Writing an AI RFP that gets the right vendor

Three ways to see it

1. 1

   An AI RFP has eight sections that matter. Background and goals: a one-page narrative of why this procurement exists. Functional requirements: what the system must do. Non-functional requirements: speed, scale, languages supported, accessibility. Data and security: residency, encryption, access. Compliance: regulatory frameworks the vendor must align with. Commercial: pricing model, payment milestones, performance bonds. Governance: review cadence, escalation, exit. Evaluation: scoring matrix with weights. Each section answers a question vendors will otherwise answer for you.

2. 2

   Way one to write the RFP: begin with the evaluation matrix and work backward. Most teams write the matrix at the end. Reverse it. If you know that 30 percent of the score is data sovereignty, 25 percent is technical capability, 20 percent is cost, 15 percent is local capacity, and 10 percent is references, every paragraph upstream is written to surface evidence for these dimensions. The vendor experience is also better; they know what to highlight.

3. 3

   Way two: read the MoITT NAIAI tender as a worked example. It is publicly available. Notice how it sequences capacity-building deliverables alongside delivery deliverables, how it specifies female participation, how it sets PWD accessibility expectations, how it stratifies risk, and how it leaves explicit room for sovereignty clauses. Copy what works for your context, drop what does not, and never copy what you cannot enforce.

The why-tree

Try this with Claude

AI-edge prompt to try: 'Acting as a senior procurement officer, draft an evaluation matrix for a PKR 200 million government AI chatbot tender. Include weights for technical capability, sovereignty, local capacity, female participation, PWD accessibility, references, and price. Sum to 100. Justify each weight.' Use it as a starter draft.