Procuring AI

  • draft
Warning This is not a comprehensive guide, but a starting point to support your AI journey. This is not intended to be any form of legal guidance or advice, and should not be used as a substitute for independent legal advice.

Modern AI systems, particularly with the rise of generative AI, have created new procurement challenges. Procuring modern AI systems often involves: 

  • ongoing API services rather than one-time software purchases  
  • sending data to third-party systems  
  • rapidly changing capabilities and pricing  
  • small number of frontier model providers  

Despite these unique characteristics, you must still: 

  • follow the Procurement Act 2023  
  • consider data protection and security from the start  
  • ensure fair competition (or document why you're using direct award)  
  • assess value for money  
  • document your procurement decisions 

Common myths:

  • "It's just a pilot so procurement rules don't apply"  
  • "Innovation means we can skip competitive process" 

Get commercial and legal advice early. AI procurement is not exempt from procurement law. 
See the Local Government Association's guidance on responsible AI procurement for building equality and data protection into your process.



Choose your procurement route

Based on your needs you have a range of options available to you:

  • If you wish to add specific technical expertise to your team temporarily, find out about Exchange schemes
  • If you wish to explore frontier technology where no proven solution exists, find out about Pro-bono pilot competitions
  • If you wish to select a supplier based on working prototypes they build during competition, find out about Competitive Flexible Procedure
  • If you wish to buy standard tools and solutions quickly from pre-approved suppliers, find out about Framework call-off
  • If you wish to procure when other routes do not fit or contract value requires full competitive tendering, find out about Standard procurement



Route: Exchange schemes

Secondments and fellowships bring specialist expertise from industry and academia into government. While not AI-specific, they're valuable when exploring how AI could address policy challenges.

How is this different from other routes?

Unlike procurement routes that deliver products or services, exchange schemes bring people into your team. You're building internal capability through knowledge transfer, not buying a solution.

Use this when: 

  • You need expert guidance to scope whether and how AI could help 
  • You need specialist expertise not currently available in your team 
  • You want to build in-house capability through skills transfer 
  • You need someone embedded in your team for 6-12 months 

Do not use this if: 

  • You need a delivered product, not knowledge transfer 
  • You're working to tight deadlines (setup can takes 3-6 months) 
  • You need a specific named individual (most schemes are competitive) 
Exchange schemes vary by organization and department. Check what's available to you.

What to expect: 

  • Timeline: 3-6 months to set up (including recruitment and security clearance) 
  • Duration: 6-12 months, part-time or full-time 

Example: Open-source AI Fellowship

In 2025, the Department for Science, Innovation and Technology (DSIT) launched a pilot funded by a $1 million Meta grant to the Alan Turing Institute to recruit open-source AI expertise. Read the press notice. The Institute recruited AI experts employed by UK organizations for 12-month secondments embedded in DSIT's Incubator for AI. 

Fellows are not affiliated with Meta and the programme is independently run by the Alan Turing Institute. Fellows work on government priorities including secure AI assistants and planning tools. All outputs are open-sourced. 

This is a one-year pilot. View the fellowship programme details.



Route: Pro-bono pilot competition

Pro-bono pilot competitions allow government to test frontier technology by having multiple companies build working prototypes before committing to paid contracts.

Companies work without payment in exchange for the opportunity to showcase their technology and engage directly with government teams. Participating does not guarantee future contracts. This approach lets government see what's technically possible before making significant financial commitments.

How is this different from other routes?

Unlike Competitive Flexible Procedure (a 2-6 week demonstration within a procurement), pro-bono pilots run for 12+ months to explore whether emerging technology can solve your problem. Multiple companies work simultaneously, and success doesn't guarantee future contracts. 

This is for exploring whether technology works at all, not buying known solutions.

Use this when:

  • You're exploring cutting-edge technology where no proven solutions exist
  • You want to see multiple working prototypes before committing significant budget
  • The technology could shape a major future procurement
  • You're willing to invest 12+ months in testing

Do not use this if:

  • You need delivery quickly (pilots take 12+ months)
  • The technology is mature and proven (use standard procurement)
  • You cannot commit to running paid procurement if pilots succeed

What to expect:

  • Timeline: 12-18 months for pilot, then separate procurement
  • Structure: Multiple companies build simultaneously, no guarantee of winning paid work
  • Your role: Provide specifications, data, and expert knowledge of the problem space

Example: GOV.UK Agentic AI Companion

In 2025, the Department for Science, Innovation and Technology (DSIT) launched a pilot to test whether agentic AI could help citizens navigate government services. Read the press release. AI companies applied to build prototypes for the GOV.UK Companion (£0 contract, 12 months) to demonstrate capabilities for major life events like re-entering the workforce. Read the notice for the 12 month pilot phase.

