The cheapest figure on a quotation can exclude the work that makes the project usable. One supplier includes installation, another provides equipment only, and a third leaves delivery unspecified. A purchasing assistant that ranks those totals immediately gives a precise answer to the wrong comparison.
The procurement owner should define scope gates before allowing recommendations. This article provides a proposed rubric for comparing quotations against an approved request. It prepares evidence and clarification questions; it does not authorise supplier selection, invent the cost of missing work or establish compliance with procurement requirements.
Convert approved requirements into comparison questions
Use the approved specification or request as the baseline, with a version and responsible owner. Break it into requirements a reviewer can check: item or service, quantity, required installation, delivery location, timing, supporting evidence and acceptance conditions. Do not let the first quotation define the scope for all others.
For each requirement, distinguish included, excluded, unclear and not evidenced. An absent sentence is not automatically an inclusion. A supplier may offer an alternative approach, but the owner must decide whether that alternative is acceptable before comparing its price as equivalent.
Keep discretionary preferences separate from mandatory gates. A helpful warranty option might inform a later judgement, while missing required installation blocks a like-for-like comparison. If the team uses weights or scores, the owner should approve them before evaluating offers rather than tuning them to favour a preferred result.
Preserve quotation wording and source references
Record the exact clause or table supporting each inclusion and exclusion. A short paraphrase can hide important qualifications such as installation by others or subject to site assessment. Preserve those qualifications and the quotation version in the evidence register.
OpenAI's file-search guide describes retrieval from uploaded documents with citations. This can help locate relevant scope statements, but does not prove that all requirements or pages were reviewed. Compare against an approved quotation inventory and record unreadable or missing sections. Source: OpenAI file search
A single supplier email may clarify a quote, but it needs a verified relationship to that quote and an owner-approved status. Do not overwrite the original offer silently. Link the clarification and record whether it changes the scope, price or validity period.
Scope-aware quotation rubric
Use this complete proposed rubric for a fictional equipment installation. The approved baseline requires ten specified units, installation, delivery to the agreed site and the requested acceptance evidence.
| Gate | Evidence to record | Pass disposition | Hold or exception disposition |
|---|---|---|---|
| Quote identity | Supplier reference, quote reference, version and valid offer status | One verified candidate offer | Conflicting versions or unknown supplier target remain unresolved |
| Item and quantity | Specified unit details and quantity with source location | Matches approved baseline | Alternative specification requires owner decision; missing quantity is held |
| Installation | Express inclusion and relevant qualifications | Required work included under approved assumptions | Equipment-only offer is incomplete for this baseline |
| Delivery | Destination, inclusion and any stated charge or condition | Approved location and treatment are explicit | Unspecified delivery generates a clarification question |
| Timing | Stated availability or delivery commitment, with status | Fits the owner-approved requirement | Tentative or conflicting timing is not treated as confirmed |
| Evidence and exclusions | Required supporting evidence plus every material exclusion | Requirements evidenced; exclusions accepted explicitly | Missing evidence or unaccepted exclusion blocks comparable status |
| Price comparison | Amount, currency and treatment approved for this common scope | Deterministic comparison of complete candidates | No invented allowance for omitted work or unknown charges |
Fictional comparison: Quote A states R9,000 for the ten units and excludes installation. Quote B states R11,500 for the units, installation and delivery under the approved baseline. Quote C states R10,000 including installation but leaves delivery unspecified. For this fixture, B is the only complete candidate at the current evidence stage. A and C remain incomplete or held; the table does not declare B the winning supplier.
Clarification queue: Ask A whether it can quote the required installation and state the resulting complete scope and price. Ask C to confirm delivery location, inclusion and any charge. Attach each question to its requirement and source reference. Draft questions remain unsent until the procurement team's communication process authorises the recipients and text.
Recommendation record: identify complete candidates, held candidates, accepted alternatives, unresolved requirements and the exact comparison rule. State why each candidate is in its group. Present a shortlist only after the owner confirms that the offers being compared satisfy the selected baseline; keep final purchasing approval separate.
Acceptance fixtures: one complete quote; a cheaper quote with an exclusion; a missing delivery statement; two versions with different quantities; and an unreadable evidence attachment. The complete quote can enter comparison, while the others produce the corresponding hold or clarification. A missing statement must never become included solely because the supplier's headline price is attractive.
The rubric is accepted when every required gate has source evidence or an explicit unresolved disposition. Reviewers should be able to reconstruct the grouping without relying on the assistant's overall confidence score.
Use structured extraction to expose uncertainty
OpenAI's Structured Outputs guide supports constrained fields and refusal handling. A quotation extraction can require separate requirement status, source reference, supplier wording and uncertainty. The schema does not guarantee that the text was interpreted correctly or that an omitted requirement is covered elsewhere. Source: OpenAI structured outputs
Check high-consequence fields against the source. Treat unreadable pages and extraction failures as coverage gaps. If the assistant cannot locate installation wording, preserve that as missing evidence rather than concluding either that installation is included or definitely excluded.
Avoid a single model-generated ranking that hides the gates. The team needs to see which scope differences prevent comparison before reviewing price. An explanation can summarise the validated table, but should not replace its evidence and disposition columns.
Work through normal, ambiguous and duplicate offers
In a hypothetical normal case, the quote explicitly matches the approved specification, quantity, installation and delivery requirement. The reviewer verifies the clauses and marks it comparable. Its amount can then be compared with other complete offers under the approved cost rules.
In an ambiguous case, the supplier says full supply but does not mention installation. That phrase alone does not establish the required work. The assistant drafts a targeted clarification and keeps the offer held. It does not fill the gap using common industry practice or a guessed installation allowance.
In a duplicate-version case, two PDFs share the quote number but contain different quantities. The assistant preserves both sources and asks the owner to confirm which version applies. It must not select whichever version yields the lowest price. Once the authoritative offer is identified, recompute the gates and mark the old comparison superseded.
Keep procurement actions under separate authority
OpenAI's function-calling guide separates a model's request from application execution and returned output. A recommendation or clarification draft is therefore not a purchase order or a sent supplier message. The application must enforce the team's actual authority for those actions. Source: OpenAI function calling
If an approved action is executed, record its target and outcome. Check existing clarification or purchase references before retrying an uncertain operation. The scope rubric should improve reviewability without quietly expanding into autonomous purchasing.
This article's decision is whether an offer is complete enough for a common-scope comparison. The separate cost-comparison problem concerns consistent treatment of tax, currency and delivery amounts after the required evidence is available. Keeping those questions distinct avoids hiding scope defects inside a calculated total.
FAQ about missing scope in supplier quotes
Should the agent always shortlist the lowest complete quote?
The owner defines the selection criteria, which may include factors beyond price. This rubric establishes comparable scope and evidence, then presents the approved comparison. It does not replace purchasing judgement or approval.
Can we add an estimated amount for excluded work?
Only through a separately approved estimating approach, clearly labelled and supported by its evidence. Do not invent an allowance and present the result as the supplier's complete quoted price. A clarification may be more appropriate.
What if the supplier offers a different solution?
Record it as an alternative and ask the scope owner whether it is acceptable. Compare it under an approved alternative baseline or keep it separate; do not silently redefine the requirements for every offer.
If your business needs help defining this process, explore Custom AI agents, the wider AI automation services, and our custom-agent workflow guide. The agents and automation comparison and custom AI agent glossary explain the terms. To discuss your records and approval rules, get in touch.