Outcome and baseline
Name one workflow, owner, current measure, target decision, users, failure consequence, and exclusions.
Enterprise AI procurement tool
What should a Hong Kong enterprise put in an AI RFP?
Ask every bidder to answer the same evidence-bearing requirements. Define the operating outcome, data and authority boundaries, evaluation gates, production responsibilities, commercial assumptions, and exit evidence before comparing price or demonstrations.
By Onyx Devs Lab · Published · Updated
9 sectionsOne comparable requirement set
Evidence fieldsClaim, method, result, and limitation
Exit-readyOwnership, portability, and handover
Requirements that make proposals comparable
Describe the operating decision and proof burden rather than prescribing a fashionable model.
Data and authority
List systems of record, sensitive data, provenance, retention, cross-border movement, and separate read, recommend, draft, approve, and execute rights.
Evaluation and acceptance
Require common, edge, high-impact, missing-data, and adversarial cases; define metrics, mandatory gates, thresholds, samples, and evidence references.
Requirements that expose the delivery boundary
The proposal should reveal who operates the complete system after the demonstration.
Architecture and operations
Require integration design, dependencies, access control, audit fields, monitoring, fallback, recovery, latency, availability, and cost budgets.
Delivery and commercials
Name responsible people, subcontractors, milestones, acceptance ownership, assumptions, exclusions, change control, recurring costs, and options.
Handover and exit
Specify source and configuration ownership, documentation, data return and deletion, portability, knowledge transfer, and termination assistance.
How to use this RFP template
Issue one evidence brief
Give every bidder the same workflow, baseline, constraints, authority map, test cases, response schema, and deadline.
Score mandatory gates first
Reject or redesign proposals that cannot meet privacy, permission, safety, legal, evidence, or exit requirements.
Compare total delivery boundary
Evaluate team, integration, evaluation, operations, adoption, recurring costs, and handover—not only model output.
Contract the proof trail
Attach accepted requirements, tests, evidence references, limitations, signatories, and change process to the engagement.
Regulatory sources and scope
These are the primary regulator materials used by this implementation guide. This page is not legal advice.
Frequently asked questions
Should the RFP name a model vendor?
Only when a verified constraint requires it. Otherwise state quality, privacy, residency, latency, availability, portability, and cost requirements.
Can a polished proof of concept replace written evidence?
No. A demonstration can support one claim, but representative tests, controls, delivery ownership, operating evidence, and limitations remain necessary.
Is this legal or procurement advice?
No. Adapt this provider-authored template with legal, privacy, security, risk, procurement, finance, and sector specialists.
Start with one concrete operating problem.
We will first clarify the workflow, evidence, data boundary, and outcome standard before recommending the next step.
Contact Onyx