The Art Detective
Is This Pissarro Print Worth $28,000—or $530? At 1stdibs, Dealers and an Algorithm Clash
Start-ups are using AI to value art, with mixed results. One project has some sellers in revolt.
Start-ups are using AI to value art, with mixed results. One project has some sellers in revolt.
Katya Kazakina ShareShare This Article
Pricing art is a tricky business—part science and part, well, art.
Many tangible and intangible parameters determine the value of a work—the artist who created it, its date and medium, condition and provenance, and more. But those are just guideposts, and when the market is in flux, as it is now, it can be especially tricky to land on a number.
Just take the Basquiat that hammered this week at Sotheby’s in London for £15 million ($19 million), its low estimate. In 2022, Portrait of the Artist as a Young Derelict (1982) was slated for auction at Christie’s with an estimate of $30 million. Then it got pulled, presumably because potential buyers deemed the estimate too aggressive; its anonymous seller, who is Paris dealer Jerome de Noirmont, did not want to compromise.
After two years of high interest rates and amid an art-market downturn, the seller lowered his expectations. He got a house guarantee, and Sotheby’s lined up an irrevocable bid. That backer bought the work and received a fixed fee for his efforts, based on its final price of £16 million ($20.3 million), which includes Sotheby’s new buyer’s premium of 10 percent for works above $5 million.
What’s the real value of this painting? Is it $19 million? $20 million? $30 million? It’s easy to imagine someone paying much more for this Basquiat one day, especially given how much the artist’s market has risen. Or it could sell for a lot less. Tastes change, and even the greatest artists of one epoch go out of fashion in another. We’ve seen this happen again and again.
Right now, prices for some artworks are in disarray, with primary prices for many popular artists surpassing their resale prices, as I’ve noted in my recent columns. That doesn’t inspire confidence in buyers. And it can really unnerve sellers.
Algorithmic Stress
Technological developments are also presenting their own bizarre issues.
This was brought home to me this week by some drama at 1stdibs, an online marketplace for collectibles and, increasingly, art.
Founded in 2000, 1stdibs went public in 2021. Its shares peaked at $19.62 shortly thereafter, but are currently down 81 percent from that high, trading below $5 on the Nasdaq. Annual sales are down 15 percent since 2022, and annual revenue declined 12 percent to $84.6 million during the same period, according to its filings.
Art sales, on the other hand, rose 7 percent in the first quarter of 2024 versus the same period a year ago, the most of all its categories, according to the company. In 2023, they accounted for about 10 percent of the platform’s $362 million in sales.
Recently, 1stdibs introduced “Pricing Guidance,” a tool for buyers and sellers that uses past sales of similar items on the platform to estimate an item’s worth.
The algorithm is wreaking havoc in some quarters of the site because of the ridiculously wide gap between what some dealers are asking for some pieces and what the algo is proposing.
A pre-Colombian gold shaman pendant offered at $27,000 should be just $630, according to 1stdibs’s tool. “Based on our pricing data, this item is $26,370 above the recommended price,” the auto-generated text next to the item said this week. Oy!
Similar language appeared next to Camille Pissarro’s etching L’île Lacroix, à Rouen (1887), offered at $28,000; 1stdibs estimated that is “$27,470 above the recommended price.”
A James Whistler, etching Old Putney Bridge (1879), is listed at $20,000, or “$19,470 above the recommended price,” according to the website.
This week, a few outraged dealers sent angry letters to the management of 1stdibs. One called the company’s recommended prices “absurd,” and another said that they threaten to “undermine and destroy” not only the gallery’s reputation but also the company’s.
“The ‘Guidance’ policy is not only unwarranted; it demeans both the site and its dealers,” Harris Schrank, a board member of the International Fine Print Dealers Association, wrote in a letter to the firm’s president, David Rosenblatt, on Wednesday. “I fear that some of my buyers who may not know me well may see such ‘guidance’ and wonder whether in past sales they’ve been overcharged.”
