Using AI to create a simple vendor and supplier comparison sheet before you sign any contract

Signing a vendor contract without comparing the fine print across all your quotes is how small businesses end up locked into bad terms for 24 months. This post walks you through a repeatable workflow using AI to compare supplier quotes — extracting, normalizing, and comparing data from messy vendor proposals. The approach works because tools like Claude and ChatGPT are capable of tabular reasoning — pulling specific data points from unstructured documents even when every vendor uses a different format.

You do not need a procurement department. You need 90 minutes and the right prompts.

What You Need Before You Use AI to Compare Supplier Quotes

Claude or ChatGPT: Either works for this workflow. Claude 3.7 Sonnet and GPT-4o both handle multi-document extraction well. Both have free tiers with usage limits; paid plans start around $20/month and give you larger context windows, which matters when uploading full contracts. Check current pricing before committing.

Google Sheets or Microsoft Excel: For building the final comparison matrix. Free with a Google account.

Your vendor proposals: PDF format preferred. You need at least two. Three or more is where this workflow pays off most.

Time required: 90 minutes to 3 hours, depending on how many vendors and how complex the contracts are.

Skill level: No coding required. Basic comfort copying text and working in spreadsheets is enough.

---

How to Prepare Your Vendor Quotes for AI Analysis

AI cannot read a scanned image well. Before you feed any document into Claude or ChatGPT, confirm it is a text-based PDF.

1. Open your PDF in any PDF viewer. Try selecting text with your cursor. If you can highlight words, the file is text-based.

2. Convert any scanned or image-based PDFs using Adobe Acrobat or Smallpdf. Run OCR to create a searchable text version. You should end up with a text-based PDF or Word document where text is selectable — either format works for re-upload, though PDF is preferred for consistency.

3. Rename each file with a clear vendor name before uploading. Use names like VendorA_Proposal.pdf rather than Scan_20240312.pdf. This matters when your AI output references source documents.

Getting clean text into the AI is the step most people skip. It is also the step most responsible for bad output.

---

Using AI to Extract Data from Messy Vendor Proposals

This is the core of the workflow. You are asking the AI to read multiple proposals and pull out the same data points from each — even though every vendor wrote their document differently.

1. Open Claude or ChatGPT in your browser and start a new conversation.

2. Upload all your vendor PDFs directly to the chat. Both Claude and ChatGPT accept file uploads on their paid tiers.

3. Paste this system prompt into the chat after uploading your files:

You are a procurement analyst. I have uploaded [number] vendor proposals. Extract the following data points from each proposal and return the results as a CSV table.

For each vendor, extract:

• Vendor name
• Base price or monthly cost
• Contract length
• Payment terms (e.g., Net-30, Net-60, upfront)
• Auto-renewal clause (yes/no — and if yes, notice period required)
• Price escalation clause (yes/no — and if yes, the mechanism, e.g., CPI-linked)
• Cancellation fee (amount or formula)
• Included services or deliverables
• SLA or performance guarantees
• Any terms that appear unusual or are worth flagging

If a data point is not present in a proposal, write "not stated." Do not infer or estimate values. Return the output as a CSV.

4. Copy the CSV output from the AI's response. You should see a comma-separated table with one row per vendor.

5. Paste the CSV into Google Sheets. Use File > Import > Paste data, or paste directly into cell A1 and use Data > Split text to columns. You should see a clean table with vendors in rows and contract terms in columns.

The reason you request CSV output specifically is that it forces the AI to normalize its formatting. Prose output is harder to compare side by side.

---

Building Your Vendor Decision Matrix with AI Help

A decision matrix removes gut feeling from the selection process. You assign a weight to each criterion, score each vendor against it, and multiply through. The vendor with the highest weighted score wins — on paper, at least.

1. Type this prompt into the same chat thread:

Based on the data you extracted, build a weighted decision matrix. Use the following criteria and weights:

• Price: weight 5
• Contract flexibility (cancellation terms, length): weight 4
• Payment terms (cash flow impact): weight 4
• Service scope (what's included): weight 3
• Risk indicators (auto-renewal, price escalators): weight 3

Score each vendor 1–5 on each criterion. Show your reasoning for each score. Calculate the weighted total for each vendor. Return results as a table.

