Before the internet, fashion trend forecasters travelled to remote corners of the world to closely document sartorial choices of resident micro-communities. This allowed them to identify the next big fashion trend. Now skip forward to the social media age where such micro-communities are easy to follow and access because they live on Instagram. It is a change that has given to new fashion trends on the most popular image-based platform. Making sense of this busy visual fest is fashion tech company Heuritech. They are trying to help fashion companies by providing them with more accurate trend forecasts. The French startup’s technology scans 3 million Instagram images on a daily basis to predict whether a specific style will sell or fail.
Proposition: Back Intuition With Data
At the outset, the Heuritech team first clarified that their solution doesn’t have a fashion eye. And that it is purely data-driven, backing (or not) the intuition and creativity of the design team and a traditional forecasting agency. The process begins with the curation of a panel of publicly accessible Instagram accounts, decided by the client brand and Heuritech. To capture both, trendsetters and followers, the panel composition is kept diverse with a mix of heavy-duty influencers as well as general fashion customers. Images are processed based on the brand brief; whether looking for skirt trends in Europe over the last two years or an analysis of the sneaker market to hone in on the most popular style.
But how is Instagram, a social platform polluted with brand-sponsored content, mined for authentic fashion trends? Well, Heuritech scans the text used in hashtags and posts and the visual content to filter out commercial posts as well as seasonal trends. On the process Tony Pinville, Co-Founder & CEO, Heuritech explained: “Despite the text and visual analysis, the results may not be fully accurate. We want to keep it transparent, and for every piece of trend diagnosis put out by our software, a confidence index accompanies the dashboard. For example, the analysis might state that Heuritech is 80% confident that the dad sneaker trend will slow down.” That accountability continues, as with each client brief the data accumulates on the AI-based platform, accessible by different departments, whether design or merchandising.
Visual Search For Fashion Trend Prediction
Heuritech’s proposition is based on computer vision (technology that extracts information from images or videos). This field has exploded in the last five years. In the context of fashion, given the nature of the product, a visual search will play a significant role, both for fashion brands as well as customers. Brands are likely to make internal use of this technology for trend forecasting while the visual search engines will target the end user. For instance, French startup Watiz uses computer vision and deep learning to suggest similar outfits for a given look to its users. Retailers like ASOS and Neiman Marcus have developed visual search technology for shopping recommendations. On the brand side, in 2018 Tommy Hilfiger teamed up with IBM Watson and The Fashion Institute of Technology (FIT) to work on an AI project for trend detection that would feed into garment design.
Even though Tony might position Heuritech as a tool that is complementary to services like WGSN and assortment planning platforms like Edited, eventually all of them can be expected to incorporate visual search over medium to long term to stay relevant in the digital world. While it operates in an increasingly crowded category, Heuritech is banking on a 5-year head-start for its algorithms. Also, the SaaS solution won the LVMH Innovation award in 2017 and had an opportunity to work close quarters with Louis Vuitton and Dior to develop a more industry-ready offer.
Starting with the accessory category, Heuritech has now expanded into apparel and is launching a trend-prediction tool targeting sportswear and fashion players in January 2019. Says Tony, “The solution can identify up to 2000 attributes in a fashion image that includes the most obvious ones like shape and colour to more complex ones like the type of people sporting the apparel/accessory in a certain image.” He further adds, “Our team is working to enhance the performance of algorithms, by being able to correlate image attributes with past sales to forecast trends.” For the fashion tool launch, Heuritech is aiming for the data-driven US market and brands like Nike, Ralph Lauren as well as sustainable fashion labels like Reformation and Everlane.
AI For Better Production
By adding a layer of data to the apparel design process, Heuritech claims that its offering will help avoid putting money behind declining trends, reduce overstock and enable apparel brands to produce in line with trends created by consumers themselves. Echoing that approach, H&M’s GM, Global Production David Sävman was recently quoted in a Vogue UK article saying, “We have to produce what people want to buy and not hope that they want to buy what we have produced.” As per the Business of Fashion(BoF)-McKinsey annual survey 2018, AI-based approach could reduce forecasting errors by up to 50% and overall inventory reductions of 20-50% is feasible.
However, in the current sustainability discourse, trends are often perceived as wasteful where products are being pushed into the market. Since technology like Heuritech can be applied to large as well as small-batch production, i.e. a 1-year prediction timeline or a much shorter one, will such digital solutions further fuel the fashion trend cycle? Fashion Futurist and Academic, Karinna Nobbs doesn’t think so. She shares, “In the future, almost every brand whether mass market or luxury will need to use data in a better way to not overproduce and that is from an economic and environmental standpoint. While we see the death of the mega trend and almost everything is acceptable, fashion will always change and a tool that predicts and de-risks that will always be useful.”
In its initial days, Heuritech was bootstrapped by 2 AI PHDs, looking to solve problems in the field of finance but spotting the potential in fashion, they changed course in 2016. The current 30 member team is equally split between techies and fashion professionals. Having received a seed investment of 1.1 million euros in 2016, the founders are working on the next round of fundraising.