For Amateur Artists Armed with AI, the Price is Right

A few years ago, she won an art contest in an online forum, and as a reward, the forum showcased 10 of her pieces in a permanent gallery. But first, she had to set prices for the pieces, so they could be listed for sale in the gallery. Sikdar was stumped. “I was thinking, I have no idea how to price my own work,” she says. In the end, the gallery did its own analysis and priced her pieces accordingly. But the experience gnawed at Sikdar, herself an avid collector of novice art, and it made her wonder how sellers on platforms like Etsy price their art.

Pricing Art Without a Safety Net

That question was the seed for a new paper recently published in the Journal of Marketing titled “Neither a Picasso Nor a Da Vinci: An Examination of Novice Artwork Pricing with Multi-Modal Data.” In it, Sikdar and co-authors Ishita Chakraborty of the University of Wisconsin-Madison and Nika Dogonadze of Google DeepMind develop an AI-powered deep learning model to understand the traits of Etsy offerings that are correlated with successful sales. Since the paper is part of a special issue of the Journal of Marketing devoted to impact-driven apps, they also built an app that may one day help novice artists figure out the best price for their Etsy listing.

The market for art is really two markets: one for “mature” artists who are well-known and brokered by galleries and auction houses, and another for “novice” artists who are relatively unknown and operate as microenterprises on online platforms such as Etsy or offline venues like street fairs. The market for mature art is driven by the name and reputation of the artist, and its artworks are considered investments with resale value. An entire ecosystem of art experts exists to put a value on these pieces and profit from their transfer. Novice art, by contrast, is mostly purchased for home or business décor and perhaps its visual aesthetics. Or, as Sikdar puts it, “As long as it matches my living room or sofa color, I’m happy, and if I don’t like it, I will probably throw it away.” The right price for novice art is, therefore, a bit murky.

Sikdar, being familiar with Etsy, reasoned that the most well-informed people on the price of novice art were probably those with a lot of experience selling it. That’s where she and her co-authors began their study: sampling 21 Etsy sellers, each with at least five years of sales, across three genres (abstract, landscape, and portraiture/still life), totaling 7,294 paintings. “These sellers have gauged the market and decided the best price,” Sikdar says, “and they know, based on Etsy’s own algorithms, how paintings surface and get noticed on the platform.” Their idea was to use the experienced sellers’ pricing to reverse engineer the artwork characteristics that could help set the prices of other work.

Using AI to Unpack Artwork Characteristics

To analyze those Etsy sellers, Sikdar and her co-authors built what’s called a “multi-modal fusion deep learning model.” It’s multi-modal because the data on an Etsy listing has two “modes:” structured and unstructured. Structured data is anything numeric that can be put into ordered rows and columns in a spreadsheet—think seller ratings, painting dimensions, price, and number of reviews. The unstructured data is everything else: the visual images and textual description of the painting, shipping and handling policies, genre, etc. Standard models can crunch structured data; unstructured data requires something more special. For this, the authors deployed a generative AI pre-trained model as an encoder to identify visual images and match them to their corresponding textual names and descriptions. It can see an apple, and the word “apple,” and know that they go together. The multi-modal model then combines the structured data and the vector embeddings uncovered from the unstructured data to achieve a holistic understanding of the Etsy listing, so all its characteristics can be mined to predict the best price. This is a cutting-edge method within marketing research.

What Sells, and Why it Matters

What did all this high-tech modeling conclude? One of the most interesting findings was that even among novice artists, reputation matters. The sellers’ rating and their number of admirers are highly correlated with robust pricing. Another key finding is that novice art has some similarities to utilitarian goods: selling prices are higher for artwork with good shipping and returns policies, a high-quality frame, and a well-made wire for hanging. Other important listing features include the medium (watercolor, oil, etc.), dimensions, certificate of authenticity, and personalization services.

After conducting the studies on predicting selling prices based on artwork features, the co-authors completed a concluding study that compared the listings’ features to how long it took for the items to sell. One important factor was the level of discounting: they found that listings that had been discounted took longer to sell. Also, a good artist reputation was correlated with a shorter time to sell. “These artists’ works are likely to sell faster, even if they’re priced higher,” Sikdar says.

Artists on Etsy gain a good reputation by making rapid sales. A key skill for Etsy sellers is, therefore, to set a price that the market deems reasonable. “Price setting is an important thing for someone just starting,” Sikdar says. The app that the authors built, which predicts the best selling price based on the listing features, could help artists determine a reasonable listing price. A reasonable listing price will in turn allow them to make more sales and thus build a good reputation online.

Sikdar is hopeful this research can help both platforms and artists. For firms such as Etsy, the research could help them design a tool that alerts the user (i.e., the seller-artist) if their price is too high at the time of listing. Currently, Etsy’s ranking algorithm sorts sellers by their reputation, so if you are a new artist on the platform, you have no chance of being highlighted. But if a user is realistic in their price setting, it could be a factor in the ranking, “so a complete newbie gets a shot at being featured on the first page,” Sikdar says.

For artists, Sikdar’s research shows that the textual description of their listing is what matters most. “The key is to ensure your textual description contains all the information about the genre, shipping terms, certificate of authenticity, etc.,” Sikdar explains. “That will make your painting much more salient.”