AWS gives us state-of-the-art applications like Amazon Rekognition that make our business possible. I got a prototype of our service up and running within four hours and into production within a week.
Artfinder can match its customers with art they will love thanks to recommendation tools built on AWS. The company is an online art marketplace, allowing thousands of artists to sell directly to buyers. It runs its website and recommendation tools using Amazon Web Services (AWS) technologies such as Amazon Machine Learning, Amazon Rekognition, and Amazon Kinesis Firehose.
Buying art can be intimidating. Most customers rely on the opinions of a small group of tastemakers and gallery owners who decide what buyers see, what is considered “good art,” and what prices they can charge. Artfinder exists to change that dynamic. It’s an online marketplace, based in the United Kingdom and the United States, that lets artists sell directly to consumers. Artists win because they get a much broader market for their work, and consumers win because they have almost half a million pieces to choose from, at much lower prices than galleries typically charge.
David Tilleyshort is the chief technology officer at Artfinder. For him and his team, the big challenge is making sure buyers see artwork they want to buy. Given the volume of work for sale, this is no mean feat, so Artfinder has had to find smart ways to match buyers with art they will love. Tilleyshort explains the difficulty his team faces: “Like many sites, Artfinder recommends pieces to customers based on what they have bought previously, what they view, and the artists they follow. Unlike many sites, however, most of the pieces we sell are one-offs, so it’s difficult to say, ‘If you like this, you’ll probably like that.’”
Artfinder is all-in on AWS, and it has been since the company was founded in 2011. AWS gives Artfinder the flexibility to start new projects like its latest recommendation engine—which uses image-analysis technology to scan every image in its library and compare it visually to every other piece of artwork, accounting for things like color palette matching, texture, brush strokes, contrasts, and object analysis.
“We were doing this on the fly,” explains Tilleyshort, “but we found that wasn’t performant enough as our traffic grew, so we decided to run the entire analysis offline. We predicted it would take two weeks with our current infrastructure. Instead, we launched a cluster of Amazon Elastic Compute Cloud (Amazon EC2) servers and crunched through it in a day or two.
That wouldn’t be possible without AWS.” This comparison engine also powers @ArtfinderEmma, a Twitter bot who recommends artwork to anyone who tweets her a picture.
Artfinder asks its artists to supply information about each work to help people find it, “but that information isn’t always the best description of the piece,” says Tilleyshort. “We’ve been experimenting with Amazon Rekognition to analyze images to see if there’s anything else we can find out. For example, if there’s a dog in the picture, we can make sure it comes up when people are looking for dogs, even if the artist hasn’t specified that information.” Artfinder also uses Amazon Machine Learning to refine its algorithms about what behavioral criteria carry the most weighting when recommending pieces.
Tilleyshort says AWS has been instrumental in getting features such as this into production: “AWS gives us state-of-the-art applications like Amazon Rekognition that make our business possible.
I got a prototype of our service up and running within four hours and into production within a week. Without AWS, we’d have to invest in hardware and software, and most importantly, the expertise to get services like this off the ground.”
Artfinder uses Amazon Kinesis Firehose as part of another recommendation tool. The tool tracks user behavior on the site to see what products they view, what artists they follow, what they buy, what they like, and what they search for. “We were trying to collect all this information within our normal transactional database, but processing it wasn’t feasible in our normal application,” says Tilleyshort.
“So now we feed all that into our pipeline and then into our Amazon Redshift data warehouse, where we can do more offline analysis with it.”
This tech-savvy approach to the art market is proving popular. Artfinder’s revenue increased by 75 percent in the U.K. in 2016 (190 percent in the U.S. in 2016), and 60 percent of traffic now comes from outside its home market of the U.K. Artfinder uses Amazon CloudFront to deliver bandwidth-heavy images from locations near its users. “Rolling out our service in the U.S. didn’t present us with any real technical challenges,” says Tilleyshort. “Tools like Amazon CloudFront let us keep a good user experience, which wouldn’t be possible in a non-cloud world.”
Learn more about artificial intelligence and AWS.