AI for sharp dressers

Last summer, Russia’s Sarafan Technology Inc. took part in London Tech Week international festival and presented its Sarafan.AI service there. The project is based on an artificial intelligence which can recognize various clothes in the photos and then find analogous items in online shops. By now, the service has 40 million visitors a month, with 150 multibrand and monobrand fashion stores connected to it. Its neural network can process 300K photos a day and the number of items on sale is over 20 million. Andrey Corhov, startup’s CEO, told Invest Foresight of the principles of Sarafan.AI operation and monetization.

Alexey Malgavko | RIAN

Andrey Corhov graduated from Siberian State Technological University and certainly could not imagine those days he would ever be involved with the fashion industry. He later visited Silicon Valley in the US where he saw achievements of the US developers of artificial intelligence and thought of an unusual application of image recognition. Sarafan.AI is based on behavioral patterns of websites visitors who watch photos looking for clothing to buy. Many women, for instance, would like to look like high society girls in the photos of paparazzi, but have no idea where such or similarly looking outfits may be acquired. Mediacontent consumption and online shopping are interrelated in such users’ psychology, hence for them it is totally natural to move from an online magazine to an Internet store.

To implement his idea, Corhov who at that moment headed Anetika, gathered a team of innovators and started teaching a neural network to recognize clothes and compare them with analogous items. The work lasted for several years, full of trials and errors. In 2017 the startup raised RUR 13 mio ($200K) of investments.

The system is based on selflearning algorithms and computer vision technology. Big Data is analyzed by cloud computation instruments (Microsoft Azure). The service can recognize not just clothes (skirts, dresses, hats), but various accessories as well. Neural network identifies the style of a clothing item, its color and material, as well as printed design. Ultimately, several matching options are selected from among millions of items of the catalogue.

Sarafan Technology Inc. invites bloggers and media portals, as well as online shops, beauty and fashion brands to joint its partner program. Technically, a partnership is accomplished through a native widget embedded into a website, which establishes a link between a photo of a product and an online store where it may be purchased. A user only needs to press Get This Look button.

Every unique transition to an online store from a website where Sarafan.AI recognized a product, results in a fee to the photo content owner. The bottom price is RUR 5 ($0.07) but it grows depending on the media outlet’s reputation and popularity, as well as on conversion of transitions into sales and scroll reach. A price of a visit may go up depending on a visitor’s activities at an online store’s page, the scroll reach and overall time spent. If a visit results in a purchase, an extra fee is added for each deal. The offer is most interesting for bloggers who write about fashion and lifestyle. A code-script for a widget installation is provided free of charge upon registration in a partner program. Sarafan.AI then automatically recognizes all appropriate photos of clothes and adds widget under every photo. No manual adjustment and content tagging is required. At the bottom of a page there may be a slider selection which displays all goods present in the photos. The service is fully paid for by online stores which in turn get additional convertible traffic. Inefficient traffic is paid at minimum rate.

Competitive advantages

Earlier, there already were some projects in the market which could recognize clothes in the photos, such as SnapFashion and Asap54, yet they required users to download photos of their interest all by themselves. Sarafan.AI therefore is a strong competitor to traditional advertizing platforms. According to experts’ views, effectiveness of such traditional platforms is now diminishing. An extra element is a harmonic addition to the content of niche websites and does not put some users off – unlike annoying ads. Get This Look button in view of consumers is not an ad but a hi-tech service. When watching photos of stars and opinion shapers, Internet users perceive the widget offered goods as personal recommendations of celebrities (influence marketing effect).

Sarafan’s widget click-through ratio CTR is three to four percent on the average. If a standard widget slider is installed, about 2.98% of Internet traffic is converted into sales with return on investments (ROI) reaching 86%. An important advantage of Sarafan is a widget deep targeting. Online shops’ owners can set geographic limits and identify specific websites where their goods are displayed, as well as arrange other settings. The service employs Smart-Click system and provides explicit statistics on each visitor attracted by Sarafan.AI. Businessmen can therefore see how rewarding the traffic from one source or another is, to adjust respective parameters accordingly.

By Christina Firsova

Previous ArticleNext Article