An image paints a thousand data

We sat down and spoke to Juliette Debois to understand how Heuritech allows brands to better connect with their customer base.


Business 22 March 2019

What is Heuritech and what’s its purpose? 

Heuritech is the first solution for predicting trends and the desirability of products that utilizes the eyes of millions of fashion influencers and consumers. Our goal is to help brands be ahead of consumer expectations, to fuel creativity and produce better at all steps of the collection development.

Our team has developed an industry-leading visual recognition technology that accurately analyses 3 million social media images every day. In any image, we can identify more than 2,000 fashion details, from a specific bag model, to a color or a pattern. We aggregate all of this data into an easy-to-use platform designed for merchandising, product and marketing teams of brands.

 How did everything begin? 

Everything started with a friendship. Heuritech’s founders, Charles Ollion and Tony Pinville, met at the Pierre and Marie Curie University where they were doing their PhD in Machine Learning. They shared the same interest in bringing science outside the laboratory and applying it to the real world and to real people’s needs. More specifically, they wanted to bridge the distance between technology and business. That’s why in 2013 they founded their company and started working in various sectors (agriculture, banking, etc.), developing tailored solutions for the firms which were their clients. They entered the fashion industry mostly by chance.

 Then what happened? 

In 2015 Louis Vuitton contacted them, asking Tony and Charles to be members of a jury tasked with evaluating innovative projects at a Hackathon. Talking to people working in this industry, they realized that there was huge potential for a tech company in the fashion industry since millions of images are shared online every day. That’s when Heuritech was born.

Did it require a new business model? 

We started as a self-funded company, then in 2016 received €1.1 million from a venture capitalist, Serena. This helped us grow a lot. Louis Vuitton was pivotal in another way. In 2017, Heuritech won the LVMH Innovation Award, and that was a crucial turning point since it brought us visibility, press coverage and ultimately new clients. We have been growing a lot from every point of view: revenue, customer base, and people employed. We grew from 15 people in 2017 to 32 people now.

What were your priorities in hiring new people? What kind of job profiles were you interested in the most?

We first needed product and marketing professionals because we had to understand the challenges and needs of the firms in the industry, and build a product that addressed their needs. But once you know who your clients are and have built a technological product tailored to their specific needs, you have to promote it. So we hired business development professionals along with a team of customer success specialists, to take care of our clients. Furthermore, we reinforced our technology by hiring data scientists and AI research engineers.

Speaking of AI, what type do you use the most at Heuritech? 

We define Artificial Intelligence as intelligent programs that achieve tasks that are usually tackled by humans. The specific branch we use is Deep Learning, a breakthrough that has revolutionized artificial intelligence since 2010. Thanks to it, the machine can autonomously learn to recognize concepts that make sense for us humans, such as a landscape, handbags, a smiling face, etc. Our industry-leading visual recognition technology accurately analyses 3 million social media images every day. Thirty-five per cent of our team members hold PhDs in Artificial Intelligence.

What is the process you apply to achieve your results? 

We count on three criteria to define a fashion trend: the intensity, the propagation and the velocity of a trend, and have created our own methodology to assess the relevance of any trend in the market. We start by segmenting audiences to differentiate early adopters from mass customers. We then monitor the spread of a trend, applying our algorithms. This allows us to classify behaviors into trend patterns. We can recognize fashion trends defined by type of product as well as its key features: colors, patterns, textures or shapes. Our research has shown that social media is ahead by months, for trends. According to a McKinsey study, an AI-based approach for demand projection could reduce forecasting errors by up to 50 per cent, while overall inventory reductions of 20 to 50 per cent are feasible.

Is it more challenging to build a classification or a clustering model, why? 

Both can be challenging, depending on what we want to cluster or classify. However, since clustering models are unsupervised, they have an additional, difficult component: there is no direct way to measure the performance of the algorithm since there is no labeled test data to assess the quality of the model. That’s why we build extra labeled test sets to measure the performance of clustering models.

There has been much talk about the disruptive effects of AI in the fashion industry: do you think these fears have some ground? 

I believe it’s necessary to have talks like these to really understand the effects that AI will have in the industry. However, I regret that these talks sometimes lead to fear: for me, AI will just be a tool to help decisions. AI is never going to replace designers, who are the ones who have the intelligence and emotion to really sense what’s happening in the market. It will never be capable of having emotions, and thus will only be a tool to accelerate. There is so much more we could do with AI to help the fashion industry become more sustainable: for example, demand forecasting and stock forecast. That’s our goal at Heuritech.

Your thoughts on Amazon? Is there any fear? 

It’s a very interesting question especially since a study published by WWD in November 2018 showed that only 38% of fashion executives are fully versed in AI and that industry executives fear the competition from Amazon. However, I don’t believe that Amazon is such a threat: consumers nowadays want to shop brands they feel they belong to, brands that share their same – that’s why so many digitally native vertical brands succeed so quickly. Brands have this very strong competitive advantage!

Looking forward, what new technology could be a game-changer in your field, and why? 

Blockchain is the next revolutionary technology. Especially because many fashion brands are faced with counterfeiting. This technology will significantly help fight counterfeiting by tracing products accurately. Artificial Intelligence, through image recognition, can also help, and both technologies can go hand in hand!