Skip to content
Compliance

AI Policy for Irish SMEs: The Four Questions Every Business Needs to Answer

Your team is already using AI. Not in a pilot you signed off, but on personal accounts and their own phones. They are drafting client emails, summarising contracts and tidying spreadsheets, using company and client information, outside anything you can see.

Nobody is doing this to cause trouble. They are trying to get through the day faster. But it means the question is no longer whether your business uses AI. It is whether anyone has set the rules.

This guide shows you how to write an AI policy for an Irish SME in an afternoon. It comes down to four questions, and there is a one-page template at the end.

What shadow AI is, and why it matters

Shadow AI is staff using AI tools the business has not approved and cannot see. It is the AI version of someone keeping client files on a personal USB stick: well meant, convenient, and a real exposure.

For an Irish SME, the biggest risk is data protection. When an employee pastes a client's personal details into a personal AI account, that data has left your control. There is no agreement with the provider, no record of where it went and no way to get it back. Under GDPR, your business is still responsible for it. Our guide to AI and GDPR in Ireland covers the data protection side in full.

There is a quieter cost too. Whatever your staff learn about using AI well stays on their phones. None of it builds up as company know-how.

The two defaults nobody chooses

Without a policy, every business ends up at one of two extremes. Nobody picks either one on purpose.

Default one

Everything is allowed

Because nothing was ever said. Staff use whatever they like, with whatever data they have, and you find out when something goes wrong.

Default two

Everything is banned

Because saying no felt safer. Use moves onto personal phones, you lose what little visibility you had, and the business gets none of the benefit.

Four answers, written down, are enough to avoid both.

The four questions every AI policy needs to answer

QUESTION 01 · SEE

What can AI see?

List the data, documents and systems AI is allowed to use. Then list what never leaves the approved setup: client personal data, staff records, health information and anything under a confidentiality agreement are the usual candidates.

Example: AI can read our shared drive's templates folder and the sales pipeline in our CRM. Client files, payroll and HR records stay out of bounds.
QUESTION 02 · DO

What can AI do?

Reading is one thing. Acting is another. Decide which actions AI can take: draft a document, update a record, send a message, move money. Make the permitted actions explicit, and assume anything not listed is not allowed.

Example: AI can draft replies and update the status of a job in our system. It cannot send an email to a client or approve a payment.
QUESTION 03 · APPROVE

What needs a human?

Match the level of review to the cost of being wrong. An internal meeting summary does not need sign-off. A quote, a client letter or anything financially material does.

Example: Anything client-facing, and any figure over €1,000, is checked by a person before it goes out.
QUESTION 04 · OWN

Who is accountable?

Every use of AI needs a named person who owns it. That person deals with exceptions, answers questions and decides what happens when something goes wrong. "The AI did it" is never an answer you want to give a client.

Example: The finance manager owns AI use in accounts. The operations lead owns AI use in scheduling.

Why rules make you faster, not slower

It is tempting to see an AI policy as paperwork. In practice it is what lets you use AI on work that matters.

Without clear rules, nobody will let AI near anything important, so it stays on low-value tasks and never pays for itself. With them, you can move quickly on the work that is worth real money. When a client or your insurer asks how you use AI, you can answer in one email. And you avoid the single incident that sets the whole effort back a year.

This is commercial, not just compliance. The businesses getting the most from AI are not the ones with the fewest rules. They are the ones whose rules are clear enough that people can act with confidence.

One approved setup, not ten tools

A policy that simply lists approved tools still leaves a gap. If each department picks its own, you end up with AI silos: marketing in one tool, finance in another, sales in a third. Different permissions, different answers to the same question, and nothing connects.

The better answer is one shared company layer, with the walls still up. Every team works from the same foundation, but each one reaches only what it should.

LeadershipBoard packs, forecasts
FinanceRestricted
SalesCRM scope
MarketingBrand, campaigns
OperationsProcess scope
SupportCustomer scope
Shared company layer DATA · SOPs · RULES · PERMISSIONS · AUDIT

Take Susan, a sales and marketing manager. In a shared setup, her AI assistant can see the CRM, campaign results and brand guidelines. It cannot see the finance team's numbers or staff records, because her permissions do not include them. What Susan's team works out about using AI well is not lost in her inbox. It becomes part of the shared layer that every team builds on.

