Every client conversation in 2026 includes AI somewhere. Sometimes it’s the headline question, “should we be doing Copilot?”, and sometimes it’s the third or fourth item on the list, but it’s always there. After eighteen months of helping SMEs work out which AI conversations are worth having and which aren’t, we’ve got a fairly clear pattern of where this stuff is genuinely paying for itself and where it’s costing money for very little return.
This isn’t a pitch for AI or against it; it’s what we see on the ground.
Where it’s earning its keep
Three categories, in roughly decreasing order of how predictably they pay back.
Transcription, summarisation, and meeting follow-up. This is the clearest win we see, with the endpoint (the laptops, phones and tablets people work on) running a meeting-transcription tool that produces a usable summary and action list inside ten minutes of the meeting ending is genuinely useful. We’ve seen SMEs cut the time their senior people spend writing up meetings by something close to half. The licence cost is small. The behaviour change is fast: people either use it from the first week or they don’t, and they almost always do.
Drafting first passes of customer-facing copy. Sales emails, proposals, product descriptions, internal comms, not the final version but a first draft that’s 70% there. The pattern that works is “AI for the boring middle, human for the opener and the close”. Where this falls down is when somebody tries to use it for highly technical or regulated copy and doesn’t realise the output is plausibly wrong in ways that look right.
Searching across internal documents. Microsoft 365 (M365, the Microsoft cloud bundle: email, Word, Excel, Teams, file storage) Copilot’s “find me the thing we wrote about X” capability, when it’s set up properly, does something the file-search bar never did. The condition is that the documents have to be in the places Copilot can see them, which in practice means SharePoint and OneDrive, properly organised, with sensible permissions. The SMEs who had their file storage in good shape get value from this fast. The ones who didn’t are getting value from the cleanup the Copilot rollout forced more than from the Copilot itself.
Where it’s not earning its keep, yet
Full Copilot rollouts on day one. The biggest mistake we see SMEs make is buying Copilot for everyone in month one and expecting productivity gains to land by month three, when they don’t. The licence cost is real, the behaviour change is uneven, and the ROI conversation gets awkward fast. The clients who got value did pilots: five to ten power users, three months, then a structured review, and only rolled wider when the pilot users couldn’t imagine life without it.
Customer service chatbots without a human escalation path. We’ve seen a handful of SMEs try to put a chatbot on their website that handles tier-one queries autonomously. None of them have stuck with it for more than six months. The technology is closer than it was, but the failure mode (sounding confidently wrong to an actual customer) is expensive in ways that don’t show up in the productivity-gain spreadsheet.
Wholesale finance or HR automation. AI tools that promise to read invoices, categorise them, post the entries, and reconcile against the bank are a real product category, but the integration overhead for an SME with one bookkeeper is rarely worth it. The 80% case works. The 20% edge cases (credit notes, foreign currency, multi-line splits) eat the time the automation was meant to save.
“AI strategy” engagements that don’t ship anything. Some SMEs are spending six-figure sums with consultancies on AI strategy documents. We’ve read a few of these. The good ones identify two or three specific workflows worth automating. The bad ones produce a 40-page deck that talks about “competitive advantage” and recommends another phase of strategy work.
What’s coming next
A few things we’re watching.
Agentic workflows that span more than one tool. The pattern where an AI does the meeting summary, then drafts the follow-up email, then adds the action items to the project tracker, then schedules the next meeting, this exists in demos and it’s getting closer to working reliably outside them. By 2027 we expect at least one or two of these to be solid enough to recommend.
On-device AI for confidential work. The data-leakage anxiety SMEs have around prompting cloud models with sensitive customer data is real, and it’s slowing some legitimate use cases. Smaller models running on a laptop’s neural processor (the dedicated AI chip in newer laptops) are starting to be good enough for things like drafting and summarisation. This will matter more than it currently looks like it will.
Verticalised tools beating horizontal ones. Generic Copilot is fine, but the vertical tools, AI built specifically for accountants, for property managers, for recruiters, are starting to outperform it for the specific workflows of those professions. We expect this to accelerate.
The compliance question. Cyber-insurance questionnaires now ask about AI exposure. By 2027 we expect them to ask in detail, the way they ask about backups now. SMEs that have a clear answer to “what data goes into which model, and where does the output land” will be in better shape than those who don’t.
What we see on the ground
The most common pattern: an SME has bought one or two AI products, isn’t quite sure which of their team are actually using them, and isn’t sure whether to expand the licences or cut them. The honest answer is usually “you can’t tell yet, because you don’t have visibility on usage”. A two-week visibility exercise, pulling usage data, asking the users directly, identifying the workflows where AI is actually changing the work, answers the question.
The second pattern: an SME hasn’t done anything with AI and feels they should have. The right answer is rarely “buy the platform”; it’s “identify two workflows where this might matter, run a four-week pilot with three people, see what happens”. Most pilots tell you something useful within a month, either way.
Practical implication for SMEs
The framework we’ve ended up using internally is three questions:
- Is there a specific workflow this is meant to change, or is the goal “we should be doing AI”?
- Who in the business will be the first three users, and are they engaged?
- What does success look like in twelve weeks, in a measurable form?
If those three questions get good answers, the project is worth running. If they don’t, the project is worth deferring until they do.
That’s our AI Enablement practice. We sit alongside SMEs working out where AI actually pays back, run the pilots, set up the visibility, and pull the plug honestly when something isn’t working.
The cost of waiting another quarter to think about this isn’t dramatic until suddenly it is. Competitors who got their pilot right twelve months ago are now ten or fifteen per cent faster on the work that matters: proposal turnaround, client reporting, internal comms. That gap compounds, and “we’ll look at AI next year” stops being a neutral position. The cheap, low-risk move is a small pilot now; the expensive move is a panicked platform purchase in eighteen months when the board asks why everybody else is using it.
Looking at AI tooling and not sure where it actually fits in your business? Drop us a note at info@jmopartners.co.uk. One of us will read it.
JMO|Partners · Enterprise IT, sized for SMEs.