ARTBnk

Applying the removal test to the product a buyer actually purchases settles this. What is sold is a value for a specific work in seconds, produced by image recognition and machine learning over thousands of data points, comparable sales and artist level market performance. Take the models out and a record of past auction results remains, which answers a different question entirely and leaves the buyer where they started, waiting on a commissioned appraisal. The curated database is the foundation the company insists on, and it is the input the models run over rather than the thing being bought.

Valuation is fully automated and instant by design, and the company positions the output as a standard benchmark for the market. What is missing is the boundary. Nothing published states where an automated value is sufficient and where a qualified appraisal is still required, which matters because lending, insurance placement, estate settlement and tax reporting each have their own expectations about who signs a valuation. No confidence interval, no flag for thinly traded artists and no escalation path to a human appraiser is described.

One methodological credential is real and worth recording: the valuation approach was developed with the academics behind a widely followed fine art index, incorporating artist level market performance over time, which is a stronger provenance than an unexplained proprietary algorithm. What is absent is measurement.

No error distribution against subsequent realised sales, no confidence bands, no backtest and no documentation for a lender's model validation function, despite realised auction prices making the error directly measurable. Compare the vendor in this index that has a reinsurer underwrite its valuation accuracy, which is what independent validation looks like on this axis.

The coverage claim is specific, artist data representing around eighty percent of the art market by dollar volume and thousands of data points behind each valuation, though it dates from the product's launch period and no updated figure was located. No bank, insurer or wealth manager is named as a customer, no valuation accuracy measure against realised sale prices is published, and no independent evaluation exists. A consultancy case study describes a deliberate shift from a consumer product to an enterprise offering, which speaks to strategy rather than to performance.

The question this product raises is whether valuations, holdings or transaction information submitted by one institution flow back into the shared database and the published indices that other clients rely on. That would improve the data and would also mean a bank's lending book informs a rival's valuations. Nothing published describes the boundary, and nothing states how the underlying auction data itself is licensed.

A collection inventory is unusually sensitive data, since it records what a named person owns and by implication where it is kept, which carries theft exposure on top of ordinary confidentiality. Nothing published states retention, access control, or whether client held collection data is separated from the market database. No privacy statement specific to institutional clients was located.

No certification, attestation or trust page was located. For a supplier holding collection inventories on behalf of private banks and insurers, an attestation is the standard procurement expectation and its absence from public material is a gap a buyer will have to close privately.

No licence is required to publish valuations and none is claimed, which is not penalised. The gap is a professional standard rather than a regulator: appraisals relied on for lending, insurance, estate and tax purposes are ordinarily expected to follow recognised appraisal practice standards, and nothing published states whether an automated valuation is offered as compliant with those standards, as an input to an appraiser who is, or as something outside that framework entirely. That is the first question a lender's credit policy will ask.

A model trained on auction results inherits the market's historical preferences, and in this market those preferences are documented and severe. Work by women artists and artists of colour has sold at persistent discounts, so a valuation engine fitted to that record reproduces the discount and then converts it into a smaller loan, a lower insured value and a lower estate figure for the people who own that work.

Thin trading compounds it, since artists with few auction records get weak comparables and the market's least liquid corners are exactly where valuation help is most needed. Nothing published addresses differential accuracy by artist demographic, region or sale volume.

An automated valuation that is too high produces an undercollateralised loan and an insurer exposed beyond its expectation, and one that is too low costs the owner borrowing capacity or settles an estate short. Nothing published describes a warranty, an accuracy commitment, a correction process or a route for an owner who disputes a value to have it reviewed. The contrast within this index is instructive, since another collateral valuation vendor warranties its numbers and pays when the asset sells for less, which converts an opinion into an instrument.

The database is described as proprietary, curated and normalised in house, which shortens the chain and is the company's central claim, but nothing states where the underlying auction records are sourced or licensed from, and auction results are compiled and sold by a small number of established data businesses. No model component, image recognition system or infrastructure provider is named either.

Delivery is a software platform with applications for art professionals and data services alongside, and one named external integration exists, a partnership with a private markets technology provider to support fractional investment in art and collectibles. No collateral management, loan origination, policy administration or portfolio system used by the institutional buyers is named, which is the connection that would matter for a lender running art secured facilities.

Described as a hosted software platform with data services and nothing further. No hosting region, tenancy model or residency commitment is published, which is a live question for European private banks and insurers holding client collection records.

No pricing, per valuation rate or subscription structure is published for the enterprise offering. This is a market where the incumbent alternative, a qualified appraisal, has a known and substantial cost per work, so a published rate would be a direct and powerful comparison and its absence is a missed argument as much as a disclosure gap.

The buyer list is enumerated across the financial functions that touch art, lending, insurance, wealth management, family offices and estates, with collectors, dealers and advisers alongside, and the company has explicitly restructured toward the institutional side. Held at B because every one of those relationships is described by category and none by name, which in a market that prizes discretion is understandable and still leaves the axis unproven.