Quick answer
Most AI RFPs are recycled ERP procurement templates with the word "AI" added — they generate proposals nobody can compare. A real AI RFP needs 7 sections (success criteria, eval-set ownership, tech-stack constraints, IP ownership, exit/data clauses, SLA structure, pricing model) and 11 specific vendor questions that filter for honesty. From 40+ enterprise AI engagements scoped between 2023-2026, the buyers who used this template selected vendors with 4x lower regret rate vs the IDC 2026 industry baseline. Free Word doc + Google Doc inside.
Why generic procurement templates fail for AI
A traditional software RFP asks "does your product have feature X". An AI RFP needs to ask "what is your eval methodology for feature X under real production traffic". The first question generates checkbox answers; the second generates honest vendor differentiation. We have read 40+ AI RFPs that were 80% boilerplate from a 2018 ERP procurement workflow — and they generated proposals nobody could compare against each other.
The 7 sections every AI RFP needs
1. Success criteria (measurable, not aspirational)
Bad: "improve customer support efficiency." Good: "deflect 50% of inbound tickets in the three categories: order status, returns, subscription changes. Maintain CSAT >4.4 (5-point scale) over a 90-day measurement window." The good version is testable; the bad version is not.
2. Eval-set ownership and methodology
Specify who owns the eval set (you do, full stop). Specify what the eval methodology looks like — labeled test cases, scoring rubric, regression thresholds. Vendors who cannot describe their eval methodology in two paragraphs are vibes-based AI shops.
3. Tech-stack constraints
Must-haves and must-not-haves. "Must use AWS Bedrock for any healthcare workload (HIPAA BAA required)." "Must not require sending PII to OpenAI's public API." "Preferred LLM tier: Claude Sonnet or higher quality equivalent."
4. IP ownership and exit data
All code, prompts, eval sets, and trained models belong to the buyer. Vendor delivers source code in a Git repository the buyer controls. Exit clause: buyer can cancel any time with 30-day notice; vendor delivers all work-product within 14 days; pro-rata refund of unused fees.
5. SLA structure
Uptime (99.5%, 99.9%, 99.95%?), response-time tiers by severity (P1 = 15 min, P2 = 4 hours, etc), accuracy floor (e.g. 92% on the eval set; vendor has 30 days to restore if it drops below). Specify what happens if vendor breaches — service credits, termination rights, dollar caps.
6. Audit evidence pipeline
For regulated industries: what logs are captured, retention period, format. "All AI tool calls logged to SIEM with prompt, response, model version, latency, confidence score. HIPAA retention 7 years. SOC 2 evidence available on request inside 48 hours."
7. Pricing structure
Fixed-price per phase, T&M for change orders, or hybrid. Specify what is included (eval set development? shadow mode? gated rollout? 90-day SLA?). Specify what triggers a change order and how it gets priced.
Want help drafting your specific RFP? 30-min call. We walk through your use case + draft the success criteria + eval methodology with you. Free, no sales sequence.
Get help with your RFP →The 11 vendor questions that filter the bullshit
- "Walk me through your eval methodology in two paragraphs."
- "How many of your last 5 engagements went live on the committed Phase 1 date? Specific number."
- "Can we see a SOC 2 Type II report under NDA before we sign?"
- "Who owns the code, prompts, and eval set at the end?"
- "What is the exit clause if we cancel at week 8?"
- "Show me one anonymized eval set you have built — even a screenshot."
- "What is your shadow-mode duration on a 50-user deployment?"
- "What is your typical model right-sizing approach (which Claude/GPT tier for what work)?"
- "Who will be on our daily standup — name and LinkedIn?"
- "What is your typical 90-day post-launch SLA? Show me a real SLA contract."
- "Why are you the wrong partner for this engagement? Tell me the honest answer."
Question 11 is the highest-signal question we have ever found. Vendors who answer it well (with a specific honest gap) tend to be the ones who deliver. Vendors who deflect or say "we are the right partner for everything" tend to be the ones buyers regret signing.
Download the template (no email gate)
Email info@braincuber.com with the subject "AI RFP template" — we send both the Word .docx and a Google Doc link you can copy. No email sequence, no follow-up sales nudges. If you want a 30-min walk-through of how to customize it for your specific use case, the link below is in our calendar.
FAQ
Can we send this RFP to Braincuber?
Yes — we respond to it the same as we would respond to any RFP. Question 11 ("why are you the wrong partner") is the one we expect you to ask. Our honest answer usually mentions specific scenarios we are not the best fit for (e.g. you need an on-site team in your city, or your use case is generic enough that off-shelf wins).
Is 11 vendor questions too many?
It is not. A bad vendor will ghost after Q4. A good vendor will answer all 11 in 30 minutes on the discovery call. The questions are filters; they take less time to ask than the questions you should be asking.
Do you have a procurement-friendly version?
Yes — the Google Doc has a tabbed procurement-friendly version with all of these as section headings in a standard RFP structure. Drop-in for most US corporate procurement workflows.
What about smaller engagements that do not need a formal RFP?
For sub-$50K engagements, skip the RFP. Use just the 6-point vendor checklist in a 30-min vendor call. Same coverage, less paperwork. Get the checklist →
RFP help, 48-hour turnaround
Send us your draft RFP, we sharpen it
Bring a rough RFP draft or just the use case. 30 minutes on a call. We draft the success criteria, the eval methodology section, and the 11 vendor questions tailored to your specific use case. You walk away with a sendable RFP. No sales sequence.
Related resources
Methodology
Template structure and the 11 vendor questions refined from 40+ enterprise AI RFPs scoped at Braincuber between 2023-2026 — both ones we won and ones we lost. The "4x lower regret rate" figure compares regret rate among 14 buyers who used the template against the IDC 2026 baseline (67% regret rate). Section coverage cross-validated against the Gartner 2026 AI Procurement Framework and the Forrester Total Economic Impact study on AI vendor selection.
Founder and CEO of Braincuber. Has scoped and shipped 500+ Odoo, AI, and cloud projects for US mid-market and global brands. Takes every founder call personally — no SDR layer between buyers and the people building the system.