Most public bodies already run Microsoft 365, so Microsoft 365 Copilot is often the assumed next step. Several things make that assumption feel settled:
- the licence can be added to an existing agreement
- the staff interface is familiar
- large UK trials have reported time savings
- NHS England has announced a rollout to 505,000 staff
For many organisations Copilot will be the right choice. It should still be a choice, made on evidence. Selecting an assistant automatically skips the questions that decide whether it will be safe and good value:
- which tasks it serves
- which content it can reach
- what it costs across the whole workforce
- how its answers perform on real work
AWS Quick, AWS's AI workspace, has been available in the London region since March 2026 and is a credible option for some public bodies. This article covers four things:
- why Microsoft is usually the default
- which assumptions are worth testing
- where each product is likely to be stronger
- how to run a fair comparison
It describes both products as their vendors document them on 20 November 2026. Inference Public works on AWS, and we say so. Our aim is that the decision is made on evidence.
Why Microsoft is the default
Microsoft's position in UK public services is built into how they buy technology, as well as how they use it:
- National pricing terms. The government's Strategic Partnership Arrangement with Microsoft (SPA24) is a five-year memorandum of understanding. It gives eligible UK public sector organisations access to cost savings across Microsoft's products. Microsoft says it includes, for the first time, Microsoft 365 Copilot.
- The competition regulator is examining the market. In its 2025 cloud services market investigation, the CMA found that Amazon and Microsoft have positions of significant market power in UK cloud services, and identified concerns about Microsoft's software licensing in cloud. On 16 October 2026 the CMA proposed to designate Microsoft as having strategic market status in respect of its Business Software Ecosystem. The CMA says a proposed designation does not find or assume wrongdoing, and a final decision is pending. Its stated aim is that customers can "mix-and-match AI software across suppliers to best suit their needs".
- Government guidance supports a deliberate choice. Where incumbent vendors are used, organisations should be aware of, and actively manage, vendor lock-in.
None of this makes Copilot the wrong choice. The CMA's findings cover AWS as well as Microsoft, and the right answer will depend on each organisation's work rather than on the market.
It does mean that adding an AI assistant because the licence is already on the agreement is different from choosing it on its merits. Where the Green Book and relevant business-case requirements apply, an AI investment should be appraised against credible alternatives, including the existing arrangement. The depth of appraisal should be proportionate to the decision, the expenditure and the risks.
Four assumptions worth testing
"We already have Microsoft 365, so we are ready for Copilot"
Copilot answers using data the user already has permission to access, so the accuracy of your SharePoint and OneDrive permissions decides what it can expose.
Microsoft is direct about the risk. Its deployment blueprint for Copilot starts with remediating oversharing, then setting guardrails and meeting regulations. The blueprint uses SharePoint Advanced Management, now included with Copilot licences, and controls such as Restricted Content Discovery to keep sensitive sites out of Copilot's reach.
Owning Microsoft 365 therefore does not mean being ready for Copilot. The same is true of Quick, which also enforces the permissions it finds. The readiness work on permissions, content ownership and withdrawal of superseded guidance is the same whichever product you choose, and it should come before the product decision.
"Copilot is cheaper because we already pay Microsoft"
Microsoft 365 Copilot
- Price. It is an add-on to an eligible Microsoft 365 subscription. The UK list price for enterprise customers at 20 November 2026 is £23.10 per user per month, paid yearly, excluding VAT. Some advanced capabilities are billed on usage through Copilot Credits.
- Copilot Chat. It is included with eligible plans at no additional cost. Copilot Chat users can also use agents that retrieve from specific organisational or connected sources, with consumption charges where they apply. Not every form of organisational grounding requires the full licence.
AWS Quick publishes per-user prices in US dollars (shown here at 20 November 2026), plus a $250 monthly infrastructure fee for each organisation account and charges for usage above each plan's allowances:
| Quick plan | Price per user per month | Included agent time |
|---|---|---|
| Essentials | $3 | 1 hour (3 hours until 31 January 2027) |
| Professional | $20 | 8 hours |
| Enterprise | $40 | 18 hours |
Essentials supports chat, research and Quick Flows. Advanced multi-step automation in Quick Automate, and authoring Quick Sight dashboards, require Enterprise.
Neither product is simply cheaper. Cost depends on three things:
- how many staff need deep use and how many need light use
- which capabilities each group actually needs
- usage above the included allowances
Model it per user group, at contract prices rather than list prices, before assuming either product is the economical choice.
"Our knowledge is all in Microsoft 365"
Often it is not. Guidance may live in Confluence, case data in line-of-business systems, and performance data in an analytics platform.
Both products can reach beyond Microsoft 365. Copilot can reference third-party tools and services through Microsoft Graph connectors or agents, and returns connector data only where the user has permission. Connectors use different models: some synchronise content into Microsoft 365, and others query the source at the time of the question. Each model has its own indexing, permission and administration behaviour. Quick builds knowledge bases over SharePoint, OneDrive, Confluence, S3 and other sources, and Quick Sight works with data held in AWS.
Spreading knowledge across several systems does not, by itself, favour either product. The question is which one covers your actual sources, with the permission behaviour, reporting and integrations you need, at an acceptable cost.
"The trials prove the value"
The UK trials are valuable, but they all tested Copilot, because Copilot is what most public bodies already had. Their results vary by population and method:
- The GDS cross-government trial reported a self-reported average of 26 minutes a day.
- HMRC's evaluation found about 60 minutes a week, which HMRC treats as a baseline estimate.
The trials say nothing about how Quick would have performed on the same tasks, and nothing about your organisation. The only way to know is to measure.
Where each product is likely to be stronger
Microsoft 365 Copilot
- Work inside Word, Outlook, Teams, Excel and PowerPoint. Copilot is embedded in those applications.
