Putting AI to work in your business.

Somewhere in your business, somebody has already used a chatbot for work. A first draft, an awkward email, a summary of a long document. That, not a strategy away-day, is where this subject actually starts. What follows covers what the tools do well, what the numbers show, the rules on client data, and the way we took a real client through it. Sources are listed at the end.

The short version.

• Around 16% of UK businesses currently use AI, rising to 36% of large firms. Four in five have not started.
• The gains are real but specific: 75% of adopters say their people got more productive. Only 12% saw revenue rise. AI saves time; it does not print money.
• Your team is probably using it already. The danger is not the tool, it is the free version: the NCSC warns that what you type into public AI tools is visible to the provider and stored.
• Data protection law already applies. Pasting client details into a chatbot is processing personal data, the same as in any other software.
• Adoption is a small project when the order is right: real tasks, a business-grade tool, a one-page policy, an hour of training, then measure.

What AI is actually useful for.

Strip away the noise and today’s AI tools excel at one family of work: reading, writing and summarising. In a small business that covers more than you might expect.

• First drafts. Proposals, letters, job adverts, website copy, the email you have rewritten four times. The tool produces a competent draft in seconds; a person makes it right.
• Long documents built from source material. The heavyweight use, and the one that changes how a business works. Give it the project brief, the survey, the site notes and the relevant policy, and it assembles a structured, referenced report as a finished, formatted file, not a wall of text to copy and paste. Documents that swallowed days become reviewable drafts in an afternoon.
• Summaries. Meeting recordings turned into minutes before the kettle boils. Email threads, long documents, the 40-page report nobody was going to read.
• Questions against documents. What does this contract say about notice periods? Which of these two quotes covers installation? What changed between these versions?
• Spreadsheet help. The formula you can never remember, written and explained.
• A patient explainer. New regulations, technical jargon, an unfamiliar industry, pitched at whatever level you ask for.

And here is what it is not: a decision-maker, a source of reliable facts, or a substitute for knowing your own business. These tools sometimes state wrong things with total confidence. Every sensible rollout keeps a person between the tool and the customer, which is exactly what 84% of adopters do.

What the numbers say.

The government surveyed 3,500 UK businesses about AI during 2025:

• 16% currently use it. Large firms lead at 36%, mid-sized at 23%, micro businesses at 14%.
• Of those that adopted, 75% report more productive people and 80% use the tools at least weekly. Once in, it becomes a habit.
• Only 12% saw revenue increase. The benefit arrives as time and quality, not as a new income line.
• Among the rest, 71% see no need and 60% say they lack the skills.

Put those together and a plain picture emerges. AI is not transforming every business overnight; it is quietly shortening the weeks of the minority who adopted it, while most of their competitors wait. A modest project, started now, buys an advantage that four in five businesses have left on the table.

Your team is already using it.

Now the uncomfortable part. Whether or not your business has formally adopted anything, some of your people are using free chatbots for work, because the tools are useful and a browser is all it takes.

Why that matters comes down to how free versions operate. The NCSC’s warning is unambiguous: what you type into a public AI tool is visible to the provider, stored, and likely to be used in developing the service. Their advice follows directly: keep sensitive information out of public tools, and read the terms before relying on one.

So the useful question is not whether to allow AI. It is whether you want client information typed into tools nobody vetted, on terms nobody read. Banning it does not stop it; it hides it. Giving people a proper tool, with a short rule about what never goes in, is what actually works.

The rules that already apply.

There is no separate AI law to wait for. If what goes into an AI tool includes personal data, names, addresses, client details, staff records, then you are processing personal data and UK data protection law applies, exactly as it does to your email or your CRM.

For a small business, four rules of thumb carry most of the weight:

• Client and personal data goes only into tools the business has approved, on business terms. Never into personal accounts.
• Confidential business information gets the same respect. Pricing, contracts and plans have no place in a free chatbot.
• A person signs off anything that leaves the building. The tool drafts; a human answers for it.
• Be able to say what you use and why. The ICO has written guidance for exactly this, and it is shorter than you would fear.

How to choose a tool.

The market is loud. The checklist is not:

• Business terms that keep your data out of training. This is the line between free and business tiers, and it is most of what you are paying for.
• Admin controls, so access can be granted and removed like any other business system.
• A fit with the actual work, established by a short trial on real tasks rather than a demonstration.

Shortlist two or three, give a few people a fortnight with them on live work, and let the results choose.

What it looked like at BlackBrick.

None of this is theory for us. BlackBrick Studio, a Cardiff architecture practice we look after, wanted to bring AI into the way they work, and an architect’s work is exactly the heavyweight case above: dense source material in, long structured documents out. So the trial matched the job. We put the leading platforms head to head on a real project brief and asked each for a full pre-application design report, then judged the results the way architects would: finished, formatted documents ready to work on, with correct references to Welsh planning policy rather than English. The winner went in for the whole team, connected to their Microsoft 365 so it works with their own project files, with guides written for the workflows the practice actually runs. Client work stayed protected throughout, and the detail that says the most: several of the best recommendations cost nothing, because the right answer included software they were already paying for.

Where this is heading.

Predictions about AI age badly, so this guide will not make any. Two things, though, are observable rather than predicted. The tools improve quickly: capabilities that decided our client’s head-to-head trial did not exist as differences a year earlier, and what you can buy today is the floor, not the ceiling. And the direction of travel is from answering questions towards doing multi-step work: reading, checking, assembling and formatting with less hand-holding each version.

Which changes what adoption means. You are not buying a product; you are joining a curve, early. The businesses that put the habits, the policy and the data discipline in place now are the ones each improvement lands on ready ground for. The four in five that have not started will meet the same curve later, steeper.

Where to start.

1. Pick real tasks. The three most repetitive reading-and-writing jobs in your week are the pilot. Not a transformation programme, those three jobs.
2. Choose a business-grade tool and write the one-page policy. Approved tools, what never goes in, human sign-off before anything reaches a client.
3. Spend an hour showing people. A demonstration on their own work beats any memo. The barrier in the government research was skill, not appetite.
4. After a month, keep what earned its place. Ask what saved time, drop what did not, and repeat the question quarterly.

The reward, when it lands, is the least dramatic thing in technology: the same work finished earlier, by the same people, with the client data exactly where it belongs.