Guidelines for AI Procurement
OUR PICK FORA complete buying process
Guidance from preparation through evaluation, award and contract management.
Visit originalTurn assessment findings into sourcing and contract decisions.
Sources reviewed through
Read the buying checklist ↓Download the editable checklist (.md) ↓View 6 resourcesAI in procurement means applying AI to purchasing work: understanding spend, researching suppliers, preparing sourcing documents and checking transactions or contracts. Buying an AI product is a related but separate task. This page covers both: start here to scope an application, then use the buying checklist and original resources below to evaluate a provider. Choose a narrow problem with a measurable outcome before comparing tools.
In spend analysis, AI can suggest categories for purchase records, identify duplicate supplier names and flag unusual transactions for investigation. The category manager should compare suggestions against known examples and reconcile totals with the source system. In sourcing, an assistant can summarize supplier documents or draft questions from approved requirements. Each summary needs a source reference; a buyer must check missing qualifications and unsupported claims before using it in an evaluation.
Contract teams can use extraction tools to locate renewal dates, notice periods or clauses for review. Purchasing teams can use assistants to draft requisitions or explain an internal buying policy. Supplier monitoring can surface changes that warrant investigation. These are assistive uses, not proof that a tool can judge a supplier fairly or interpret a contract correctly. Practitioner context includes the CIPS report on AI in procurement and supply; availability and suitability still depend on your data and process.
Back to page sections ↑Do not allow an untested assistant to exclude suppliers, award contracts, change bank details or release payments. Those actions carry consequences that a fluent explanation cannot justify. Separate read-only research, draft recommendations and executable actions in the proposed scope. A human review step is meaningful only when the reviewer has the evidence, time and authority to challenge the output before an action occurs.
Keep confidential bids, negotiated prices, personal information and supplier banking data out of tools that have not been approved for those data categories. Check the actual account, deployment, retention settings and access permissions; approval for public research does not imply approval for confidential tender documents. Treat instructions embedded in supplier documents as untrusted content, and prevent extracted text from authorizing tool actions.
For supplier scoring, retain the published criteria, original evidence and reasons for any override. Investigate whether missing data or document format disadvantages some suppliers. If a reviewer cannot reproduce why a recommendation was made, use it as a research lead rather than a decision record. Our suggested boundary is to pause the affected activity when evidence or control is insufficient; your commercial, security and legal owners determine the applicable internal restrictions.
Back to page sections ↑Start with AI use-case intake: name the purchasing problem, process owner, affected suppliers, data and proposed level of autonomy. The procurement lead owns the business outcome; security and privacy reviewers examine access and data handling; the commercial owner checks service and contract terms. Use your existing authority matrix to name the person who can approve a limited pilot and the person who can authorize wider use.
Before the pilot, agree a small representative evaluation set, the manual baseline and acceptance criteria. For invoice classification, record both correct categories and costly mistakes, rather than reporting only an overall accuracy score. Include uncommon suppliers, incomplete documents and a case where the tool should decline. Measure review time as well as processing time, since an apparently faster draft may require extensive correction.
Keep one linked evidence record containing the intake ID, product and configuration, permitted data, test results, reviewer findings, unresolved gaps, decision and conditions. Use AI vendor assessment to distinguish supplier claims from demonstrated behavior. Record who accepts each remaining risk, when approval expires and what would trigger a pause. Add an approved pilot to the AI system inventory, with an owner and review date. A model, integration or autonomy change should return to the relevant reviewer before its expanded use.
Back to page sections ↑AI Governance Library is an independent AI procurement repository of selected guidance, practical records and original publisher links. The six resource cards below provide buying guidance and checklists; they are resources to compare, not a procurement portal for submitting bids or placing orders. We do not host a GPT procurement repository of models or endorse the suppliers mentioned by linked publishers.
Use the AI governance templates guide to choose a record, then the AI governance policy template to assign decision rights. Return to the checklist below when preparing requirements or resolving a buying question. Keep publisher guidance separate from your organization’s evidence and approvals, and verify current source terms before adapting materials.
Back to page sections ↑Written by AI Gov Library · Editorial guidance reviewed 3 November 2026. Source review dates describe our checks, not publisher updates. Suggested fields and decisions below are our own starting points; adapt them to your organization.
Download the editable checklist (.md) ↓
For the buying side of AI procurement, use this checklist to evaluate an AI system or service, including tools used by a purchasing team. Copy these fields into your buying brief. They are AI Gov Library suggestions, not a reproduction of any publisher checklist. Record the answer, evidence link, responsible person and unresolved action for each item.
A useful handoff is a one-page buying brief plus an evidence log, not a count of checked boxes. If a material question is unanswered, record an action and an owner instead of marking it complete.
Back to page sections ↑RECOMMENDED RESOURCES
Compare type, format and framework. Every link opens the original publisher.
OUR PICK FORA complete buying process
Guidance from preparation through evaluation, award and contract management.