Government will only proceed to a national-scale paid procurement in 2026-27 if pilots prove the technology works. This approach manages risk while exploring frontier technology.



Route: Competitive Flexible Procedure

A multi-stage procurement process that lets you see working prototypes before choosing a supplier. This helps reduce risk by allowing you to select the winner based on proven capability, rather than written promises. The winner receives the contract to deliver the full solution.

How is this different from other routes?

Unlike pro-bono pilot competitions (which run for 12 months to explore emerging technology and future capabilities), this is a short demonstration of current capability as part of a single procurement process. One company wins the contract at the end.

Unlike standard procurement, you evaluate working demonstrations instead of choosing based on written proposals alone, reducing risk.

Use this when:

  • You want to reduce risk by seeing working solutions before selecting a winner
  • Technology is evolving rapidly and written descriptions will not show true capability
  • You're unsure which approach will work best and need to see different solutions demonstrated to make an informed choice

Do not use this if:

  • Requirements are routine and well-defined (use framework call-off)
  • You can confidently assess capability through written proposals (use standard procurement)
  • Building a meaningful prototype would take longer than 6 weeks

What to expect:

  • Timeline: 6-10 weeks from publishing the tender to awarding the contract (compared to 24+ weeks for standard procurement)
  • Structure: You publish requirements and receive written proposals. You shortlist several suppliers based on their proposals. Each shortlisted supplier builds a working prototype over 2-6 weeks at their own cost. Suppliers demonstrate what they've built. You award the contract to the winner.
  • Your role: Provide clear specifications, access to relevant data or systems for testing, and expert knowledge to evaluate the demonstrations

Example: AI for Augmented Planning Decisions

In 2025, the Ministry of Housing, Communities and Local Government (MHCLG) and the Department for Science, Innovation and Technology (DSIT) launched a procurement for an AI tool to speed up planning decisions, aiming to reduce processing time from 8+ weeks to 4 weeks. Read the tender notice.

The procurement includes a 2-week "Accelerator" phase where shortlisted bidders build working prototypes and demonstrate their solutions at a "demo day". This allows government to evaluate real capabilities across the full planning workflow (policy research, compliance assessment, and decision reasoning) rather than selecting based on written promises. The evaluation also rewards collaborative approaches, encouraging partnerships between frontier AI companies and UK SMEs (small and medium-sized enterprises).

The winner of the accelerator phase receives the contract to build and deploy the full solution across local planning authorities.



Route: Framework call-off

A way to buy commonly available products and services from pre-approved suppliers. Frameworks are agreements where suppliers have already competed, been vetted, and set their prices and terms.

How is this different from other routes?

You're buying from suppliers who've already competed and been approved. This is faster than running your own competition, but limited to what's already on the framework.

Use this when:

  • You know what product or service you need (for example, cloud hosting, AI software licenses, consultancy)
  • A framework already exists that covers what you need

Do not use this if:

What to expect:

  • Timeline: 2-12 weeks depending on method

  • Two ways to buy:

  • Direct award: Pick a supplier directly from the framework. Fastest option but not always available.

  • Mini-competition: Ask several framework suppliers to bid for your work.

  • Your role: Find the right framework, check how to buy from it, follow the process in the guidance

Accessing frontier AI models:

OpenAI, Anthropic, and Google DeepMind are not on frameworks. To access frontier AI models, buy cloud services from AWS, Azure, or GCP (who are on G-Cloud and resell AI model access).

Finding frameworks:

Search for frameworks on the Crown Commercial Service (CCS) website. Each framework has guidance explaining how to buy from it. Main AI frameworks are:

  • CCS AI DPS (RM6200): 219 suppliers offering services for AI discovery, consultancy, implementation, support. Mini-competition only. (DPS stands for Dynamic Purchasing System)
  • G-Cloud 14: 4,000+ suppliers offering cloud hosting, software, support including AI model access. Direct award or mini-competition.



Route: Standard procurement

A traditional competitive procurement where you publish requirements, suppliers submit written proposals, and you evaluate and award based on those submissions. This is the default route when other options do not fit.

How is this different from other routes?

You run a full competition from scratch, giving you complete control but taking significantly longer (24+ weeks vs 6-10 weeks for Competitive Flexible Procedure). You evaluate written proposals only, without seeing working demonstrations.

Use this when:

  • Your requirements are too specific or complex for existing frameworks
  • Your needs are well-defined
  • You are looking to procure an established technology

For AI procurement, these alternatives are often better:

What to expect:

  • Timeline: 24+ weeks from publishing tender to contract award
  • Process: Define requirements, publish tender notice, evaluate written proposals, conduct clarifications, score bids, award contract, observe standstill period

Your role: Design the entire procurement, draft all documents, evaluate all submissions, manage the process end-to-end

Back to top