The Practical Impact of AI on Business

By Lionel Naidoo, Director, Dragon Information Systems

Ask a group of business owners who has used ChatGPT, Copilot or something similar in the last week and most hands go up. Ask who has changed how their business actually runs because of it, and far fewer do.

That gap – between using AI and running on AI – is where the practical impact lies. No robots, no science fiction: just what AI is doing inside real businesses this year, where the money is, and what can go wrong.

AI in plain English

Traditional software follows rules a programmer wrote. AI has instead learned patterns from enormous amounts of text, images and data, and predicts the most useful response to whatever you ask. That’s why it can draft an email, summarise a contract or spot a trend in a spreadsheet without anyone writing a rule for that job.

It helps to think in three layers:

Models are the engines – GPT, Claude, Gemini, Grok. You’ll rarely deal with them directly.
Assistants are the chat windows built on top – ChatGPT, Microsoft Copilot, Claude. You ask; they answer, draft and analyse. This is where most SMEs are today.
Agents are the new layer. Give them a goal and they plan the steps, use your apps and files, and complete the task, with you approving the important bits.

Each layer needs less from you and does more for you.

This is no longer early adoption

British Chambers of Commerce and University of Essex research found in March 2026 that 54% of UK SMEs now use AI, up from 35% in 2025 and 23% in 2023. Opinium research for OpenAI and Enterprise Nation found small business users saving 5.2 hours per person, per week. And the Department for Science, Innovation and Technology found 75% of AI-using firms report higher productivity.

The telling number is this: 60% of businesses that haven’t adopted AI say the barrier is skills – not cost. The question has moved from “should we?” to “how far behind are we?”

Four places AI pays back fast

Time back. Emails, quotes, proposals, meeting notes, reports – the admin that eats evenings. AI drafts it in seconds; you review it in minutes.

Better decisions. Ask questions of your own data in plain English: “Which customers have gone quiet?” “Where did the margin go last quarter?”

Customer experience. Respond faster, in any language, around the clock, with personalised follow-ups a five-person firm couldn’t manage before.

Punching above your weight. Marketing, first-draft contracts, financial analysis, even software – capabilities that once needed a department now sit on a laptop.

Bolting AI on versus building AI in

This is the mistake I see most often. Firms “adopt AI” by keeping the same process they had in 2019 and adding a chatbot at one step. Every task still starts and ends with a person. You’ve made the typist faster, and you’ll gain 10–20% at best. Because nobody owns it, it fades once the novelty wears off. It feels safe, but it’s the slow lane.

The alternative is an AI-first workflow. You redesign the process on the assumption that AI does the first 80% – gathering, drafting, checking, formatting – and people move to the judgement points: approving, deciding and building relationships. That’s where you see gains of two to ten times across the whole process. And someone owns it and measures it every month.

Take client onboarding. The bolt-on version: someone types the welcome email with AI help. The AI-first version: the signed proposal lands, an agent creates the folder, drafts the contract from your template, sets up the project, prepares the welcome pack and books the kick-off – and one person reviews and clicks approve. Same outcome, around 90% less human time, and better consistency.

It feels uncomfortable. It’s also where the value lives.

A thought experiment

If you started your business today with AI available from day one, would it look anything like it does now? Same headcount, same admin, same turnaround times?

For most of us, the answer is no. And the competitor who starts up next year won’t have to imagine it – they’ll build it that way from the start, with lower costs and faster service. The only defence is to become that business before they do.

Small businesses can pivot. Big ones can’t.

Here’s the good news. AI is moving faster than any technology we’ve seen, and the winners will be those who iterate fastest.

For a 5,000-person company, changing a process means a steering committee, procurement and legal review, a departmental pilot, a change programme and a 12–24 month roll-out. By the time they finish, the tools have changed twice.

A 15-person company can decide on Monday, pick a tool by Wednesday, redesign one process, go live by Friday and learn from it the following week. Ten experiments in the time a corporate holds one meeting.

That speed is the real value proposition for SMEs: compounding small improvements every week while larger competitors are still writing the business case. If you don’t use it, you’ve thrown away your only structural advantage.

The tools driving the impact

The leading tools are closer to each other than the marketing suggests; the difference is where they live and what they do best.

Microsoft Copilot puts AI inside Word, Excel, Outlook and Teams, grounded in your company data and respecting existing permissions. Roughly £20–25 per user per month – best if you live in Microsoft 365.
ChatGPT is the broadest all-rounder, with Deep Research and an agent mode – the easiest on-ramp for a whole team.
Claude excels at long documents, careful writing and analysis, and Cowork and Claude Code turn it into an agent working across your files and apps. We’ve used it to build internal tools that would once have been a five-figure development project.
Grok has real-time access to X for live news and sentiment, though its citation accuracy is weaker, so verify before you act.

Google Gemini, Perplexity and the AI features your existing software vendors are adding are also worth a look. My advice: don’t agonise. Pick one, get everyone on it, and revisit in six months. The habit matters more than the brand.

From AI that answers to AI that acts

You describe the outcome: “Chase every invoice over 30 days, in our tone, and flag anyone who’s disputed before.” The agent plans the steps, opens your apps, reads your files and writes the drafts. One person reviews the exceptions and clicks go. Judgement stays human; the legwork doesn’t.

This isn’t a roadmap – these agents are available today, and at Dragon IS we’ve built our own for billing and bookkeeping. It’s what makes “AI does the first 80%” a real design principle for a small team.

The challenges ahead

The risks are real too:

It’s confidently wrong. AI still makes things up – less often than before, which arguably makes it more dangerous. Anything you’ll act on needs a human check.
Data leakage. Staff pasting client data into free tools, or Copilot surfacing files over-shared on SharePoint for years.
Skills and fear. Staff worry about their jobs and quietly avoid the tools. It’s a change problem more than a technology one.
Tool sprawl and cost creep. Five subscriptions, nobody owning any of them.
Regulation. UK GDPR applies, the EU AI Act matters if you sell into Europe, and “the AI did it” is not a defence.
The pace of change. What you set up in January is out of date by June.

None of these is a reason to wait. All of them are reasons to adopt deliberately rather than by accident.

Four rules that cover most of the risk

Use business accounts, not personal ones. Free consumer tools may train on what you paste; business plans don’t.
Know what data goes where. Decide what’s allowed in which tool and write it down on one page. That’s your AI policy.
Keep a human at the judgement points. AI drafts, people approve – especially anything legal, financial or customer-facing.
Treat AI as part of your IT. Identity, permissions, backups and UK GDPR all apply. Involve your IT provider, ideally before you switch Copilot on.

Three things to do this month

Pick one process and go AI-first. Quotes, onboarding, invoice chasing or meeting follow-ups. Redesign it so AI does the first 80%, and measure the time before and after.

Get everyone a business licence for one tool. Copilot if you’re a Microsoft business, otherwise ChatGPT or Claude. One tool, everyone on it, thirty minutes of training.

Write your one-page AI policy. What data goes where, who approves what, which accounts to use.

The businesses that win won’t be the ones with the best AI, as everyone has access to the same tools. They’ll be the ones that redesigned themselves around it first. And a small business can do that faster than anyone.

You may also be interested in:

Three step plan to AI adoption for SMEs