A governed evidence-review tool for municipal procurement, grounded in AI governance research

Supplier Evidence Review

A significant share of a municipality's carbon footprint can sit in what it buys. This includes construction and capital projects, purchased goods and services, contracted services, and waste-related activities, many of which fall under Scope 3 emissions. Procurement becomes a meaningful climate lever when staff can verify which supplier claims are credible.

This tool uses a structured review process to help staff organize supplier evidence, flag gaps, and prepare follow-up questions. Final judgment stays with the reviewer.

You have an unfinished review.

Step 1 of 3

What are you reviewing?

Pick the closest category. This loads the sustainability claims worth checking for that kind of contract.

Have the proposal as a PDF? Drop it here to get a head start (optional)

The file stays on your computer and is never uploaded or sent anywhere. The tool reads the text on your device, suggests a category, and pre-highlights the claims it appears to touch, so you confirm rather than start from scratch. Works best on digital PDFs; scanned or image-only PDFs may not contain readable text.

Now: matching is by keyword, so confirm each highlight. When AI is live: a model reads the proposal and proposes the claims with their supporting text.

The company whose proposal you're reviewing.

The name of the tender or contract, for your records.

Step 2 of 3

Log what the supplier claimed

Click a common claim to add it, then answer two quick questions about the evidence. The tool suggests a status and drafts a follow-up question from fixed rules; you adjust both. Added claims are highlighted.

Common claims for this category. The leaf marks climate and emissions claims.

Step 3 of 3 · the result

Your evidence review

The picture at a glance, the follow-up questions to send, and a one-page brief for your file.

Send these to the supplier to clarify weak claims.

Compare suppliers

Each supplier you add appears here, side by side, on the same requirements. This shows evidence status only. It does not total, score, or rank, the decision stays with your team.

Build the ask

Use this before you publish an RFP. It gives plain wording to paste into the sustainability section, so suppliers know what evidence to bring. If you don't ask for evidence, there's nothing to review later.

How rigorous should the ask be?

Paste into your RFP

Where AI fits

This tool was built deterministic-first: today every suggestion comes from fixed, inspectable rules, not a model. The map below shows what each function does now, what AI would do once live, and what stays with the human reviewer either way. The boundary is the point: AI is introduced only where its output can be checked against a named source, and never where it would make the award decision.

Now · rule-based, no AI When AI is live Always human
Reading the proposal & finding claims
NowThe tool extracts the PDF text on your device and you click sentences to turn them into claims.
When liveA model reads the proposal and proposes the candidate claims and where each one sits in the document, so you confirm rather than hunt.
AlwaysYou confirm or reject every proposed claim. Nothing enters the review without your click.
Suggesting an evidence status
NowStatus comes from a fixed lookup: the evidence form and condition you select map to a status by rule.
When liveA model reads the actual evidence text and suggests a status with its reasoning, flagging ambiguity a fixed rule would miss.
AlwaysThe status dropdown is yours to set. The suggestion is a starting point, never the record.
Drafting the follow-up question
NowQuestions are selected from pre-written templates matched to the claim by keyword.
When liveA model drafts a question specific to the wording and gaps in this supplier's claim, which you edit before sending.
AlwaysYou edit and send. The tool drafts language; it never contacts a supplier.
The award decision
AlwaysNo AI now or later. The tool organizes evidence and surfaces gaps. It does not score, rank, or recommend an award. That decision, and its justification, stays with the evaluation team. This boundary does not move when AI goes live.

How it's governed

The controls that keep this tool a support for your judgement, not a replacement for it.