A 12-person consultancy I spoke to recently had done nearly everything right. They’d demoed several tools, picked a well-reviewed AI assistant, and ran a briefing session for the whole team. Three months in, the usage data told a different story: two people used it daily, eight had never logged in. The tool worked well. The fit did not.
Choosing an AI tool for your team is less a technology decision than a workflow decision. Get the match right and adoption follows almost without effort. Get it wrong and you’ll spend months coaxing people back into something that genuinely works but sits outside their day.
What choice are you actually facing?
UK SMEs that see genuine results from AI tend to run a compact stack: one general assistant for drafting and analysis, one AI layer inside the productivity suite they already use, and one tool built for their specific sector or function. The first question worth asking is where friction actually lives in your team’s day, not which AI model scores highest on capability benchmarks.
The UK government’s DSIT guidance on AI for SMEs makes this explicit: start from a business problem, then look for the tool that addresses it. Buying a platform and hoping adoption follows tends to become a lengthy IT project where you wanted a quick win.
Advisory work with UK SMEs points to two factors that consistently predict whether a tool will be used: it fits into a workflow the team already runs, and staff can see a result within their first week. When both are present the rollout largely takes care of itself. When neither is present, no upgrade will fix the gap.
When are off-the-shelf AI tools the right starting point?
Off-the-shelf AI assistants suit teams doing text-heavy, repeatable work: drafting proposals, writing meeting summaries, generating first drafts of customer communications. They set up quickly, need no IT project, and at £25–100 per user per month they’re cheap enough to pilot and discard if the fit isn’t right. Advisory data suggests around 44% of UK SMEs find this path sufficient before they need anything more complex.
This path works when your team runs common stacks (Microsoft 365, Google Workspace, Slack, HubSpot) and can connect tools via a browser extension; when the biggest time drains are emails, proposals, meeting notes, and similar documents; and when you want results this quarter rather than after a lengthy IT project.
The main adoption risk with standalone tools is fragmentation. If every team member adopts a different AI app, you end up with multiple data-handling arrangements and no admin controls. The NCSC recommends integrating AI usage into your existing access management and device controls wherever possible. In practice, that means approving one or two tools centrally, being clear about what is not approved, and backing the policy with short practical training rather than a dense acceptable-use document.
The working benchmark from UK SME advisory practice: a tool worth rolling out should save at least two hours per user per week on a specific task, and a pilot of two or three people should verify that within a 60–90 day window.
When does platform-integrated AI make more sense?
If your team spends their day inside Microsoft 365 or Google Workspace, a standalone AI tool means asking them to open something else every time they need it. Platform-integrated AI like Microsoft 365 Copilot or Google Workspace Gemini turns up inside Outlook, Teams, Docs, and Sheets, where the work already happens. For firms handling client data in regulated sectors, the enterprise-grade audit trails and access controls carry real compliance weight.
This path suits teams already standardised on one productivity suite who want AI to work across email, documents, spreadsheets, and meetings without learning anything new. Microsoft 365 Copilot adds around £23 per user per month on top of an existing subscription; for a ten-person team that lives in Outlook and Teams, the shorter adoption curve often more than offsets the higher per-seat cost.
The NCSC specifically recommends controlling AI access through existing authentication and device management where possible. Platform-integrated tools make this straightforward because they inherit your existing admin controls, retention policies, and role-based permissions.
The trade-off is lock-in. The CMA’s ongoing work on AI foundation models flags concentration risk among a small number of providers. SMEs committing to one platform’s AI layer may face switching costs later. Choosing tools with clear data export options and open APIs is reasonable long-term caution. If your sector involves financial advice, health, or legal services, the FCA’s outsourcing guidance also requires you to treat AI tools as third-party arrangements with appropriate oversight and exit planning.
What does it cost to pick the wrong tool?
A 20-person firm on a £25 per month AI licence spends £6,000 a year. Add a Copilot-style platform add-on and the figure roughly doubles. Over a three-year cycle, licence spend alone commonly reaches the mid-five figures for a small team. If the tool sees low adoption, much of that is effectively written off, alongside the staff time spent on onboarding and any integration work.
Larger custom builds amplify the risk further. UK AI agencies typically quote £10,000–£50,000 for implementation projects once discovery, configuration, training, and integration are counted. A poor fit at that investment level means the spend restarts from scratch.
The compliance dimension adds a second category of cost. The ICO expects organisations using AI to process personal data to identify a lawful basis, carry out a Data Protection Impact Assessment where processing is high-risk, and hold a UK GDPR-compliant Data Processing Agreement with any vendor. Staff pasting customer data into a public chatbot without proper contracts in place is a specific misuse risk the ICO has flagged, and it is a common failure in the first weeks of an unmanaged rollout.
There is also a subtler cost: change fatigue. Failed technology rollouts make the next one harder. Staff who have sat through one poorly chosen AI tool will be slower to engage with the next attempt, and that wariness is earned.
What should you ask before you commit?
Before signing up for anything, ask your team to log a week of their time: emails, proposals, meeting notes, data entry. Then ask which apps they open every morning. The tool you choose should address the top two or three time drains and sit inside the apps your team already uses. If it fails either test, the best model in the world will not drive adoption.
For vendors, the questions that matter are about data, not features. Where is data stored and processed? Is it used to train models? Can the vendor provide a UK GDPR-compliant Data Processing Agreement? For regulated firms, does the tool fit within your outsourcing framework? These questions are more important than any live demo.
For your pilot, keep it narrow. Pick two or three people and one workflow. Set a clear benchmark before you start: at least two hours saved per person per week on that specific task. Run it for 60–90 days. If the benchmark is not being met, the problem is the use case, not the tool, and a more expensive version will not fix that.
The last question is about getting out before you go in. What are the export options if you cancel? What happens to your data? The CMA has flagged concentration risk among the major AI providers, and selecting tools with open APIs and clear exit paths is straightforward long-term caution for any SME making a three-year licence commitment.
