Compare Vendor Proposals Against a Requirements Checklist

Compare vendor proposals against a requirements checklist by asking AI to build a source-linked table, then checking each claimed match against the proposal. Treat missing answers as unknown, not as proof that a vendor meets or fails a requirement.

A fixed checklist makes proposals comparable

Start with your own requirements checklist, not a summary of whichever proposal is easiest to read. Vendors may describe the same service in different terms, and a polished proposal can draw attention away from an unanswered requirement. Write each requirement as a separate row before asking AI to compare anything.

Make each row specific enough to test. Instead of asking whether a vendor offers good support, state the support hours, response arrangements, or contact method your team needs. Separate requirements that must be met from preferences that would be helpful. Keep pricing terms, delivery arrangements, security commitments, and contract exceptions distinct, even when a vendor discusses them together.

Gather the proposals, appendices, and any written clarifications you are allowed to consider. Give each file a clear name and note which version is current. Decide whether a clarification changes an earlier proposal or merely adds detail. Do not mix material from different versions without recording which one supplies the evidence.

Before uploading confidential proposals to an AI tool, check your workplace rules and the tool’s data-handling settings. The AI acceptable use policy checklist can help you identify what your team needs to approve. If uploading is not allowed, you can still use AI on a redacted extract or build the table manually.

The requirements matrix needs evidence and an unknown status

An AI requirements matrix from multiple proposals should show the requirement, each vendor’s status, and the source behind every status. A practical spreadsheet can give each vendor its own set of columns for status, evidence summary, proposal location, and reviewer notes. Keep the original requirement wording in a separate column so the comparison does not quietly change the question.

Use clear statuses such as met, not met, partial, and unknown. Mark a requirement met only when the proposal explicitly supports it. Use not met when the proposal explicitly rules it out or states a conflicting term. Use partial when the proposal covers only part of the requirement. Use unknown when you cannot find an answer. Silence is not a refusal, but it is not a commitment either.

For each conclusion, record the file name and the page, section, or other location that lets a colleague check it. Where your workplace permits it, add a link in your working spreadsheet to the saved proposal or relevant passage. A file link may not open at the exact page, so retain the location in text as well. If the tool cannot identify a reliable location, leave the reference unverified rather than inventing one. This table is an evidence record, not an automatic ranking.

A narrow prompt keeps ChatGPT focused on the checklist

You can compare vendor proposals with ChatGPT or another AI document assistant by supplying the checklist and the proposals you are permitted to share. Ask for a draft matrix, not a winner. If the files are difficult to read or too large for your chosen tool to handle reliably, provide relevant extracts in clearly identified batches and consolidate the results in your spreadsheet.

A useful prompt says: Create a comparison table using my requirement wording without changing it. For each vendor and requirement, choose met, not met, partial, or unknown. Base each status only on that vendor’s supplied documents. Give a short reason and identify the file and page or section supporting it. If the documents do not answer the requirement, write unknown. If evidence conflicts, flag the conflict rather than choosing an interpretation. Do not use general knowledge about the vendor to fill gaps.

Then ask the assistant to review its table for unsupported claims and missing references. Treat that second pass as another draft, not independent verification. Copy the results into a spreadsheet you control, preserving the original requirement and the source location. For a small set of short, clearly structured proposals, a manual spreadsheet may be simpler than an AI upload. AI is most useful when it helps you locate and organize evidence, not when it decides what a vendor must have meant.

Proposal summaries must preserve missing and conflicting answers

To summarize supplier proposals against a checklist, summarize what each proposal says about each requirement, including what it does not say. A general proposal summary can be useful for orientation, but it is a poor substitute for row-by-row checking. A statement that a vendor offers flexible service does not establish that it meets your required service hours.

Watch for answers buried in attachments, exceptions, or contract terms. An attractive feature description may be narrowed elsewhere by an eligibility condition or an extra approval step. Ask AI to identify both the supporting passage and any passage that appears to limit it. Keep conflicting passages together in the reviewer notes and mark the row for human review.

Distinguish missing evidence from a vague answer. Missing evidence means no relevant passage was found in the supplied material. A vague answer exists but does not establish the precise requirement. Both may remain unknown, but the follow-up question differs: ask a vendor to answer an omitted item, or to make an existing statement specific. If a proposal refers to an outside policy or an unsupplied appendix, record that dependency rather than treating the reference as evidence you have checked. The goal is a usable list of questions, not a confident-sounding summary.

Source checks turn a draft table into a decision aid

Check every row that could affect the decision against the original proposal before sharing the matrix. Open the recorded location, confirm that it belongs to the right vendor and document version, and read enough surrounding text to catch conditions. A page reference alone is not proof that the table interpreted the passage correctly.

Review hard requirements and unknowns first. If AI marks a requirement met, ask whether the cited passage actually commits to the full requirement rather than merely mentioning a related feature. If it marks a requirement not met, check for a direct contradiction. Correct the status and note why whenever the evidence supports a narrower reading. The AI output review checklist offers a broader way to check AI-generated work before others rely on it.

Ask vendors to clarify unresolved requirements in writing, using the same question for each vendor where possible. Add their replies as new evidence with a date or version identifier, then update the affected rows. If proposals or clarifications arrive as revised files, compare versions before replacing your references; the guide to comparing SOP versions describes a useful approach to checking changes without assuming the newer file altered everything. Keep the original matrix so colleagues can see what changed.

The final comparison separates evidence from judgment

Use the verified matrix to support a decision, but keep the decision separate from the evidence. A vendor can meet a requirement and still be a poor fit on a preference or an unresolved contract term. Conversely, an unknown row may simply mean you need an answer before comparing that vendor fairly.

Before discussing a recommendation, agree on which requirements are essential and who can accept an exception. Do not let AI silently assign importance based on how much space a vendor devotes to a topic. If your team uses a score, record the rule for scoring unknown and partial answers before applying it, and show the underlying statuses beside the result. A score should not hide a missing answer to an essential requirement.

Share the comparison table with access to the underlying proposals where permitted. Invite a colleague to challenge the most consequential statuses by opening their references, not by asking another AI tool whether the summary sounds right. Record vendor follow-ups, reviewer decisions, and any accepted exceptions in the same working file. The finished result is not a declaration that AI chose the best supplier. It is a traceable account of what each vendor stated, what remains unknown, and which judgments your team still needs to make.