What makes retail, and fashion retail in particular, an exciting playing field for data science? It might not be obvious at first sight, but it has all the ingredients to make it ready for a data-driven revolution: it is data rich, it is fast moving, and it is undergoing significant changes.
Even retailers with low degree of process automation have lots of data, which in most cases, is not used to its full potential. At the very least, retailers keep information about sales (down to individual sales transaction), stock levels, product descriptions and, typically, lots of image assets that they use for promotional marketing activities. In many cases, the information is far richer than that, and yet, very little is used to optimize internal processes or to gain detailed knowledge about customer behavior.
It’s not, of course, that they haven’t tried to use that data, but in most cases they’ve lacked the skills and tools (“big data”, in the sense of extracting information that is “hidden” under huge volumes of mostly useless data; artificial intelligence technologies to leverage semi-structured data -as product catalogues, social media feeds, marketing studies, etc.- and unstructured data -as images, collection reviews, etc.-; modern machine learning to improve prediction and optimisation capabilities, etc.), as well as clear ideas of what might be possible. Of course, the best way of show “what might be possible” is by doing it and showing it, and that’s exactly what we do in Nextail.
Being a data-rich environment and having exciting new possibilities of exploiting that data is all fine, but unless there is enough competitive pressure and those data-driven innovations can really help retailers deal with it, there might not be compelling reasons to invest on them. After all, many retail groups have been much more focused on brand positioning than in optimizing operations, and they might think, maybe rightfully so, that that’s best left to human experts.
But all of that has changed in the last 10 years or so, and today’s retail environment is one in which long-established brands practically disappear in one or two seasons, while new brands manage to gain market share at unprecedented speed. The key theme seems to be “agility” and the ability to deliver better shopping experiences, be it via faster renovation of the collections, omni-channel capabilities, more intimate knowledge of the customer or whatever means. Innovating, being up to speed with the competition is not an option: yesterday’s “good enough” is today’s “out of business”, and the list of casualties keeps growing. Retailers that want to thrive are hungry for innovation.
Now, we have a data-rich environment, opportunities to leverage that data, and pressing market conditions for retailers. The remaining question is, of course, this: Are those opportunities to leverage all that data what can make the difference for a retailer in those difficult market conditions? We strongly believe so, as proved by looking at some of the most successful retailers today, and as well as the top and bottom-line impact of Nextail solutions in our clients.
What are some specific applications of advanced data science that we are working on in Nextail? Here are some interesting examples:
– Forecasting demand and sales is at the very core of what we do: We need it to drive many other processes. We’ve moved from simple ARMA models to sophisticated ones that include hierarchical views for stores and products, censored data analysis (one can consider “sales” to be a store-availability- censored view of “demand”), rich modelling of seasonality effects, and much more. And we keep working on it, particularly focused on making it more robust, now that minimising the highest error has become more important than lowering the average error; and improving season-long forecasts.
– Managing stock levels at the point of sale is another intriguing problem. It is well known that stock information is extremely inaccurate at most retailers despite being one of the most important operational indicators. Some retailers report over 30% of stock data to be inaccurate. We cannot simply “believe” what our clients tell us about their stock, but instead we have developed a probabilistic model of what the real stock might be. This model allows us to anticipate some problematic scenarios like “phantom inventory”, “picking and packing errors”, etc. and react earlier.
– We use artificial vision techniques, augmented with meta data information, to detect similarities between products and automatically classify and estimate how well a new product might do. Of course, sensing sales is a much more reliable source of data, but that information might not be available early enough. We expect artificial vision to play an increasing role in some of our algorithms in the near future.
– Even with the best machine learning techniques in place, there is as much art as there is science to retail. We try to leverage the collective wisdom of the retailer’s sales force to enhance forecasts, selectively gathering information directly from store managers, area manages and so on, and incorporating that information into our calculations. So far, our forecasts are more accurate than employee forecasts… but only when running “business as usual”! On the contrary, employee forecasts are more accurate in exceptional circumstances like very early or very late in the season, or when introducing a new product that is very different from previous ones. Incorporating this feedback in our predictions not only improves them, but it makes them much more likely to be trusted by the end user / retail employees.
– The sales information we collect (store location and characteristics, size distribution, promotion activities, etc.), plus information we plan to progressively gather and use (buyer demographics, weather, timing of purchases in the day, joint purchases, etc.), open the door to extremely rich analysis of client preferences and buying patterns, some of which we already perform as an on-demand service to our clients. This knowledge will eventually enable us to solidly support crucial retail activities such as assortment definition, store layout design and overall season planning.
This is just a sample; the list keeps growing every day! I should point out that all the items on this list are directly related to the specific functional area Nextail is sharply focused on today: stock management. I haven’t even touched on end user behavior, store level optimization, collection design, brand management and so many other retail areas that can, and will, be radically changed by machine learning sooner rather than later. As Nextail, we might participate on some of those areas directly, or we might collaborate with third parties in the process of supporting retailers leap forward to their next stage. But one thing I am certain: we won’t stay on the sidelines watching others do it.
Visit our website to learn more about Nextail!
