Nadezhda Anisimova, marketing and PR manager at Sarafan Technology Inc., talks about how easy and unobtrusive it is to nudge a media portal reader to buy clothes of a particular brand.

Nadezhda, what does your company do?

Sarafan Technology Inc. is in business since 2016. We have developed artificial intelligence called Sarafan.AI and taught it to recognize objects in photos and videos.
Our research showed that photos and videos most often inspire people to buy fashion-goods.

We decided to start with the recognition of these sources.

Today we work with media portals. Our system analyzes photos and videos published on the site and searches for goods that appeared in the frame. All found goods are displayed in the widget. The user clicks on the product in the widget and is redirected to the online store. The only thing the media portal needs to do is install our widget. All work is handled by artificial intelligence.

Nowadays, many companies opt to release their application, why did you not choose this path for yourself?

Initially, we developed a mobile application which you could use to find clothes shown in a photo.

But a year later we decided to give up this idea. First of all, users took photographs of the goods themselves, and amateur photos can’t always boast high-quality. It was difficult for artificial intelligence to process them. And it’s no wonder. How, for example, would it understand what color a light cloak is, if the photo was taken in the dark under lamplight? Those poor-quality photos were numerous, as a result, users did not get the desired matches and deleted the application.

There was a second reason why we abandoned that strategy: users downloaded the application only when it was necessary to find a thing and after some time they forgot about it, ultimately deleting it altogether.

How can your technology be useful to an “ordinary user”?

Let me explain to you how Sarafan.AI works in a typical case. Imagine an article about fashion trends.

In one of the photos there is a blonde girl in a blue jacket and a dress with polka dots, she wears red lipstick and earrings in her ears. The system processes this information and finds similar dress, jacket and earrings. At the same time, the system analyzes the model’s make-up: the AI determines the color of the lipstick, skin tone, hair type, and then selects recommended goods for creating this look: red lipstick, nude face powder and a shampoo for blond hair. The photo has a button “Buy this look”, or small pins on each product. The user presses the button or pin to see what products are shown in the photo, find out their price, brand and, if desired, can go directly to the purchase.

The system continuously checks and updates links – products that are out of stock in the online store are deleted from the widget. And in this case, a user does not need to install any apps. They simply visit a site and read the article. An example of how the widget works you may find on the Cosmopolitan portal.

What benefits does the media portal get from this cooperation?

Our technology is used for internet marketing. Sarafan.AI embeds advertising into the content, and the site gets an additional monetization tool. It does not compete with contextual or banner advertising, our technology creates new advertising opportunities. The site receives payment from the advertiser for redirections to the store from the widget (CPC model), purchases from the links provided by the widget (CPA), or 1000 views of the goods in the widget (CPM). Sarafan earns commission from the profits of the site.

For media portals, Sarafan widgets are not only an advertising tool, but an interactive addition to the content. With the help of the widget a user can interact with each photo. Accordingly, user’s engagement with the content increases.

Indeed, it is very convenient that you can show unobtrusive recommendations to a user, and the bottom line is that all benefit from it. But still, are there any stumbling blocks in how the system operates?

Yes, there are some pitfalls. Artificial intelligence has no intuition. The machine can only see what is shown. For example, it cannot guess that the photo, where the model is shown down to the waist, shows a dress and not a top or a blouse. Therefore, the quality of content is very important for the correct operation of our technology. Large media portals have no problems with that, but smaller sites may have difficulties. Due to the poor quality of the photo, the system may make a wrong guess with the color or even the type of product.

Automatic advertising also does not always guarantee safe context for the brand. To prevent widgets from appearing in articles with negative content, we taught our artificial intelligence to “read”, i. e. our technology determines the tonality of the text and does not add advertising to articles with a negative context, for example, “Top ten celebrity looks gone wrong” or “TV show star died in a terrible accident”. Text recognition is a very difficult task to handle. All languages have words with the same spelling, but with different meanings (for example, Object and objEct). A human being can distinguish between their meanings, but a machine can’t.

Therefore, in controversial cases, our assessors take over and manually remove the article from the markup. The advertiser can also provide a list of stop words which the brand should not be associated with.

What are the prospects for the development of your technology?

Just a little while ago, we launched a new advertising format, it is called In-image. Like I said, our technology is very precise with fashion and beauty products, but it can also recognize other objects.

For example, Sarafan.AI can find all photos with cars or a Christmas tree on a site, and then embed a suitable advertisement in them. Car photos will advertise a minivan, and Christmas tree photos will promote a gift shop. Turns out that the advertisement is targeted by the objects shown in a photo. Objects for targeting are determined by the advertiser.

Tell us about the development of the company, did you attract investments?

Since the beginning of 2017, we have been residents of “Ingria” Business Incubator. It is notable for its great and inspiring job climate. Every month it hosts meetings with foreign investors. We participated in meetings with investors from the Baltic States, Finland, Japan, China, Belgium and the Netherlands. “Ingria” also organizes events with potential customers. Not so long ago there was a very interesting and valuable meeting with Concept Group. It was a unique thing. It was the first time when the brand came to hear out startups, and not vice versa.

We now have two offices, one in St. Petersburg, the other in New York. All technological developments are conducted in Russia, and in the USA we looking for partners and sales opportunities. We have already penetrated the European market. We have connected partners in Germany, and now we are negotiating with media companies in the UK, Italy and Poland.
As for obtaining investments, we are also working in this direction. We decided to look for investors for the seed stage at competitions and large conferences. The project was pitched at different sites, where we met our first investors, The Untitled Ventures fund and Sergey Dashkov. They invested $400,000 in the project. After the first investment, we did not specifically go and look for similar opportunities, the project began to attract money all by itself. In September 2018, we closed the second round and attracted 1.3 million USD from Admitad Invest, a German fund. The Untitled Ventures and Sergey Dashkov again invested in the project.

What competitive advantages do you have?

We have no competitors in Russia, in just one year we have entered into a partnership with all the major Russian media houses. Now we are working our way onto the markets of Europe and the US. There is where competition is already, but we are continuously working to be one step ahead. For example, we offer partners a flexible business model, when a media can choose any monetization model. Of course, we are improving technology, too, now we are training our artificial intelligence to recognize goods in streaming videos. So far, no other technology can do this.

Source: vc.ru