top of page

The Ledger in the Basement

7 hours ago
7 min read

AI and the art market's unread record: an insider's archive, a silence at the core, and a financial layer built on the thinnest part of the record.


In 1992, Rufus Bird spent a month as an intern at Christie's. He did not spend it in the saleroom. He spent it in the basement, in the Muniments archive, among the records of every Christie's London sale since 1766: the catalogues, and the Day Books holding stock numbers, client information, consignments and sale details, kept on rolling shelves. His job was the "look-up": answering queries about stock numbers by hand, from the paper.


Bird has now written about that basement, in a piece first printed in The Critic in October 2026 and republished on his Substack, Ornament, on 2 November 2026. His argument is simple. The way the auction houses sell has "remained unchanged since the eighteenth century", so their record runs unbroken across more than two centuries. And the two great houses have not digitised it.


His piece does not mention AI once. It speaks of big data, of coding, of digitalisation. That is exactly why it matters here. For five months our monitoring has looked for AI at the core of the art market and recorded silence. Bird, from inside, describes what lies beneath that silence.


Two silences, one floor

The record on our side is plain. By mid-November 2026, fourteen consecutive counts had passed without a major auction house announcing an in-house AI tool. Beneath that line, the cataloguing vendors keep multiplying and the houses that use them stay unnamed. Across the whole field, AI tools appear where knowledge is already written down in structured form (watch references, trading cards, wine labels, auction price records), and they stop where knowledge is held by people: attribution, taste, advice, private sales.


In October 2026, we put this in one sentence. The art world has not yet worked out what it needs AI for, and the knowledge AI would need is not there in a form it can use.


Bird gives that sentence a physical address. The knowledge exists. It sits on rolling shelves, in ledgers, in a form no machine can read. An AI system learns from records, and at the centre of the art market the records are paper. The silence at the core is not only reluctance. There is very little at the core for a machine to read.


The market's own figures fit the picture. Bird notes that global sales have stayed at about $60bn for roughly a decade, with "no dramatic change in profit or trading volume". A decade of online sales, digital platforms and AI announcements has not moved the total.


The naming problem

Digitising the ledgers would only be the first step. Bird leans on David Kusin, a former curator at the Metropolitan Museum and an art-market economist since 1998, whose book Art as an Asset in the 21st Century argues that the market has never turned its records into "actionable knowledge". Kusin counts about 150 submarkets, each with its own habits. The obstacle, in his words, is that "the current chaotic state of nomenclature means that it is impossible to sort data."

Bird adds his own experience: as a curator working with about half a million objects, he found that search only worked because simple, robust rules had decided how each object was first described. The museums already have such rules in the Getty Vocabularies, which Bird says the commercial market has overlooked. One reader of his piece pointed out that scholars have had a classification of sale catalogues for a century, in Frits Lugt's Répertoire. What the market has never had is a commercial one.


Our monitoring has one clear case of what happens when a machine meets this problem. On 1stDibs, a machine-learning model placed its own price beside a dealer's on a Pissarro print listed at $28,000, and judged the print worth about five hundred dollars. The model could not tell a lifetime impression from a posthumous one. To the model, both were the same image under the same name. The difference that moves the value by an order of magnitude lived in the specialist's knowledge, not in the record. The platform withdrew its number, starting with the most expensive works.

That is the naming problem in practice. Where names are loose, a machine price is weakest exactly where values are highest.


What the account leaves out

Bird lists the contents of the Day Books in passing: stock numbers, client information, consignments, sales. It is worth reading that list again. It is not only a history of prices. It is a record of who sold, who bought, when, and on what terms, across two and a half centuries of private wealth.

Bird explains why the ledgers stay on paper in terms of habit: a market whose model is still "moored to the ancient strategy", fed by consignments that come from death, debt and divorce. He does not ask what digitising the Day Books would expose.


That question is ours, not his, and nothing in the record shows what the houses intend. But the trade runs on discretion. Consignors appear in catalogues as "a private collection". Buyers bid anonymously. Private sales are, by their nature, kept out of the published record. A searchable archive of clients reaching back to 1766 would be something to guard as much as something to sell.

Seen this way, the undigitised archive fits a pattern our monitoring has met all year. The houses stay silent about AI on the surface, while operational work goes on, unnamed, below it. What they keep closed may be less their technology than their ledger.


