AI and Its Role in Collection Management

Introduction

In November 2024, the Zurich-based Germann Auction House became the first auction house to sell an artwork authenticated solely by artificial intelligence. It collaborated with the Swiss company Art Recognition, which authenticates artworks using AI. Germann offered three works from the collection of the late curator Martin Kunz — by Louise Bourgeois, Marianne von Werefkin, and Mimmo Paladino — accompanied by AI-generated authenticity certificates that replace traditional expert connoisseurship. The von Werefkin watercolour, which had no prior documentation of authenticity, sold for nearly double its high estimate. Carina Popovici, co-founder and CEO of Art Recognition, described the sale as a "pivotal test case" for the technology's capacity to influence real market transactions (Lawson-Tancred, The Art Newspaper, 31 October 2024).

This event is one of many that signal that artificial intelligence has moved from the margins of the art world – often mentioned but little understood - to its operational core. For collectors, whether they hold a few pieces or manage a substantial collection of artworks, understanding what AI can and cannot do now is part of responsible collecting practice. This article examines the practical applications of AI in collection management: how it is used for identification, valuation, cataloguing, provenance research, market analysis, and insurance, as well as its limitations.

What AI Actually Does

At its most basic, artificial intelligence in the context of collection management refers to software that can recognise patterns across large datasets, such as auction prices, images, museum catalogues, marketplace listings, and acquisition records, and return results that would take a human researcher considerably longer to produce. For collectors, the practical capabilities of these automated processes fall into three main areas.

Identification. Image recognition tools can analyse a photograph of an object and compare it against millions of reference images. Google Lens, widely available and free to use, can identify not only paintings or drawings but also pottery marks, silver hallmarks, and printed signatures in a short time. The app Magnus, founded in New York in 2013 and sometimes described as a "Shazam for the art world," allows users to photograph a work and is then able, so it claims, to retrieve the artist's name, title, exhibition history, and — uniquely among comparable platforms — prices from both auctions and galleries (magnus.net). Not to mention image apps that let potential art buyers visualise a work in their home, connecting realms of collecting and interior design.

Valuation. AI-driven pricing tools simultaneously scan live and historical auction results, online dealer listings, and other secondary-market platforms, providing a picture of what comparable objects are actually selling for rather than what sellers are asking. This kind of real-time market intelligence was previously available only to dealers, auction specialists, and well-connected private collectors, based on years or decades of experience and close, ongoing monitoring of the market. The democratisation of this data is one of the more significant shifts in the market over the past decade, although there are still questions, of course, about the unique traits of each artwork and how much these comparisons actually mean.

Practically, it means that entry-level collectors, for example, can search for artworks within their budget. Marek Claasen founded Limna (www.limna.ai), which promises to be an "AI-powered art advisor in your pocket".

However, while some AI tools offer direct support for budding collectors, others significantly support the strategic management and risk assessment of large high-end collections and provide important support for art professionals. Certain proprietary technologies can analyse large data sets to deliver insights into art market trends, facilitating up-to-date and ongoing risk assessments of collections and enabling data-driven collecting advice. Although these offer assurances that go beyond personal opinion and experience, they still depend on human interpretation and explanation. As AI-supported strategic collection advice becomes more essential, it will provide vital tools for risk analysis, helping financial providers and insurance companies serve their clients more effectively.

Authentication support. One area where AI has produced significant results is the authentication of artworks. AI can detect visual inconsistencies in an object's surface, construction, or markings that may indicate a fake, a misattribution, or a later reproduction. Art Recognition AG, a Swiss firm founded in 2019 by Dr Carina Popovici and Christiane Hoppe-Oehl, uses machine learning, computer vision, and deep neural networks trained on verified datasets of both authentic works and known forgeries. Art Recognition can assess the probability that a given work was made by a specific artist. In May 2024, the company identified counterfeit works attributed to Monet and Renoir being offered on eBay (https://art-recognition.com/). Its authentication certificates now carry sufficient market credibility to support auction sales, as demonstrated by the Germann case. A comparable company focused specifically on the high-end paintings market is Hephaestus Analytical, which combines AI pattern recognition with scientific material analysis. Both companies are clear that AI authentication is most valuable when used alongside, rather than instead of, physical examination and scholarly research (https://www.hephaestusanalytical.com/).