Finance sees financial data. Sales sees the CRM. Client and employee records stay restricted. Your policy then has one setup to describe, not ten.

The EU AI Act and staff training

The EU AI Act mostly regulates the companies that build AI systems and the high-risk uses of AI, such as recruitment and credit scoring. For most Irish SMEs, one part applies directly to everyday use.

Article 4 covers AI literacy. It has applied since 2 February 2025 and asks businesses that use AI to look after the AI knowledge of the staff who use it. The EU's AI Omnibus, in force since July 2026, softened the wording. Businesses must now take measures to support AI literacy among their staff, rather than guarantee a particular level of it. Doing nothing at all still falls short.

In practice, that means making sure the people using AI understand what it does, where it goes wrong and what your rules are. A written policy, a short training session and a named owner go most of the way. Teaching people what the AI is actually doing, not just which buttons to press, is also what makes adoption stick. Our six-step AI readiness guide covers that in more detail.

This is a plain-language guide written from build experience, not legal advice. No law in Ireland requires a written AI policy, but GDPR and the AI Act's literacy duty still apply. If you process sensitive data at scale or use AI in a high-risk area such as recruitment, take advice from a data protection or legal professional as well.

A one-page AI policy template

Copy this into a document, fill in each section and share it with your team. For most SMEs, this is the whole policy.

[Company name] AI Policy

VERSION 1 · [DATE]
Approved tools

The AI tools staff may use for company work, and the accounts they must use them through.

e.g. Our company AI assistant, accessed with your work login only. No personal AI accounts for company or client information.
What AI can see

The data, documents and systems AI may use, and what never leaves the approved setup.

e.g. Templates, the CRM and published price lists. Never: client personal data outside the approved setup, payroll, HR or health records.
What AI can do

The actions AI is permitted to take. Anything not listed is not allowed.

e.g. Draft documents and replies, summarise calls, update job status. Not: send client emails, approve payments, delete records.
What needs a human

The outputs a person must check before they are used.

e.g. Anything client-facing. Any figure over €1,000. Anything involving legal, medical or financial advice.
Who owns each use

A named person for each area where AI is used, responsible for exceptions and incidents.

e.g. Accounts: [name]. Sales: [name]. Operations: [name]. Overall policy owner: [name].
Training

How staff learn what the AI does, where it goes wrong and what these rules mean.

e.g. A one-hour session for everyone using AI, and a refresher whenever a new tool or action is added.
Review date

When this policy will next be checked, and what triggers an early review.

e.g. Every six months, or sooner if we add a new tool or give AI a new action.

Common questions about AI policies

Is an AI policy a legal requirement in Ireland?

No law in Ireland requires a written AI policy. GDPR applies whenever AI processes personal data, and Article 4 of the EU AI Act asks businesses that use AI to take measures to support AI literacy among their staff. A short written policy is the simplest way to show you have covered both.

What is shadow AI?

Staff using AI tools the business has not approved and cannot see, usually on personal accounts or their own phones. The risk is that company and client information ends up somewhere with no agreement, no audit trail and no control.

What should an AI policy include?

Approved tools, what AI can see, what it can do, what needs a human, who owns each use, how staff are trained and a review date. For most SMEs that fits on one page. Use the template above as a starting point.

Should we just ban staff from using ChatGPT?

A ban usually pushes use onto personal phones, where you have even less visibility, and the business gets none of the benefit. Approving one setup with clear limits works better.

How often should we review our AI policy?

Typically every six months, and sooner whenever you add a new tool or give AI a new action. More answers on how we approach this are on our FAQ page.

Set the rules, then put AI to work

A policy is the starting point. The next step is an approved setup that matches it: one shared layer, with the right permissions, that your team can actually use.

That is what our AI consultancy sets up. Book a free 30-minute call and we will look at how your team is using AI today, what the rules should be and what a sensible approved setup would cost.

Thirty minutes, on Dublin or Atlanta time.