The fashion industry听is notoriously difficult听to predict despite the data revolution sweeping other sectors.听
Trends quickly take a hold and, as suppliers scramble to service the听newfound听demand, can听again听become听redundant听within a few short听weeks.听
Enter fashion prediction platform Dressipi, fifth on our recent London Tech 50 innovation ranking. Co-founded by Donna North and Sarah McVittie around 2012, the business is working with high听street听brands听John Lewis, River Island听and Topshop to听help them predict what customers will buy and not return.
鈥淔ashion and clothing听is听much more complex than most other products,鈥� North tells听老九品茶Cloud. 鈥淓very product has a very short lifecycle, only about six to eight weeks, and there is always a 鈥榗old start鈥� problem.听听
鈥淭he physicality of the consumer makes a real difference as to whether they can wear that item,听from a size or shape perspective;听then you get into their preferences;听and then听there is听the context.听听
鈥�So听there听are a lot of challenges听that prevent you from taking off-the–shelf models or algorithms and using them within the fashion domain.鈥�听
And North should know. With a strong background in digital media, the entrepreneur – who will speak at our free virtual event on fashion tech (click below to register) – was 鈥渆arly听to tech鈥� in the 1990s.听
/events/meet-the-fashion-tech-trailblazers/
At IMG she听oversaw听the creation of technology to听enable听new听business models,听digital rights听and听content across the听media giant鈥檚听portfolio, including for the International Olympic Committee and football federations.听
鈥泪t鈥檚 not very well known, but听IMG had a very big investment in fashion and modelling agencies. That’s where the idea for Dressipi first started,鈥� she explains.听She teamed up with听McVittie 鈥� a听former city analyst who had sold her听text message question answering service听Texperts听to the owners of 118 118 in a multi-million-pound deal 鈥� to听found听the start-up.听
鈥淎s more and more retailers听went听online, we could see that there was a big problem around navigating choice. Dressipi initially tried听to solve that problem: how do you edit down from a million听available听players to the ones that are the best match for the听consumer?听听
鈥淭he volume of data needed to build smart algorithms effectively听forced us听very quickly听to听move the business from a听B2C model听to听B2B, as that听enabled us to start getting access to volumes of transaction and activity data听and听build the smarts.鈥�听听听
Dressipi now claims to hold the largest and most accurate customer and garment dataset in the industry, powering billions of product and outfit predictions every year.听River Island says customers who interact with its personalised content are twice as likely to convert to a purchase, while their average order value increases by an average of 20 per cent.听
A team of 25 today, it听employs mostly听data scientists and engineers听who work with experts to gain that crucial human insight into fashion.听
鈥�Very early on, we听paired our data science and engineering teams听with fashion stylists so we could learn and implant听those kind of nuances听into the algorithms,鈥� says North.听
鈥�If a product type suddenly starts getting bought, a stylist might say ‘that became a trend three weeks ago’.听Suddenly听customers who said that they didn’t like polka dots and flares last month听are听going to purchase that item.听听
鈥淣o one other than somebody close to fashion will be able to pick up nuances like that, which are the small percentage difference in performance from a pattern perspective.鈥�听
Up to听40 per cent of clothing bought听online is听returned, compared with 5-10 per cent of in-store purchases.听鈥泪f you’re just looking at the data听and听seeing听a听pattern听of returning听that is illogical, a stylist might be able to say,听鈥�the reason that item is being returned is the neckline is very high – and if you look at the consumer base that are returning, they’ve all got very large bust size鈥�.鈥�听
Every product is attributed a听set of听25-50听data points听to create a 鈥榙igital twin鈥� of听every feature of a garment. 鈥淔or example, there could be听detail on the bust, transparency on the back,鈥� says North.听听
鈥淲e may then combine features to get a trend.听If polka dots and flares combined with a certain material make a trend, then we can annotate every garment with that set of features.听Then when it starts behaving differently to听how听we may expect, we would understand why that is.鈥�听
According to North, another problem unique to clothing is that around 40 per cent of products created are never sold at full price – and therefore are oversupplied.听听
Identifying true demand is also tricky.听鈥淎 retailer might sell out a product very, very quickly in a small number of core sizes. It is then a 鈥榝ragmented product鈥�:听they听can no longer promote听it听so it is听hidden and听it becomes听very difficult for a customer to get to that product thereafter.听
鈥�If you use personalisation on the front end听鈥撎�getting听customers to products as quickly as possible that best match their needs and their preferences听鈥撎�then听you have a view on what is true demand. And then you can start solving the supply side: what supply do I need for that true demand?听听
鈥淭he two sides are symbiotic:听we come in at both ends.鈥�听
North, left, with McVittie
COVID-19听saw a shift in online purchasing habits from evening听wear such as dresses听to听more casual clothing.听鈥�Our retailers saw a big shift in what the consumer was purchasing. We also saw a big shift in their听need to have more nimble access to their data,鈥� says North.听听
鈥�We听built data products which听enabled them to get a view in real–time about what was going on in their business so they could听access听the听insights they needed to react in the moment.鈥�听听
The company has mostly focused on the UK听but counts听OVS, the largest fashion retailer in Italy, among its clients. It is soon to announce clients in other countries, with听European, Australian听and US听expansion on the cards.听
The question of sustainability is also firmly in the founders鈥� thoughts. 鈥�Fashion is responsible for about 10听per cent听of global carbon emissions. Part of what excites us around making the supply chain more efficient is this ability to reduce wastage in the system,鈥� says North.听
鈥�In fashion the margins are very, very slight and it’s moving very, very fast. It’s very difficult for these retailers to stop and take a step back and say,听鈥�Okay, how听can听I do this differently?鈥�听听
鈥淭he way to tackle some of these issues is to help them improve their profitability;听get rid of some of the wastage;听create more optimum inventory in supply chains;听then that frees up a little bit more profitability and time to then really start looking at this huge problem.听听
鈥�It’s very difficult to solve it when everyone’s breathing down your neck and you’re a low–margin business.鈥�听