When it comes to e-commerce, Amazon rules the roost, not just because of its size, but also because of how it uses it to collect large amounts of data, which it then uses to keep feeding the machine with sophisticated product recommendations, relevant advertising, and more to keep people finding and buying things.
Today, Depict.ai, a Stockholm-based startup that has also developed a product recommendation tool — which it believes can help any retailer sell like Amazon — announced $17 million in funding to fuel its own growth in the United States and Europe, after signing up 60 customers including Office Depot and Staples.
Tiger Global is leading the Series a round, which also includes Initialized Capital, EQT Ventures, Y Combinator, and a long list of high-profile angel investors. It follows a $2.8 million seed investment from Initialized, EQT Ventures, Northzone, and Y Combinator last year, during which Depict.ai was part of the first cohort to complete the program amid a Covid-19 lockout. Depit.ai’s main idea, as described by CEO Oliver Edholm (who co-founded the firm with CTO Anton Osika), is that Amazon’s algorithms function so effectively because they have so much data on their platform about what you, and people like you, are buying.
It provides Amazon the power to decide what to offer you on a marketplace with millions of products, as well as what to store and grow as product categories and how to price those things. He claims that most other retailers have adopted the same mindset. “This is the same system that everyone else has used,” Endholm explained, “but they’re usually only looking at their own historical data,” which will never be as large as Amazon’s collection and won’t include information about active transactions.
Depict.ai’s solution has been to collect a much larger trove of data by aggregating data from across the internet; developing its own deep learning-based platform to “read” it in relevant ways (for example, in a search for recommendations after a customer searches for a dress, identifying data that relates to other dresses rather than models that look like the model in the initial search a customer made); and then ordering it to fit searches made on its customers’ sits. According to Edholm, the data is gathered by scraping a variety of websites across the internet. Scraping has sparked debate — some sites go to great pains to make it difficult or outright restrict it, while others have gone so far as to take legal action against scrapers — but Edholm points out that it isn’t illegal and is actually quite common practice in the business world.
“We train on huge amounts of data from the web,” he explained, “but there are a number of algorithms that do that.” “You can pick up on abstractions pretty quickly.” In any event, Depict.ai scrapes such a wide range of sites that even if one or two or ten were to stop it, there would still be a vast quantity of data to mine, and Depict.ai has already gathered a massive amount of data. “We’re not reliant on any specific site like LinkedIn or Craigslist,” he explained, referring to two platforms that have been scraped extensively for primary data that has been repurposed by others over the years.
“We normally want to find a lot of information on e-commerce products, and there are a lot of ways to achieve that, so I’m not concerned about barriers.” And we’ve previously trained our models and can do it again if necessary, as well as radically alter the data set.” The recommendation engine can then be integrated into the backends of its customers via an API. It claims that its technology may boost customers’ e-commerce revenue by 4% to 6% “without requiring any sales data at all.”