AI for Auction Houses: Consignments, Lots and Settlement Reminders
A practical guide to AI agents for auction houses: consignment intake, lot records, settlement reminders, and clear limits on valuations and payouts.

An AI agent for an auction house can take consignment intake requests, create and maintain lot records, and send settlement reminders to consignors and internal staff. It works from the house's own consignment terms, catalog rules and payout calendar. Valuations, reserve decisions and payout approvals stay with named specialists, who review the agent's drafts before anything reaches a client.
Key takeaways
- Give the agent intake, record keeping and reminders, and keep valuation, reserve and payout approval with named staff.
- Build one lot record that every other step reads from, so consignor, description, status and settlement stage never disagree.
- Write the consignment terms and approved wording into the agent's instructions before it drafts a single message.
- Use settlement reminders to request missing information, never to state an amount or promise a payment date.
- Judge the agent by the share of intakes that reach a specialist complete, and by how many consignor questions it closes without staff rework.
What an auction house agent should own
Auction work contains three kinds of decision. Some are administrative, such as confirming that a signed agreement is on file or that photographs were received. Some require expertise, such as estimating value, judging condition or setting a reserve. Some carry money, such as commissions, refunds and payouts. An agent is well suited to the first kind, can prepare material for the second, and should not decide the third. Drawing that line on the first day is more important than any feature list.
The useful first role is usually narrow: a consignment coordinator that handles intake questions, collects the documents named in the house's checklist, keeps lot records current and drafts reminders for staff to approve. Many houses already run this work by email, shared spreadsheets and phone notes. That existing trail is where the agent gets its inputs, and it is also the test of whether the agent is doing the job correctly.
How does consignment intake work step by step?
Intake is the place where a vague offer becomes a record the specialists can act on. A workable sequence looks like this:
- Receive the offer through the channel the house already uses, such as a web form, a general inbox or a call note, and open a new intake.
- Request the minimum facts: the owner's name, a description of each item, provenance documents, clear photographs and any earlier consignment history with the house.
- Check the agreement status: whether terms are signed, which commission schedule applies, and whether the owner has accepted current conditions.
- Create a provisional lot record and mark each missing field as missing rather than filling it with a guess.
- Route the intake to the right specialist with a short summary that separates what the owner said from what has been verified.
- Tell the owner what happens next and when a specialist will reply, using wording the house has approved.
Illustrative scenario: an owner sends photographs of six ceramic vases and a letter describing an estate origin. The agent opens the intake, notes that no signed agreement is on file, asks for the estate documents listed in the checklist, and passes the specialist a summary with the claims marked as unverified. It does not estimate value, does not comment on authenticity and does not suggest a sale date. The same intake pattern, with records and flags kept current, appears in car rental fleet work, where returns and damage notes need a single trusted record.
Lot records: one source that every step reads
The lot record is the spine of the system. A practical record holds the lot number, consignor reference, description, condition notes, attached documents, document status, the estimate range set by a specialist, reserve status, sale status and settlement stage. Each field needs one owner. The agent can draft descriptions and track document status. Specialists set estimates and reserves. Staff control any change to sale status. When a field has no owner, the agent should leave it blank and flag it.
Permissions matter as much as instructions. The OWASP Authorization Cheat Sheet makes the general point that access should be decided by the system, not by what a prompt asks for. Reserve values should be invisible to any identity whose role does not need them, including the agent's. Consignor documents also need care. A scanned letter or an email can contain text written to steer an assistant, so the agent should treat document contents as data to summarize, never as instructions to follow. The OWASP prompt injection guidance covers this risk in detail and is worth sharing with whoever writes the agent's rules.
Settlement reminders without promising a payout
Settlement is where trust is won or lost. The agent can follow the stage of each sale: closed, invoice drafted, consignor documents requested, payout scheduled by finance, payout confirmed. It can send reminders asking a consignor for missing bank details, tax forms or identity documents, and it can remind staff of items waiting for their approval.
The limits should be written down and tested:
- It must not state a payout amount, net figure or commission deduction to a consignor unless finance has released that figure for that sale.
- It must not promise a payment date that finance has not set.
- It must not create, edit or delete bank details or payout status.
Reminders also need to be sent once per event. Payment providers handle this with idempotency keys, so a retried request does not create a second transfer, as described in Stripe's documentation on idempotent requests. The same principle applies to reminders: each message should be tied to a settlement event and a key, so a retry does not send a consignor the same request three times. When a finance system confirms a payment, the agent should update its record only after verifying that confirmation, which is the purpose of signed webhook delivery in Stripe's webhook guide. Verification is an engineering task, so it belongs in the implementation plan, not in a prompt.
Which valuation and payout decisions need a named person?
A short decision guide keeps everyone on the same page. Write it into the agent's instructions and the permission settings, then review it whenever staff roles change.
- Valuation or estimate range: a specialist decides. The agent may attach comparable notes the specialist has supplied.
- Authenticity or condition conclusions: a specialist decides. The agent records what the owner reports and what documents show.
- Reserve prices: restricted to named staff. The agent does not see, repeat or infer them.
- Commission or fee changes: management and finance decide. The agent refers the question and does not answer it.
- Payout approval and bank detail changes: finance approves through its own process with a second check. The agent drafts and flags only.
- Disputes over sale results or refunds: a specialist handles them. The agent gathers the timeline and the documents already on file.
The NIST AI Risk Management Framework offers a practical structure for this work: name the risks, assign an owner to each, decide what must be measured, and record who approves changes. Marketplaces face similar questions about trust, disputes and seller records, which marketplace operations examines from a different angle.
How does Be Human run this in a managed engagement?
Be Human starts with one defined role, such as the consignment coordinator described above. The first engagement writes down the memory the agent needs (consignment terms, the intake checklist, approved wording), the tools it may call, the approval boundaries for each decision in the guide, and one success criterion that staff can observe, such as the number of intakes that reach specialist review with a complete lot record. Managed operations then cover monitoring of the agent's activity, recurring evaluations of its drafts against examples the house has approved, permission reviews when staff join or leave, incident response when something goes wrong, and controlled changes to instructions, one at a time, so the effect of each change is visible. The service is remote and available in English and Portuguese. If you want to find the first task in your auction house worth handing to an agent, take the free Be Human AI diagnostic below.
Frequently asked questions
Can an AI agent set or suggest estimates for lots? Not in a way that binds the house. Estimates depend on expertise, comparable sales and condition, and they should be set by a specialist who signs off on them. The agent can hold the specialist's estimate in the lot record and show it to the right staff, but it should not produce its own valuation for a consignor.
Does the agent need access to consignor bank details? No, and in most designs it should not have it. The agent can see whether bank details are on file and whether they have been verified by finance, which is enough to send a useful reminder. Creating, changing or reading the full details belongs to finance's own system and process.
What happens when a consignor disputes a settlement? The agent stops at the point of dispute and does not argue the figures. It gathers the sale record, the settlement timeline and the documents already on file, then passes them to the specialist or finance contact with a short summary. The consignor receives an acknowledgment with a named next step that staff have approved.
Which records should we clean up before starting? Start with the consignment agreements, the current commission schedule and the lot records for sales still open, because the agent will read from them every day. Fix duplicate consignor entries and agree on one status vocabulary for intake, cataloging, sale and settlement. Messy source records produce messy drafts, however carefully the agent is written.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