Schrank, who specializes in what he terms “exceptional examples” of fine printmaking from 1490 to 1940, is demanding that the policy be terminated ASAP. “If not, I’m afraid I’ll have to drop the site,” he wrote.
Here’s one example of what upset Schrank: While he is trying to sell a Martin Lewis (1881–1962) print made during the artist’s life, Rainy Day, Queens (1931), for $48,500, 1stdibs estimates it at $750.
It’s the priciest of 27 works by Lewis, an Australian-born American etcher known for his urban scenes, listed for sale on 1stdibs. (Five of them are marked “price on request.”) One of them, White Monday, printed in 1980, 18 years after the artist’s death, is priced at $700 (compared to guidance of $310).
Edward Pollack, a prints dealer based in Portland, Maine, sent a letter to Rosenblatt with an analysis of the Lewis listings on 1stdibs. The average price is $22,116, and the median price is $24,000, Pollack wrote in a letter viewed by the Art Detective. He noted that he had sold a work by Lewis for $60,000 at auction (the buyer paid $75,000 after fees).
However, 1stdibs shows $500 or $700 as the “median” for Lewis’s prints, Pollack wrote.
The tool “makes no distinction between posthumous and lifetime prints, nor does it take into account the desirability or lack thereof of one item versus another, which, as anyone ought to know is the most significant factor in establishing a price,” he wrote.
Pollack has been in business for five decades and has spent seven years selling on 1stdibs. The platform accounts for about 30 percent of his business, he told me this week. As an owner of the company’s shares, the stakes are even higher for him, he said.
“As a stockholder I have seen my investment shrink by more than two-thirds,” Pollack wrote in the letter. “While I understand that many factors are involved, it cannot be helpful to the success of the company that it is undermining its own dealers and its own reputation with a Price Guidance policy based on an algorithm that doesn’t produce valid results.”
I reached out to 1stdibs for comment about the “pricing guidance” complaints from the fine prints dealers.
“We have been testing and optimizing the pricing guideline tool to improve its accuracy,” the company’s chief commercial officer, Matthew Rubinger, said in an email. “To that end, we have a number of tests planned and in progress that address concerns from sellers.”
Rubinger said that 1stdibs already does not show recommendations for items over $50,000, since it does not have enough transaction data to draw on. “In addition,” he said, “we will be removing pricing recommendations in a variety of contexts, including where the distance between the list price and the recommended price is significant. We are also working on adding more item attributes to improve the recommendation accuracy.”
It appears that 1stdibs already implemented some changes: for one, the tool no longer lists the exact, glaring difference between the dealer’s price and recommended price.
A Brave New World
Can algorithms and AI accurately determine the value of art? A growing number of players are working on this, but the industry remains resistant, a prominent art advisor told me.
Live Art, co-founded by Adam Chinn and Boris Pevzner, offers such estimates by feeding an AI model with numerous factors, ranging from the work’s size, year, and palette to past auction prices and economic conditions, including interest rates and the price of gold. Its estimate for Basquiat’s Portrait of the Artist as a Young Derelict this week was $20.3 million, the same as the final price of the work (which reflected the addition and subtraction of various fees).
Another company, Artory, is using AI to help financial advisors appraise large art troves for clients who have other assets under management. The purpose is to get a rough estimate of what these holdings may be worth.
“It’s not a one-stop magic bullet,” said Timothy Kompanchenko, chief technology officer at Artory, which has partnered with Winston Art Group to create an AI-powered database of art prices for internal use, including private sales.
AI models, even the most advanced ones, like ChatGPT, are powerful tools but require careful guidance to be effective within specialized fields like the art market, Kompanchenko said. Their accuracy depends on the quality of the data they’ve been trained on and how well they are fine-tuned to the nuances of art valuation. To achieve reliable results, these models need ongoing refinement by art professionals.
“AI is very smart,” Kompanchenko said. “But it requires experts to inform its development and usage. You can ask ChatGPT for a price, and it will provide an amount, but the critical question is whether it’s the right amount.”