AI RFP template: 10 sections and the questions to ask

An RFP (request for proposal) for AI software should cover ten areas: features and performance, infrastructure, security and compliance, implementation and integration, onboarding, off-boarding, support, vendor stability, roadmap, and price. For AI tools, add questions on accuracy with your own data, human review of the output, and how the vendor uses your data. Copy the template below and swap the examples for your own workflow.

How to use this template

1. Write the context first. Who you are, the problem, the goal and the systems the tool has to work with.
2. Keep it short. Ask only what changes your decision. Long RFPs get long boilerplate back.
3. Ask open questions. "How do you..." gets you more than "Do you support...?"
4. Ask for proof. Request a demo on your data, a pilot result or a reference for each key claim.
5. Set the scoring weights before the answers come in.

Part 1: Context for vendors

• About us: size, industry, regions, and any regulated data you hold.
• Objective: what you want the software to do, in one or two sentences.
• Current problems: what is slow, costly or error-prone today.
• Priorities: for example ease of use, price, security or speed to value.
• Current stack: the systems the tool must connect to.
• Timeline and process: due date, pilot window, decision date and a contact.

Part 2: The ten sections

1. Solution features, usability and performance

• Describe how the product handles [workflow]. Show it on our sample data.
• What accuracy do you achieve on our samples, by field or case type? How did you measure it?
• How does a person review, correct or approve each AI output?
• What are the known limits? Where should we not use it?

2. Infrastructure, reliability and redundancy

• Where is the service hosted, and in which regions?
• What uptime do you commit to in the contract, and what are the remedies?
• Which models do you use? How will you tell us when a model changes?
• Can it run in our cloud, self-hosted or air-gapped? What stops working if it does?

3. Security, privacy and compliance

• How long do you keep our data, prompts and outputs?
• Does any model provider train on our data? Show us the contract terms.
• Which certifications and audit reports can you share?
• Do you keep a time-stamped log of user and agent actions? Can we export it?
• How do you govern AI risk? Map your answer to the NIST AI RMF functions: Govern, Map, Measure, Manage.

4. Implementation, customization and integration

• Which of our systems do you connect to natively? How deep does each connector go?
• Which APIs can we use? On what schedule does data sync?
• Who does the setup work, how long does it take and what does it need from our IT team?

5. Onboarding and training

• What training do admins and users get? Is it in the price?
• How long does a customer our size typically take to go live?

6. Future off-boarding

• How do we export our data, configuration and logs if we leave?
• In what format, and at what cost?
• When do you delete our data after the contract ends?

7. Customer support and service quality

• What support hours and response times are in the contract?
• Will we have a named contact?
• Give us two references at companies like ours.

8. Leadership and stability

• Who leads the company, and how is it funded?
• How many customers use the product in production today?

9. Vision, maintenance and roadmap

• What shipped in the last 12 months?
• What's on the roadmap for the next 12, and which items are committed?

10. Price, financing and key terms

• What drives the price: seats, usage or outcomes?
• What will year one and year three cost at our volume?
• What are the renewal caps, terms and notice periods?

Add market-specific questions

Generic questions miss what decides a deal in a given category. Airframe publishes the top buyer questions for each market it tracks at airframe.ai/markets, ordered "most decisive first, from pre-purchase questions and RFPs." Three examples:

AI Audit Agent (37 vendors tracked, 486 case studies read, Oct 2026)

• "How do you test controls and track pass and fail results?"
• "How do you check full ledgers with AI and flag odd posts?"

FP&A (17 vendors tracked, 922 case studies read, Oct 2026)

• "How do you link driver moves to profit, cash, and balance results?"
• "How do you perform on large models with many rows, users, and complex rules?"

AI Code Review (20 vendors tracked, 130 case studies read, Aug 2026)

• "Which Git platforms and self-managed versions do you support, what permissions do you request, and how do you handle a monorepo?"
• "How do you build and refresh whole-codebase context, and can you show a cross-file defect that a diff-only review missed?"

Score the responses

Score each section from 1 to 5 and multiply by its weight. Write down the evidence behind each score. The weights below are an example; set your own before the RFP goes out.

Questions about AI RFPs

What is an AI RFP?

A request for proposal that asks software vendors how their AI product would meet your needs. It adds questions on accuracy, human review and data use to a standard software RFP.

What should an RFP for AI software include?

Context about your company and goal, then sections on features, infrastructure, security, integration, onboarding, off-boarding, support, vendor stability, roadmap and price.

What AI-specific questions belong in an RFP?

Accuracy on your own samples, how people review output, whether model providers train on your data, how model changes are announced, and whether actions are logged.

Where can I find RFP questions for a specific software category?

Airframe's market pages at airframe.ai/markets list the top buyer questions for each market it tracks, such as AI audit agents, FP&A and AP automation.

How many questions should an RFP have?

As few as you can. If an answer wouldn't change your decision, leave the question out.