How Art Collector IQ Values Art

Every valuation and bid recommendation is grounded in a proprietary database of five years of secondary-market auction results from Christie's, Sotheby's, Phillips, and Bonhams — hammer prices, estimates, dimensions, editions, sale dates, and lot details, across 33,000+ artists.

Two rules we never break: bought-in (unsold) lots are never counted as prices, and missing data is never faked — where estimates don't exist, estimate-based metrics say so instead of inventing a number.

Every number you see — which comparables qualify, price per square centimetre, the quartiles, the fair market value band, the all-in cost of a bid — is computed in code against the database, not generated by a language model. The AI writes the reasoning on top and is instructed to cite those numbers, never to derive its own. If the AI ever disagreed with the database, the database would win.

Comparable sales are tiered deterministically. Key comparables — sales of the same work or title, and the closest sales by subject, medium, period, and size — set the valuation range. Market context lots inform statistics like sell-through and the series backdrop, but never set the range. Both tables appear in every report, each sale linked to its auction record, so the band is auditable.

For editioned works, comparables are matched on the details line — signed versus unsigned, edition size, publisher — because the same image can trade five figures apart on those attributes alone.

Estimated FMV range — the band within which a willing buyer and a willing seller would most likely agree, anchored on the key comparables.

Most likely value (point estimate) — the single most probable price inside that range; the median of the anchor sales.

Confidence — a data-quality score built from sample size, sell-through, comparability, and price consistency. It describes the evidence, not a guarantee. With fewer than three comparables we won't issue a signal at all.

Three jobs, none of them arithmetic: a vision model reads your photograph — signature, edition number, condition, margins — and reconciles what it sees against what you declared, saying honestly when something isn't legible; live web research checks whether the exact work has surfaced before and what moved the artist's market this month; and a writing model turns the computed numbers into reasoning a collector can act on.

The database knows what sold and for how much. It cannot see condition beyond a photograph, it cannot authenticate, and it doesn't know that a particular provenance makes a particular buyer pay up. Owner declarations are treated as unverified claims, and reports say so. Art Collector IQ is decision support — for insurance, estate, or donation purposes you need a licensed appraiser with the object in hand. We make that person faster, not redundant.

• No black-box pricing — every band ships with the sales that set it.
• No authentication or condition guarantees from a photograph.
• No smoothed index numbers presented as the price of your specific work.
• No AI-derived figures — statistics come from the database, always.