One important element of authenticity is provenance research. Provenance — the documented history of an object's ownership — has always been one of the most labour-intensive aspects of serious collecting. Tracing a work through successive auction appearances, dealer records, estate inventories, and exhibition catalogues can require weeks, months, or even years of correspondence and archival research. AI tools are significantly shortening this timeline. Machine learning systems can now simultaneously cross-reference digitised documentation and can return preliminary results in hours. The ability to surface provenance gaps more efficiently has made due diligence more accessible and, in some cases, has helped identify works with contested ownership histories before purchase rather than after. The Art Loss Register, which maintains the world's largest private database of stolen, looted, and missing cultural property, now offers AI-assisted search tools that can be queried prior to acquisition (for a fee).

These lead to practical applications on the following areas:

Market Intelligence and Price Analysis

AI-driven market analysis tools offer collectors something that was previously the preserve of auction house specialists and professional art advisors: structured, data-based insight into price trends and market behaviour.

Price-tracking platforms such as the Artnet Price Database and Artprice allow collectors to examine how a particular artist, maker, or collecting category has performed at auction over time. Alert systems across multiple platforms notify collectors when comparable objects appear at auction or in dealer inventories, helping them identify undervalued lots before wider attention inflates the price (www.artnet.com; www.artprice.com).

These tools are most useful when applied with knowledge of the field. Data on price development tends to originate from the upper end of the market, and conclusions drawn from blue-chip auction results do not always translate to the mid-market or specialist collecting categories. The deflation of the ultra-contemporary art market (also called the Wet Paint market) since 2024 is a reminder that algorithmic pattern recognition cannot substitute for an understanding of taste, critical reputation, and the longer cycles of collecting history.

Cataloguing and Collection Management

However, for many collectors, the practical challenge is not acquisition itself but the organisation of a collection. The lack of collection databases, even among serious collectors, and the consequences of inadequate documentation and missing information — for insurance, estate planning, and eventual sale — can be significant.

AI-supported collection management platforms have substantially reduced the friction of maintaining a proper catalogue. Photograph an object, and the software can automatically propose a title, category, date range, and relevant descriptive tags, drawing on reference databases specific to the collecting category. Platforms such as Catalogit are designed for general use across collecting categories; more specialised tools exist for coins, stamps, wine, watches, and books (https://www.catalogit.app/). Artlogic, widely used by galleries and private collectors, combines inventory management with CRM functionality and has become one of the more comprehensive all-in-one solutions (https://artlogic.net). Arte Generali, for example, uses professional tools such as ProCollect3 (licensed by SpeakART) and ProRisk (developed by SpeakART) to support client needs. ProCollect 3 streamlines the issuance of insurance certificates, the generation of inventories, and the management of condition and status reports within a comprehensive database platform. ProRisk, on the other hand, accelerates the production of insurance quotes, equally enabling smoother and better support.

The practical benefits of a well-maintained digital catalogue include:

• A searchable inventory filterable by medium, date, value, condition, and location with easy filters.
• Automatic generation of insurance schedules, updated as new acquisitions are added.
• Secure, cloud-based storage of provenance documents, condition reports, and purchase records.
• The ability to share a curated view of the collection with advisors, insurers, or prospective buyers without exposing the full database.
• Ongoing and up-to-date risk management.

What AI Cannot Do

A balanced assessment of AI in collection management requires clear acknowledgement of its limitations.

Image recognition is only as good as its training data. Objects from peripheral regional traditions, lesser-documented makers or artists, or categories underrepresented in digitised collections may return unreliable or absent results. Academic research has focused on the biases inherent in AI categories. Research from the University of Sheffield states that "AI is enmeshed with and maintains historic and ongoing colonial power relations, biases, and harms" (https://sites.google.com/sheffield.ac.uk/museums-ai-toolkit). A confidently stated incorrect identification is more dangerous than no result at all, particularly when it informs a purchasing decision. The unresolved dispute over the de Brécy Tondo Madonna — assessed as authentic Raphael by one AI system and as inauthentic by Art Recognition's — illustrates how different training datasets can produce contradictory conclusions from the same image (https://news.artnet.com/art-world/exhibition-of-ai-attributed-raphael-2341923).