2. Review each score and the reasoning behind it. You should see a table with raw scores, weights, weighted scores, and totals.

3. Adjust the weights in the prompt if they do not reflect your actual priorities. Resubmit and compare results. A business with tight cash flow should weight payment terms higher than a business with a healthy reserve.

Adjust the weights to match your situation. A 5-point weight on price makes sense for a commodity supplier. It makes less sense when you are buying specialized software where switching costs are high.

---

The Red Flag Check: Finding Hidden Contract Traps

Small businesses most often lose money on three specific contract terms: auto-renewal clauses, price escalators tied to CPI, and cancellation fees buried in definitions sections.

1. Paste this prompt into the chat:

Review all uploaded vendor proposals specifically for hidden or high-risk contract terms. Flag any of the following if present:

• Auto-renewal clauses with notice periods shorter than 60 days
• Price escalation tied to CPI or any external index
• Cancellation fees exceeding one month's contract value
• Liability caps that favor the vendor significantly
• Dispute resolution clauses that require arbitration in a different state or jurisdiction
• Undefined or vague performance metrics
• Terms that limit your right to switch vendors or use competitor services

For each flag, cite the specific clause and the page number or section reference. If you are uncertain whether a term qualifies, include it and explain why.

2. Read every flag the AI returns. You should see a numbered list of specific terms with source references.

3. Locate each flagged term in the original document manually. Confirm it exists and read the surrounding paragraph for context.

This step exists because AI models occasionally misread numeric values or misattribute a clause to the wrong vendor. You are not trusting the flag — you are using it as a search shortcut.

---

When Something Goes Wrong

The AI returns "not stated" for values that are clearly in the document. The most likely cause is that the PDF uploaded as an image rather than text. Re-export the document as a text-based PDF using Adobe Acrobat or Smallpdf and re-upload.

The AI assigns different scores to the same vendor across two separate matrix requests. AI models do not produce deterministic outputs. If you rerun a scoring prompt, scores will shift slightly. Run the matrix once, review it, and adjust criteria weights directly rather than regenerating until you like the result.

The AI combines data from two different vendors into one row. This happens when vendor documents have similar formatting or when filenames are ambiguous. Rename your files clearly, start a new chat, and re-upload with the vendor names explicitly stated in your opening prompt: "The three files I am uploading are from Vendor A, Vendor B, and Vendor C."

---

What to Do Next

Take the flagged red-flag terms from the AI's review and consider bringing them to a contract attorney before you sign — particularly for high-value or long-term agreements. The AI found the terms — a qualified human decides whether they are acceptable.

If you want to build this into a repeatable procurement process, read how to automate your vendor onboarding workflow with AI.

---

FAQ

Can I use AI to compare supplier quotes if the proposals are in different formats? Yes. Claude and ChatGPT are built for this. Their tabular reasoning capability lets them extract specific data points from unstructured text even when each vendor uses a different layout. The key is using a prompt that specifies exactly which data points to extract and instructs the AI to return "not stated" rather than guessing.

Is it safe to upload vendor contracts to ChatGPT or Claude? Check your organization's data policy first. OpenAI and Anthropic both offer options to turn off training data use. For sensitive commercial contracts, consider using the API with data processing agreements in place, or use a local model. Do not upload contracts containing customer personal data without confirming your compliance obligations first.

How accurate is AI at reading contract numbers? Accurate enough to be useful, not accurate enough to trust without verification. AI models can hallucinate numeric values — misreading a figure or transposing digits. Use the AI to identify where the numbers are, then verify the actual figures yourself against the source document before making any decision.

What is a weighted decision matrix and do I actually need one? A decision matrix is a scoring table where you rate each vendor against criteria you care about, then multiply each score by how much that criterion matters to you. It forces you to define your priorities before you see the results, which removes post-hoc rationalization from the decision. For any significant annual contract, it is worth the 20 minutes it takes. HBR covers the procurement decision-making framework in more depth here.

Can AI analyze how payment terms affect my cash flow? Current AI models support reasoning chains that go beyond comparison. You can ask Claude or ChatGPT to simulate the cash flow impact of Net-30 versus Net-60 payment terms given your average monthly revenue. It will not replace an accountant's analysis, but it produces a useful first-pass estimate you can take into that conversation.