There is a further twist. In May 2026, Artprice published a manifesto arguing that without exhaustive archives there can be no lasting trust or fair pricing. It is a vendor's case for being treated as the market's archive. The most exhaustive archive Bird describes is in none of the vendors' databases.


Who wants the record

Look at who is asking for the record to be standardised, and why. Every voice faces outward, toward finance.


Kusin wants the major operators to hire people who combine numeracy, coding and connoisseurship, so they can compete with other wealth-management firms, such as private offices. He argues that financial intermediaries cannot apply their usual standards to measure risk and transaction costs in art, and he imagines art price futures that would trade continuously. He is also the founder of Kusin & Company, an art-economics firm. Standardised art data is his field and his business, and his case should be read with that in mind.

 

The state is asking too. A panel at The Art Business Conference announced UK SIC 47.921, a Standard Industrial Classification code for "activities of retail auctioneering houses for art works, antiques and antiquities", reported by The Art Newspaper in September 2026. Its purpose is to show government how large the auction sector is. Bird thinks lenders and analysts may use it later. It is unlikely to begin before 2031.


The commercial bid is the one Bird names directly: the merger of Artnet and Artsy, he writes, "may create a Bloomberg for art". In our record that merger was completed in April 2026 under Beowolff Capital, with predictive analytics stated as an aim. It remains a plan, not a product.


Meanwhile the bridge from art to finance is already carrying traffic. ARTDAI has published collecting-category indexes on the Bloomberg Terminal and LSEG since February 2026, according to its own releases. Addepar, the wealth-reporting platform, takes art valuations by API from Overstone, an art-finance firm founded by a former Christie's specialist. In November 2026, Artsignal, backed by Christie's Ventures, named art-backed lenders among its customers for the first time, without naming any of them.


And the one lender named anywhere in our monitoring, Citi Art Finance, values artworks with its own people, lends at around half of that value, and inspects each work in person when the loan begins and every year after.


The shallow ledger

All of this new financial layer stands on the same foundation: published auction results. Hammer prices, catalogue descriptions, sale dates. The shop window, not the Day Books.


So the data layer that finance is starting to rely on is forming on the thinnest, most public part of the record. The deep record — who consigned, who bought, the stock histories, the private sales — stays in the basement. Many of the people building the bridge come from the auction houses themselves. They know where the deep record is. What they can carry across is the shallow one.


This matches what our monitoring has seen crossing into finance all autumn. Value crosses. A price, a range, a risk figure. The history of an object travels only as far as it moves that number. Everything else that makes a work matter to the family that owns it has no field in a lender's file.


It also sets a limit on every AI valuation offered to a lender today. It rests on public prices, coarse naming, and no private sales. It is no surprise that the one named lender still sends its own people to look.


A market that has not moved

Bird reads the flat $60bn as the cost of a market that never built its data. That may be right.


Our record allows another reading alongside it. Much of the art market's value sits where no data reaches: in a ledger in a basement, in a specialist's eye, in a private sale that never enters the published record. AI has arrived at the edges. Finance is building at the surface. The core is untouched, and the core is where the market's most valuable knowledge sits.


"Futures in Rembrandt, anyone?" Bird asks at the end, half in jest. Futures would need a ledger. The one that matters is still on paper, and it has not yet been opened to any machine.


Founder & CEO of SMA Crown Confidential


Bespoke AI Intelligence for Private Decision-Makers


This article is part of an ongoing series by SMA Crown Confidential on the intersection of private wealth, cultural intelligence, and the future of bespoke AI.


Sources: this article takes as its starting point Rufus Bird, "Fine Art and Big Data", Ornament (Substack), 2 November 2026, first printed in The Critic, October 2026, which draws on David Kusin, Art as an Asset in the 21st Century. It also draws on The Art Newspaper (UK SIC 47.921, 11 September 2026) and SMA Crown Confidential's Art & Collectibles AI Deployment monitoring (June–October 2026) and Art & Collectibles AI — Use and Bridge monitoring (from 18 October 2026), including the reporting cited in "Where the Signature Sits" (13 November 2026): Artnet News and Business of Home (1stDibs), Cultured (Artsignal, 10 November 2026), Citi Art Finance (June 2026), ARTDAI and Artmarket.com releases (interested parties), and Addepar and Overstone materials. Figures and quotations should be confirmed against the primary sources before publication

Comments


bottom of page