Visit originalOUR PICK FORStructuring a procurement programme
A multi-part toolkit for planning and evaluating government AI purchases.
Visit originalOUR PICK FORA review checklist for a live purchase
Question-led workbook for applying AI procurement guidance to a specific purchase.
Visit originalOUR PICK FORA staged procurement workflow
Organizes AI buying around plan, source and manage stages with expected outputs.
Visit originalOUR PICK FORA compact procurement check
Checklist for buying AI products and services, including governance and ongoing oversight.
Visit originalOUR PICK FORPublic-service impact and data protection
Questions about equality and data protection for commissioning AI-based technology.
Visit originalNo resources match these filters. Clear filters to see the full collection.
The planning, evaluation and contracting steps specific to buying AI systems or services.
Buyers, commercial leads, counsel and service owners.
From requirements through contract award and ongoing supplier management.
AI procurement means buying an AI system or service, including AI embedded in an existing product. Start with a defined buying decision: what problem must be solved, who owns the outcome and what evidence would justify a purchase? For applications inside the purchasing function, start with the AI in procurement guidance above.
Our recommendation is to use one main process guide and add specialist questions only where your purchase needs them. Six overlapping documents should not become six mandatory assessments. The selections below are editorial judgments, not rankings issued by the publishers.
| Resource | Use it for | Produce | Watch for |
|---|---|---|---|
| UK Guidelines for AI Procurement | Preparing a public-sector buying process, from planning through ongoing management. | A buying plan with evaluation and governance checkpoints. | Published in 2020; check current local procurement rules separately. |
| WEF AI Procurement in a Box | Organizing a procurement programme and involving the right functions. | A shared plan for commercial, technical and governance reviewers. | A broader toolkit; select relevant parts rather than treating all material as mandatory. |
| WEF AI Procurement in a Box Workbook | Turning guidance into questions for a specific purchase. | A purchase-specific question and evidence record. | A worksheet cannot settle your acceptance criteria or contract terms. |
| NSW AI Procurement Essentials | Organizing work around plan, source and manage stages. | Stage outputs and named handoff owners. | Public-sector context may need adaptation for a private business. |
| Australian AI Procurement Checklist | A compact check of buying and oversight considerations. | A short list of unresolved procurement questions. | Review the original file and its context before adopting its fields. |
| Local Government Association: Responsibly Buying AI | Commissioning with equality and data-protection concerns in view. | Questions to take to service, privacy and legal reviewers. | This lens complements technical assessment; it does not replace it. |
Open the original-source cards for publisher links and formats. For a small team buying a support assistant, we would begin with a brief buying plan, then use the workbook or checklist to record gaps. For a public-service decision tool, we would also involve the service owner and specialist reviewers before evaluating suppliers.
Back to page sections ↑Example: A customer support team wants suggested replies, with staff approving every response. Its procurement record should include a test set of typical and difficult enquiries, handling of personal data, the ability to disable suggestions and a reviewer responsible for the pilot result. A polished demo alone would leave these buying questions unanswered. This is an illustrative scenario, not a completed assessment.
Back to page sections ↑Fictional teaching example: a proposed support drafting assistant. No evidence or approval is supplied.
| Question area | Answer / proposed condition | Evidence status | Owner / open action |
|---|---|---|---|
| Data retention | No customer data until retention and training use are reviewed. | Not verified; supplier documents requested. | Support Lead: obtain product-specific terms. |
| Human review | Propose staff review of every draft; no automatic sending. | Not tested; pilot evaluation plan required. | Support Lead: agree reviewer tasks and acceptance criteria. |
| Exit | Propose export, access removal and deletion verification. | Not demonstrated; contract and technical evidence missing. | IT owner: request an exit demonstration. |
No. Using AI in purchasing concerns tasks such as spend analysis or document review; buying AI concerns selecting and contracting a system. This page covers practical uses first, followed by resources for the buying decision.
Record the claim as unresolved, name an owner and request evidence for the actual product and configuration. Escalate a material gap before approval; a completed form is not evidence.
Adapt the record to your decision and existing responsibilities. Public bodies should also use the original official guidance and current local procurement rules. This checklist does not replace legal review.
Start with the framework and working example, assign authority in the policy, and keep the decision in your inventory.
Back to page sections ↑UK Guidelines for AI Procurement ↗ supports the public-sector procurement context and the sequence from planning to contract management. NIST AI RMF Playbook: Manage ↗ provides context for managing third-party AI risks. Our comparison recommendations, buying fields and support-assistant example are original editorial guidance. They are not legal requirements or an exhaustive buying standard.
For policy ownership, see the free AI governance policy template. For forms across the workflow, use the AI governance templates guide.
Back to page sections ↑OFFICIAL GUIDANCE
These sources explain the underlying expectations; the practical templates help teams apply them.
COMPARE RESOURCES
Government checklists are practical for process; longer guides cover evaluation and contract lifecycle in more depth.