Check Model Numbers Before AI Compares Specs.

A showroom salesperson has a client’s product selection, a supplier quote, and a manufacturer specification sheet. Before asking AI to compare dimensions or features, check whether those records describe the same model. A tidy comparison can hide a mismatch in the documents it started from.

Datum recommends a separate model-number check at the start of product research. Ask the assistant to show the identifier in each source, preserve every suffix, and flag any match it cannot establish. A person confirms which documents belong together before the specifications enter a selection record or an order.

Give the assistant a specific product to find

Start with the selected manufacturer and the complete model identifier from the project record. Include the finish, size, configuration, or other option already recorded. Keep missing fields visible. If the client supplied only a photograph or a collection name, the task is to prepare a question for the salesperson.

Gather the relevant manufacturer documents and supplier records in a company-approved workspace. Record their URLs or file locations and any revision dates shown. Keep the original selection beside them so a suggested alternative cannot quietly replace the requested product.

Make each match traceable

Anthropic’s current Citations documentation describes answers linked to passages in supplied documents. For PDFs, those citations identify page ranges. It also states that image citations are not supported. That capability can help someone check where a model number or specification came from; it does not establish that separate documents refer to the same product.

For this workflow, ask for a short evidence record for each document:

• Requested product: the manufacturer and complete identifier in the project selection.
• Document identity: title, source location, and revision date when provided.
• Identifier found: the exact model text and the page or passage where it appears.
• Applicability: evidence that the document covers the requested variant, including any table heading or footnote that limits a specification.
• Open question: the mismatch, missing suffix, unreadable text, or absent document that needs a person’s attention.

If a sheet covers several models, keep each specification attached to the correct row and column. Ask the assistant to preserve the original unit and label beside any extracted value. A nominal product size and an installation opening have different meanings even when their numbers look similar.

Test the suffix before trusting the summary

Consider a hypothetical showroom packet with invented model identifiers. The client’s selection names MODEL-A-L, while the supplier quote names MODEL-A-R. The brochure heading says MODEL-A and shows both variants. An assistant that shortens every identifier to MODEL-A could report that the packet matches.

The useful output preserves all three identifiers and asks what the suffixes mean. It should not assume they indicate handing, finish, or another option. The reviewer checks the manufacturer’s explanation or asks the supplier, then records whether the quoted variant matches the client’s selection.

If the quote is correct and the selection record is incomplete, update that record through the normal approval process. If the quote is wrong, request a correction. Keep the unresolved mismatch open until the evidence supports a decision.

Keep availability and compatibility as separate checks

Finding the right specification sheet does not establish that a product is available to order. Record a supplier’s availability statement with its date and any conditions. Likewise, matching model identifiers only establishes which documents to review; installation compatibility still needs the relevant requirements and project conditions checked by the responsible person.

The AI draft should stop at the unanswered question. A related model may be worth considering, but it belongs in a proposed substitution review with its own approval.

Try a packet your team has already checked

Use a past product selection with a known, verified model match. Include a similar model’s sheet and a family brochure so the trial tests document selection as well as extraction. Have the reviewer compare the AI record with the originals.

Record false matches, missed differences, and the time needed to verify the output. Pay particular attention to a confident match built from an incomplete identifier. The trial should show whether the assistant makes the review easier without concealing the question the salesperson still needs to answer.

Bring a redacted product packet and the selection record your team uses to a conversation with Adam. Datum’s Training work helps your staff build and run tools for a defined task, including the checks needed before AI output becomes an instruction to purchase.

• Prepare a product substitution review
• Review supplier order acknowledgments
• Explore Datum Training
• Book a Conversation

Next step, if this note maps to a problem on your desk: book a conversation with Adam. Thirty minutes, weekday evenings, Eastern time.