AI vendor adoption is not a question for the future. Staff across the organisation are using AI tools today, sometimes through sanctioned channels and sometimes through personal accounts. Personal preference is doing most of the deciding in IT leadership conversations across the sector, which is neither unusual nor a strategy.
At a Glance
Six engines dominate the conversation, with five more worth knowing about. Their jurisdictional positions, cost profiles, and risks vary significantly. A structured adoption pattern for a 300-person organisation typically runs £137,000 to £167,000 (€157,000 to €191,000) per year, around £456 to £556 per user, with material variation by user tier. This is the kind of document IT and Cyber Security functions should be putting in front of executive teams, and what executive teams should be asking for if they are not already receiving them.
Where We Are in May 2026
The vendor landscape shifted materially in April 2026: OpenAI's models arrived on AWS Bedrock, Microsoft and OpenAI dropped the exclusivity clause, Google rebranded Vertex AI to the Gemini Enterprise Agent Platform, and Mistral shipped Large 3. Most organisations finalised their 2026 AI strategy before any of this happened. The pace of change is the dominant feature of this market.
What a Structured Adoption Pattern Looks Like
The defensible pattern blends tools by user need rather than imposing one tool on everyone:
- Microsoft 365 Copilot for the productivity tier, where the data already lives in M365 and the integration value is real.
- Claude Team Premium for power users working on analytical material and bid drafting, where output quality directly affects win rates.
- ChatGPT Plus for users needing multimodal or agentic workflows.
- Mistral via La Plateforme reserved for workloads where European sovereignty is a hard constraint.
Underneath all of it: an AI governance forum, a properly resourced training programme, and a SharePoint oversharing audit completed before Copilot rolls out at scale.
The Questions That Matter
- Structured versus ad-hoc adoption. A structured approach reduces compliance exposure and delivers better outcomes, but requires deliberate effort to keep pace with vendor changes.
- Information governance before Copilot. Copilot surfaces documents users can access but never realised they had. A SharePoint estate fifteen years deep produces uncomfortable conversations until labelling and access controls are reviewed.
- An AI governance forum. Sign off on use cases by data classification, with the Data Protection Officer involved. Increasingly expected by major donors.
- Hidden cost categories. Training, governance, and shadow IT typically add 30-50% to subscription costs and rarely appear in vendor decks.
- Data classification for AI use. No tool should see commercial-confidential or beneficiary data without explicit classification and sign-off.
An Illustrative Cost Picture
The average smooths over real variation. Subscription-only costs are around £180 per year for Copilot-only users, £360 for Copilot plus ChatGPT users, and £1,512 for Copilot plus Claude Team Premium users. The Claude Team Premium seats carry most of the per-user cost weight and must justify it through measurable output quality.
Risks the Organisation Is Managing
Shadow IT and data exposure: staff use personal AI accounts when official tools do not fit, and commercial-confidential or beneficiary data can enter unreviewed inference paths. Donor and partner expectations: major donors increasingly ask about AI use in proposals and delivery, and several partner countries have data sovereignty positions that affect what tools can be used. Competitive position: competitors with structured AI adoption are producing better bids faster, and the gap compounds quickly. Pace of change: the landscape changes faster than annual planning cycles, so designing around capabilities and risk classes rather than product names is the only pattern that survives.
The engine choice is a small part of the picture. The components that must withstand regulatory or contractual scrutiny are the architecture, deployment patterns, data classification, and governance.
Closing
That work belongs to an AI governance forum, with executive visibility maintained through periodic briefings rather than annual project reviews. If your organisation does not yet have a briefing of this kind landing in front of the executive team, that is the conversation worth having.
