AI can help a small business draft, summarise, analyse and automate. But buying a tool is the easy part. The harder question is whether the business can give it reliable information, appropriate access and a useful job to do.
That matters as adoption gathers pace. The Government’s UK Business Data Survey 2025–26 found that 41% of businesses handling digitised data used AI for at least one purpose. Yet just 17% of AI-using businesses reported having an AI policy or guidance, and only 5% had a formal written policy.
For an SME, readiness is not a grand AI programme or a reason to delay indefinitely. It is a practical review of the business foundations that determine whether a tool will be useful, affordable and supportable. Start with the operating model, not the shopping list.
1. Start with a job worth doing
Before comparing tools, choose a specific business task that causes friction. Perhaps staff spend too long locating current procedures, preparing routine finance summaries or turning service notes into a first draft for review.
Describe the task as it works today: who does it, what information they use, where delays occur and what a better result would look like. Then set a measure. That could be minutes saved per case, fewer manual hand-offs, faster reporting or a reduction in avoidable rework.
This is the difference between a business case and an enthusiastic demo. The Government’s AI adoption research found that productivity is a leading motivation for adopting AI, but reported revenue effects were less immediate. A pilot should therefore measure outcomes the business can actually observe, not assume that a clever tool will automatically improve the bottom line.
Keep the first use case narrow enough to evaluate. If you cannot explain what success means, it will be difficult to judge whether the tool deserves a wider rollout.
2. Make the data usable before making it available
AI can help find patterns in business information, but it cannot quietly turn inconsistent records into dependable ones. Duplicate customer entries, missing fields, old price lists and several competing “final” spreadsheets remain business problems, now with a faster route to a confident-sounding answer.
Map the information needed for the chosen task. Identify its source, owner, update frequency and level of sensitivity. Check whether the same information appears in multiple systems and which version staff should trust. For an AI pilot, improving a small, relevant dataset is often more valuable than launching a costly clean-up of everything.
This is particularly important when a business’s finance and operational records are central to its decisions. A considered review of Microsoft Dynamics 365 Business Central can help establish whether finance, stock, purchasing and sales information is structured consistently enough to support better reporting and future AI use. A Dynamics 365 Jump Start should be about fit, process and readiness, not simply switching on another feature.
The same thinking applies where a business relies on older platforms. If IBM i or AS/400 data forms part of a critical workflow, check how it is owned, maintained and made available before designing an AI use case around it. This calls for specialist IBM i Management expertise, not an assumption that legacy means unusable.
3. Check access, records and responsibility
A useful AI tool may draw on material that employees can already access. That makes the quality of permissions important: outdated folders, broad access groups and records with unclear owners can all complicate a rollout.
Microsoft’s Copilot requirements guidance recommends establishing a secure, well-governed data foundation. It also explains that Microsoft Copilot uses the user’s existing permissions. Review access to relevant SharePoint, OneDrive and other Microsoft 365 content before enabling new experiences; permissions that were merely untidy can become much more visible when information is easier to search and summarise.
Set out who owns the pilot and who can approve a wider use. Keep a simple record of the tool, the intended task, the data it may use and the person responsible for reviewing the results. Decide how staff should raise an issue and what happens if the system produces incorrect or unexpected output.
The Government’s AI Management Essentials self-assessment is designed primarily for SMEs and start-ups. It offers a practical prompt for reviewing matters such as AI records, policies, data management, risk and issue reporting. It is a starting point for improving management practice, not a certification or a substitute for tailored advice.
4. Make the licence and the whole cost add up
AI tools are often sold as subscriptions, add-ons or features within products a business already uses. That does not make their total cost self-evident. Check which users need access, which work they will use it for, what existing subscriptions already provide and whether extra services or configuration are involved.
For Microsoft Copilot, licence requirements vary by experience: Microsoft’s current guidance distinguishes between a Copilot licence with a qualifying prerequisite licence and Copilot Chat, which is available to users with an eligible Microsoft 365 subscription. Confirm the current eligibility and terms for your tenant before buying. Avoid licensing everyone by default when a measured pilot can answer the business question first.
Include the less glamorous costs too: preparation, configuration, training, ongoing support and the time needed to check outputs. A Technology Expense Management review can help identify duplicate subscriptions, unused seats and costs that could be redirected towards a worthwhile pilot. The aim is to improve return on technology spending, not to add another line item because the sales presentation was persuasive.
5. Test the plumbing and the support model
A cloud-based AI experience relies on more than the application. Identity, devices, internet access, network configuration and other Microsoft 365 services can all affect whether it works as intended. Microsoft’s network guidance highlights required endpoints and connections, including *.cloud.microsoft, and provides connectivity tests for organisations preparing a deployment.
Check the actual locations and working patterns of staff. If people work across offices, shops, clinics or warehouses, test from those environments rather than only from the main office. Reliable Network Solutions, Cloud Solutions and Voice and Connectivity can support an AI-enabled working environment, but they are building blocks within a wider strategy, not an AI strategy in themselves.
Plan for disruption as well as normal use. If a cloud service or connection is unavailable, staff should know how to continue priority work. Review Disaster Recovery arrangements for the systems and information on which the pilot depends.
Finally, decide where users will go for help. A new tool will prompt questions about access, results and day-to-day use. A clear route to Helpdesk Support helps distinguish an application issue from a data, process or training issue.
6. Run a small pilot, then decide
A sensible pilot has a named owner, a limited group of users, an agreed use case and a review date. Record the starting point before introducing the tool. During the trial, collect feedback and note whether it saves time without reducing quality or creating new work elsewhere.
Set boundaries for the task: which information is appropriate, what a user must verify and which decisions remain with a person. For finance, healthcare or other sensitive work, get the relevant internal and professional advice before processing personal or otherwise restricted information.
At the end, decide whether to stop, adjust or expand. If the pilot works, document what changed and what further investment is justified. If it does not, the business has gained useful evidence without turning a small experiment into an expensive permanent fixture.
Strategy first, technology second
This is where a consultancy-first approach matters. The Fractal IT Director perspective looks across business goals, data, systems, people and costs before recommending an implementation. David’s practitioner background (including Google-certified IT project management, multi-national onboarding and IBM i system administration) supports a strategy grounded in real operational experience.
Traditional Managed IT Services, connectivity, hardware, software licensing and support can then be added as execution services under that strategy. They are useful bolt-ons; they are not substitutes for deciding what the business needs to achieve.
For SMEs considering AI, the first question is not “Which tool should we buy?” It is “What business problem are we solving, and are our information, permissions, costs and infrastructure ready to solve it well?” Answer that first, and the tools have a much better chance of earning their place.
For a broader view of the Evestaff group’s services, visit evestaff.co.uk.
Cloud Solutions, Disaster Recovery, Dynamics 365 Jump Start, Helpdesk Support, IBM i Management, Managed IT Services, Network Solutions, Technology Expense Management, Voice and Connectivity, AI adoption, SME AI readiness, Microsoft Copilot, AI governance, data quality, digital transformation.
Join The Discussion