Use AI to Prepare a Product Substitution Review.

A supplier offers a replacement for a delayed fixture. The finish looks right, the description sounds similar, and the email says it is available sooner. Before the team accepts it, someone still has to compare the exact products and find out what the change affects.

Datum’s recommendation is to use AI to prepare that comparison for review. Give it the approved product record, the proposed replacement, and the project requirements. Ask for documented differences and unanswered questions. Approval stays with the people responsible for the design, installation, purchasing, and client decision.

Start with the exact product that was approved

A product family name is too broad for this task. Record the manufacturer, complete model number, finish code, selected options, intended location, and the document that established the approved choice. Keep the proposed replacement in a separate record with the same fields. If an option is missing, leave it unresolved.

Use dated manufacturer documents for the technical comparison and a dated supplier quote for the offer. Record which revision you supplied. A general product webpage may describe several configurations; it should not silently replace the specific configuration the client selected.

Compare against requirements, not just another brochure

Procore’s materials-management guidance recommends discussing substitutions early and coordinating purchase decisions around project scope, schedule, and budget. It also notes that purchase price alone does not capture installation and other costs. The article is established guidance, updated in 2025, rather than a new AI feature announcement.

Applied to an AI workflow, that means the project’s requirements belong beside both product records. Ask the reviewer to name the requirements that matter before the comparison starts. Depending on the product, these might include dimensions, connection details, required accessories, finish, service access, or documentation the project requires.

• Requirement: what the project needs, with a reference to the approved record or relevant document.
• Approved product: the documented value and its source location.
• Proposed product: the documented value and its source location, or “not found in supplied documents.”
• Difference or question: what the reviewer needs to resolve and which responsible role should check it.

Keep “not found” separate from “does not meet the requirement.” Missing evidence establishes a gap in the review, not a product defect. Likewise, matching a few listed values does not establish that the products are interchangeable.

Use citations to make the review checkable

Anthropic’s Citations feature links generated answers to passages in supplied documents. That is useful for a comparison where the reviewer needs to trace each reported value. A source link helps someone check the claim; it does not establish that the source applies to the selected model or that the comparison is complete.

For example, in a hypothetical bathroom remodel, a proposed faucet has the requested finish but the supplied documents do not establish compatibility with the already selected sink. The AI draft should identify the missing connection between those records and direct a question to the responsible designer or installer. It should not turn “same finish” into an approval recommendation.

Some decisive information may be in a drawing, footnote, or option table. Ask the workflow to identify pages it could not read. Have the reviewer inspect those sections in the original document, especially where a value depends on a diagram. Do not fill unreadable dimensions from memory.

Keep commercial questions and approval visible

A supplier’s availability statement needs its date and conditions attached. Record any quoted difference in product cost separately from unresolved labor, accessory, freight, or return costs. Do not present an incomplete comparison as the total cost of making the change.

The review should end with a recorded disposition: more information requested, rejected, or approved by the authorized person with any conditions attached. Client acceptance, technical review, and release to purchase may involve different people. The AI draft must not treat one person’s favorable comment as all of those decisions.

Try it on a completed substitution first

For a proposed trial, choose a past substitution with a documented outcome. Have the reviewer list the known differences before seeing the AI draft. Include a missing option code, a document for the wrong model, a detail shown only in a drawing, and a supplier statement that has expired. These are suggested test cases, not reported results.

Track missed differences, unsupported claims of equivalence, wrong source references, and the time spent checking the comparison. Compare the full review effort with the current process. Keep the trial read-only until the team has evidence that it produces a useful review packet.

• Preserve client approval in an AI selection log
• Prepare an RFI someone can answer
• Explore AI training for your team

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