A registrar entering a single 1970s gouache into an off-the-shelf online shop hits a wall inside two minutes. The template wants a size, a colour, a stock quantity. The work is one of one, on consignment from an estate, priced differently for a collector who saw it privately none of it fits.
That mismatch, not any appetite for novelty, is why galleries commission software rather than configure it. Shopify reported GMV of $115.57 billion in Q2 fiscal 2026, up 31.6%, credited partly to new AI commerce features momentum built on a product model designed for goods that can be reordered.
Key takeaways
- Shopify's Q2 FY2026 GMV hit $115.57 billion, up 31.6%, B2B GMV up 76% catalogue machinery built for wholesale.
- US Census Bureau data puts e-commerce at 17.1% of US retail sales, hard for the trade to treat as ancillary.
- Defensible AI uses in a gallery are clerical: metadata extraction, vocabulary mapping, draft condition reports, translation.
- Attribution and provenance are contractual warranties; a fluent machine-written error carries costs no efficiency gain offsets.
- Shopify Plus fees start near $2,300 a month, specialist builds near 80,000 a floor under who should commission one.
Why does standard commerce software break when the product is an edition of one?
Because template commerce assumes replenishment. Product records are built around variants and a stock number that can be topped up. A unique work has no variants, no reorder path, and a price that may legitimately differ between two buyers looking at the same page. Those are structural mismatches, not missing features.
Behind the price sit fields of their own consignment status, provenance chain, exhibition history closer to a registrar's model than a retailer's catalogue. Private-view pricing is the other break: showing a figure to one collector while withholding it from everyone else is a permissions problem, not a discount code.
What is AI integration for eCommerce, applied to art dealing?
AI integration for eCommerce is the practice of connecting machine-learning services to the systems that run an online shop catalogue, records, enquiries, logistics so that defined clerical tasks are performed or drafted automatically. In art dealing it functions as a back office technique applied to records that already exist, not as a sales channel.
The phrase usually means consumer recommendation engines. Where inventory sells once, "customers also bought" has nothing to compute the useful surface is internal.
Where does AI genuinely earn its place in a gallery's back office?
In four places, all of them clerical: extracting structured data from existing paper and PDF records; mapping inconsistent in-house terminology onto controlled vocabularies such as the Getty Art & Architecture Thesaurus; drafting condition reports for a conservator to correct; and producing first-pass translations of catalogue text for overseas buyers.
Forty years of index cards and exhibition catalogues are a genuine transcription problem. OCR combined with a language model produces candidate records artist matched against Getty ULAN, medium against AAT for a registrar to approve or reject. What is saved is typing, not judgement.
Translation is the other honest win: cross-border selling often stalls simply because nothing is written in the buyer's language. Terms of art must be locked, not translated freely "attributed to" rendered as a flat claim turns a hedge into a fact.
Why can't a checkout quote for a created painting?
Because the cost depends on decisions no cart can make: crate construction, whether the work travels flat or rolled, courier against consolidated freight, customs status, temporary import paperwork, and whether a technician installs at the far end. A weight-and dimensions rate table cannot approximate any of it.
The workable build is a quote-on-request path through a carrier service integration, routing the enquiry to Momart, Gander & White or a regional equivalent, holding the order pending. Fair-bound works need ATA Carnet or temporary admission paperwork. AI's contribution is small: parsing quotes into line items, flagging insurance thresholds.
How do consignment splits and resale rights end up in the code?
Because a sale is not one transaction. It allocates money between gallery and consignor on agreed terms, may trigger an artist's resale royalty depending on jurisdiction and sale value, and must reconcile against a signed consignment agreement. If the store cannot record any of that, somebody rekeys it into a spreadsheet after.
Resale right is the unforgiving example: introduced across the EU by Directive 2001/84/EC and collected in the UK by DACS, it applies to qualifying resales on a sliding scale, and liability turns on facts a template shop never captures. This is why serious custom eCommerce development begins at the ledger, not the storefront.
Why is the trade right to be wary of machine-written attribution?
Because attribution and provenance are warranties, not descriptions. A catalogue entry stating that a work is by an artist rather than after them is a representation a buyer can rely on and, if wrong, litigate over. A plausible-sounding generated sentence has no evidential basis and no accountable author behind it.
Language models are fluent by construction and confident by default the failure mode provenance research cannot absorb. Gaps in an ownership chain, particularly across 1933 to 1945, are meaningful and must be presented as gaps: a generated "private collection, Europe" is worse than a blank line. One wrong provenance line can cost decades of relationships against a saving measured in registrar hours.
Who should not commission a custom build?
Galleries turning over a handful of works a year, and anyone whose records are not yet in order. Software formalises an existing process; it does not supply one. If provenance currently lives across three inconsistent spreadsheets, an integration will propagate that inconsistency faster and more expensively.
The published numbers set a sensible threshold: Shopify Plus fees start near $2,300 a month, specialist builds nearer 80,000. For a dealer selling largely by enquiry, a constrained template plus proper records is the better first purchase catalogue hygiene, then integration, then automation.
Frequently asked questions
Does AI integration for eCommerce mean replacing our collection management system?
No. The records system stays the source of truth; integration synchronises a subset of fields to the catalogue and pushes enquiries and sales back.
Can AI write condition reports?
It can draft one from photographs and registrar notes, in house format, for a conservator to correct and sign. It cannot examine a surface, and no drafting tool substitutes for raking-light inspection.
How do we handle prices that only some collectors should see?
Through customer segmentation and gated catalogues, not hidden pages or discount codes. Access control at record level is a build decision taken early, not added later.
Figures cited are drawn from Shopify Inc.'s Q2 fiscal 2026 earnings disclosures, US Census Bureau retail e-commerce reporting, and published platform and agency pricing. Resale right references are to EU Directive 2001/84/EC and its UK implementation; this article is informational and is not legal advice.