- Existing investment. Organisations that have invested in Microsoft Purview, sensitivity labelling, E5 compliance and Microsoft administration can build on it directly.
- NHS organisations covered by the national agreement, where licences are being allocated centrally.
- In-country AI processing. In an April 2026 update to its November 2025 announcement, Microsoft says local inferencing for Microsoft 365 Copilot is expected to become available in the UK by the end of 2026. We have not confirmed that it is generally available yet. Confirm it for your tenant, the features you need and your contractual commitments. In-country inference is also separate from where connected external data, optional services and some model choices are processed. Quick in London stores data in the UK, but processes AI requests within Europe and sends web search to Ireland.
- Organisations with little AWS presence, where a second platform would add operational burden.
AWS Quick
Quick's potential advantage is a separate AI workspace that combines research, knowledge, applications, workflow automation and AWS analytical capabilities. This depends on regional availability, permissions and licensing. In practice:
- Research across a large body of broadly readable material, through Quick Research. Knowledge bases with document-level permissions cannot currently be used by Research.
- Reporting and analysis over data already held in AWS, through Quick Sight, where that fits the reporting need.
- Workflows across AWS-hosted systems, through Quick Flows and, at Enterprise, Quick Automate.
- Workforces where many users need only light use, if Essentials covers what they need.
- A deliberate second option, where an organisation wants to manage its dependence on a single vendor for both productivity tools and AI.
An existing AWS landing zone helps, but it does not make Quick simple. Quick brings its own identity, knowledge base, encryption and operational decisions, several of which cannot be changed later.
The actual decision
The choice comes down to which product covers your required information and tasks with fewer unsupported assumptions, a lower total operating cost and stronger measured outcomes.
Questions that should decide it
| Question | Why it matters |
|---|---|
| What tasks, for which staff? | In-application drafting suits Copilot. Cross-source research, reporting over AWS data and AWS workflows may suit Quick |
| Which sources must it reach, and with what permission behaviour? | Coverage, indexing and permission checks differ by connector in both products |
| How accurate are permissions today? | Both products expose what users can already see. Remediation comes first either way |
| Where must data be stored and processed? | The products differ, and Microsoft's UK processing is a dated expectation that needs confirming |
| What will it cost per user group, at contract prices? | Licence tiers, included allowances, usage charges and connector costs change the comparison |
| How will records, FOI and retention work? | Both create prompts, outputs and logs that are recorded information |
| What skills and platforms do you already run? | A second platform has a real operating cost |
| How do the products perform on your own work? | Measured results are the only evidence that settles quality |
How to run a fair comparison
A bounded comparison settles the question faster than a long debate. Run it in three parts, and include your current way of working as a third comparator.
1. Shared knowledge tasks Use the same approved sources, users and questions for both products, with at least 20 real questions and agreed answers. Fix permissions and content first, so neither product is judged on the state of your SharePoint. Measure:
- correctness
- grounding and citations
- appropriate refusal when the answer is not in the sources
- access enforcement, including users who should be refused
- response time
2. Product-specific work Test what each product is designed for, within your organisation's agreed requirements:
- Copilot: drafting in Word and Outlook, Teams workflows, spreadsheet work and meeting follow-up.
- Quick: cross-source research, Spaces, Flows, supported connectors and any relevant analysis of data held in AWS.
Score these separately from the shared tasks. Never mark a product down for lacking a feature outside your agreed requirements.
3. Operational and commercial evaluation Compare:
- real subscription costs, agent usage and connector costs
- implementation effort and administrative burden
- records handling and processing location
- incident response
Measure output quality, correction effort and completed work as well as claimed time savings. Record the decision and the evidence, including why each alternative was not chosen.
None of this effort is wasted. The permission remediation, the question set and the evaluation method carry over to whichever product you choose.
Using Quick alongside Microsoft 365
For most public bodies the realistic question is whether AWS Quick adds value alongside Microsoft 365, rather than whether to leave Microsoft. Quick reads SharePoint and OneDrive and offers Microsoft 365 extensions, so staff can keep working in Microsoft 365 while Quick serves the specific jobs it is chosen for.
Some organisations will run both: Copilot for in-application work, and Quick for research or reporting over AWS-hosted data. That can be sensible, but it doubles some governance work:
- two identity integrations
- two sets of permission behaviour
- two records and retention regimes
- two cost lines
Decide deliberately which tool is used for what, and tell staff.
What does not change
Whichever product you choose, the quality and safety of answers depend on the content behind them and the controls around them:
- owned and current documents
- accurate permissions
- matching identities
- withdrawal of superseded guidance
- evaluation against real work
Our article on treating RAG as a data product explains why. The AWS Quick FAQ covers Quick's identity, encryption and data location decisions. Build RAG, use Bedrock Managed Knowledge Base, or adopt AWS Quick? covers the wider architectural choice, including purpose-built services for the public.
The free AWS Quick readiness checklist covers the groundwork, and most of it applies to Copilot as well. If you want help running a fair comparison as part of a Discovery, see AWS Quick Readiness, Implementation & Assurance.
Further reading
Strategic Partnership Arrangement 2024 (Government Commercial Agency)
CMA, actions on business software and cloud services
CMA, SMS investigation into Microsoft's business software ecosystem, roadmap (16 October 2026)
Microsoft, oversharing deployment blueprint
Microsoft, SharePoint Advanced Management for Copilot
Microsoft 365 Copilot UK enterprise pricing
Microsoft 365 Copilot privacy, data and connectors
Microsoft, in-country data processing for Copilot (April 2026 update)
GDS, Microsoft 365 Copilot cross-government findings
Microsoft, five-year agreement with UK government
