate to recommend one style of dress over another. This approach has genuinely allowed us to lessen the gap between the pure algorithm-driven recommendations and real people’s tastes and preferences.
So the lesson we learned is that data is great for the ‘what is’ but not always for the ‘what is not yet’. In businesses that are innovating, data is only part of the picture. It is the canvas on which you can design your products and services, but you have to bring the paints.