AI has no physical engagement with objects. Experienced dealers and collectors have for hundreds of years described the tactile knowledge that underlies sound judgement — the weight of a piece of silver, the texture of a glaze, the feel of paper stock — and this dimension of connoisseurship remains beyond the reach of image-based tools. The leading British art historian Bendor Grosvenor, commenting on the Germann Auction House sale, noted that while "AI will play an increasingly important role in helping us to recognise who painted what, and when," the track record of AI attributions "is patchy, to say the least," and that the market still "prefers the judgment of academic research, the human eye, and technical analysis" (ARTnews, December 2024).

The data on which AI valuation tools are trained reflects past market behaviour, not present conditions. Markets move, tastes change, and the reputations of individual artists or makers can shift considerably over short periods. Valuation tools should be understood as one input among several, not as a substitute for current specialist advice.

Finally, meaningful questions about the regulatory and ethical dimensions of AI in the art market remain unresolved. The provenance of the data used to train these tools, the rights to the data, the accuracy of the results they return, and the accountability of the companies providing them are all areas in which due diligence by collectors is warranted.

Looking Ahead

The integration of AI into collection management practice is still at an early stage, and the pace of development is rapid. For collectors, the most important shift may be cultural rather than technological: the expectation that a well-managed collection will be properly documented, regularly valued, and actively monitored is becoming standard. The tools to do this more efficiently than ever before now exist and are affordable. The question is whether collectors and their advisors choose to use them.

Practical Recommendations

• Use AI identification and valuation tools as a starting point or a second opinion, not as a conclusion. They are most useful for initial research and for flagging questions to put to specialists.
• Adopt a digital collection management platform early. The cost of building a proper catalogue retrospectively is significantly higher than maintaining one from the outset.
• Before making significant purchases, conduct AI-assisted provenance searches using resources such as the Art Loss Register, in addition to traditional due diligence. Alternatively, ensure that you buy from a platform (such as an auction house or an art fair) that has its objects vetted.
• Treat AI-generated market data with the same critical eye you would apply to any single-source financial report. Understand what segment of the market the data reflects and whether it is applicable to your collecting category.

Further Resources

Platforms and Tools

Art Loss Register (stolen and looted art database): https://www.artlossregister.com

Art Recognition AG (AI authentication for paintings): https://www.art-recognition.com

Artclear (provenance fingerprinting and blockchain certification): https://www.artclear.io

Artlogic (collection management and CRM for galleries and collectors): https://artlogic.net

Artnet Price Database (auction price research): https://www.artnet.com/price-database

Artprice (market data and price indices): https://www.artprice.com

Catalogit (digital cataloguing for collectors and institutions): https://catalogit.app

Magnus (art identification and price transparency): https://www.magnus.net

PSA / Collectors (authentication and grading for sports cards, coins, and collectables): https://www.psacard.com

Publications

Deloitte Private and ArtTactic. Art & Finance Report 2025. Deloitte Luxembourg, 2025. https://www.deloitte.com/lu/en/services/consulting-financial/research/art-finance-report.html

Deloitte Private and ArtTactic. Art & Finance Report 2023. Deloitte Luxembourg, 2023. https://www.deloitte.com/content/dam/assets-zone2/lu/en/docs/services/financial-advisory/2023/art-finance-report-2023.pdf

Hiscox and ArtTactic. Hiscox Art and AI Report 2024. ArtTactic, September 2024. https://arttactic.com/reports/hiscox-art-and-ai-report-2024

Lawson-Tancred, Jo. 'A real leap of faith: Swiss auction house to offer works authenticated by AI'. The Art Newspaper, 31 October 2024. https://www.theartnewspaper.com/2024/10/31/a-real-leap-of-faith-swiss-auction-house-to-offer-works-authenticated-by-ai

Lawson-Tancred, Jo. 'Is the Art Market Ready for AI Authentication?'. Artnet News, 5 December 2024. https://news.artnet.com/market/first-a-i-authenticated-artwork-sells-big-at-auction-in-pivotal-test-case-2580910

McAndrew, Clare. The Art Basel and UBS Global Art Market Report 2024. Art Basel and UBS, 2024. https://www.artbasel.com/art-market