A conversation I have fairly regularly starts with a business owner telling me they do not really use AI yet.

A few questions later it turns out someone is drafting emails with ChatGPT, someone else is running meeting notes through a summarising tool, and a third person has installed a browser extension that helps them write faster. The business has not adopted AI. AI has arrived anyway.

Hi, I am Andy from Initial IT in Lichfield. We look after Microsoft 365, cyber security and IT support for small businesses across Staffordshire and the West Midlands, and I want to be clear up front that this is not a story about people doing something wrong. It is a story about how quickly useful things spread.

How it happens, one reasonable decision at a time

Picture someone on your team receiving a long, rambling customer email. They paste it into a public AI tool to help draft a reply. The answer comes back in seconds and saves them fifteen minutes.

Tomorrow they do it again. A week later they upload a proposal because they want a summary. A month later a colleague is using a different tool to analyse a spreadsheet.

At no point does anybody feel they are taking a risk. They are trying to get through their work. Yet customer information, financial data, internal documents and commercial plans have started flowing into systems nobody has reviewed, approved, or in some cases heard of.

Why AI spreads differently from other software

This is the bit I think gets underestimated.

A new accounting package goes through a buying process. A CRM gets discussed, costed and argued about before it arrives. AI tools are found by an individual during a working day and adopted the same afternoon, usually for free, with no purchase order and nothing to notice.

So by the time a business sits down to write an AI policy, people already have workflows built around specific tools. You are not introducing a rule into a vacuum. You are asking people to give up something that is currently making their job easier.

Why a blanket ban usually backfires

If somebody believes a tool is helping them do their job well, being told to stop rarely lands as a security instruction. It lands as the business getting in the way.

What tends to follow is not compliance. It is the same activity on a personal laptop, a personal account, or a phone, where you have no visibility at all. The risk has not gone. It has just moved somewhere you cannot manage it, which is the worst of both outcomes.

This is the same pattern we saw with file sharing a decade ago, and the businesses that handled it well were the ones that gave people a sanctioned option quickly rather than the ones that wrote the strictest policy.

What to do instead, this week

Start with a question rather than a rule. Ask the team which AI tools they are using and what job each one does. Make it plainly a fact-finding exercise and not a disciplinary one, or you will get a very tidy and completely useless set of answers.

Then three things:

• Name what is approved. Even a short list gives people somewhere legitimate to work. If you already pay for Microsoft 365, you may have a sanctioned option available that keeps data inside your own tenant.
• Name what must never go into a public tool. Customer personal data, anything covered by a client confidentiality clause, financial records, credentials. Keep it to a handful of items people can actually remember.
• Say what happens when the approved tool does not fit. Give people a route to ask. Most unapproved tools exist because the sanctioned one could not do the job and nobody had anywhere to raise it.

If your business is in a regulated sector, or you handle client data under a professional obligation, this is worth doing properly rather than informally. It is the sort of thing we build into cyber security reviews, and for firms rolling out Microsoft tooling it sits alongside the Copilot and AI work.

If you have never asked your team which AI tools they use, the answers will probably surprise you. That surprise is the useful part.

Quick check: is your business ready for tools like this?

Our free AI & Copilot Readiness Quiz takes two minutes and shows you where your business stands: what is ready, what is missing and what to fix first. No marketing list.

Take the 2-minute AI Readiness Quiz

If you would like help working out what is actually in use across your business, and whether it is helping or quietly creating risk, we are happy to go through it with you.

Frequently asked questions

Is it against the law for staff to use ChatGPT at work?

Not in itself. Where it becomes a legal problem is if personal data goes into a tool without a lawful basis or the appropriate safeguards, which is a UK GDPR question, or if you have client contracts or professional rules restricting where their information can be processed. The tool is not the issue, what gets pasted into it is.

How can I see which AI tools are being used on our systems?

A good starting point is the list of third-party applications that have been granted access to your Microsoft 365 tenant, which often shows more than people expect. Browser extensions and anything used on personal devices will not appear there, which is exactly why asking the team remains part of the job.

Is Microsoft Copilot safer than public AI tools?

For business data, generally yes, because it works within your Microsoft 365 tenant and your content is not used to train public models. That is a meaningful difference from pasting a client document into a free consumer tool. It still needs configuring properly, particularly around who can see which files.

What should an AI policy actually cover?

Keep it short enough that people read it: which tools are approved, what must never be entered into a public tool, who to ask when the approved option does not fit, and the expectation that a person checks AI output before it goes to a customer. A one-page policy people follow beats a twelve-page one nobody opens.

– Andy Price, Founder & MD, Initial IT · 01543 524 594 · hello@initialit.co.uk