Skip to content
Sanfeliu Software

Guides

AI usage policy for companies: what it should say and how to make it stick

An AI usage policy is a short document, one or two pages long, that tells the team which tools and accounts they can use artificial intelligence with, which data must never go into it, that everything that comes out of the AI is reviewed before it’s sent, and who to ask when in doubt. You need one because the team already uses AI, with or without a policy: without rules, everyone decides for themselves what to paste into it.

Updated

Thirty minutes, free and with no obligation, with the engineer who would lead the project.

Why you need one, even if you’re a small team

In almost every company someone already uses an AI assistant to write emails, summarise documents or prepare quotes. Often with a free personal account, and without anyone having told them what they can put into it.

The risk isn’t AI, it’s not having talked about it. A quote with a client’s prices, a contract or a list of employees pasted into a free account are company data in a place the company doesn’t control. A one-page policy prevents that. And along the way it tells the team that they can use it, which is what many don’t dare to ask.

What it should say

  • Which tools can be used and with which accounts: the company’s, not personal ones.
  • Which data never goes in: personal data of clients and employees, passwords, bank details and anything covered by a confidentiality agreement.
  • That everything is reviewed. What the AI writes is a draft: whoever sends it answers for it, as if they had written it.
  • When to say that something was made with AI: to clients, in official documents or in images that look real.
  • What to do with what isn’t yours: other people’s texts, images and code aren’t given to the AI to copy.
  • What it isn’t used for: decisions about people (hiring, assessing or dismissing) without a person reviewing them.
  • Who to ask when in doubt, and who to tell if something went in that shouldn’t have.
  • When the policy is reviewed: the tools change, and the document has to change too.

A one-page outline

This outline is a starting point. Change whatever you need and keep it short: a policy nobody reads is no use.

  1. What we use AI for at this company, in two lines.
  2. Authorised tools and accounts, and who hands out the licences.
  3. What never goes in, with examples from our own work.
  4. Review: what must be checked before sending or publishing anything.
  5. Transparency: when we say that AI was involved.
  6. Point of contact, and what to do if there’s a mistake.
  7. Version date and when it will be reviewed again.

Personal data and confidentiality

If personal data goes into the AI, data protection law still applies: you need to know which provider you’re working with, on what terms and whether it uses what you put in to train its models. Business plans usually give these guarantees by contract; with free accounts, you can’t take it for granted.

We’re not lawyers: the policy has to fit your record of processing activities and the contracts you have with clients. If in doubt, show it to your adviser before handing it out.

How to make it stick

  • Explain it in person, with examples from each department’s work. A PDF sent by email changes nothing.
  • Give company accounts to those who need them: if the right option is easy, nobody uses the wrong one.
  • Appoint a point of contact who gathers doubts and takes them to whoever can answer them.
  • Review it every few months, with the questions that have come up.

Policy and training go together: a document without training doesn’t get read, and training without a document gets forgotten. On top of that, the EU AI Act asks companies for AI literacy measures for the people who use it.

How we do it at SANFELIU

At SANFELIU, a custom software consultancy in Barcelona, the policy isn’t a separate document: it comes out of our AI adoption support. When a question from the team keeps coming up, we turn it into a company guide that says how AI is used in your work and what must never go into it. The guide is yours, even if you stop one day.

The support also includes a monthly meeting with management, a training session every three months at your company and the whole team’s questions, with no limit. We’re not lawyers: the legal side has to be reviewed by your adviser.

Thirty minutes, free and with no obligation, with the engineer who would lead the project.

Frequently asked questions

What is an AI usage policy?

It’s a short document that tells the team which tools and accounts they can use artificial intelligence with, which data never goes in, that what comes out is reviewed, and who to ask when in doubt.

What should never be put into AI?

Personal data of clients and employees, passwords and keys, bank details and anything covered by a confidentiality agreement, unless it’s in a company account with a contract that allows it.

How long should it be?

One page, two at most. It has to be short enough for everyone to read and specific enough to be useful: examples from your own work, not general principles.

Who should write it?

Management, with someone who knows the tools and, for the personal data part, your legal adviser. And it should be gone over with the team, who are the ones who know what they use it for.

Is emailing it round enough?

No. A policy that is only emailed doesn’t get read. It has to be explained in person, with examples from each department, and reviewed every few months with the questions that have come up.

Tell us about your case in thirty minutes.

Book a meeting with the engineer who would lead the project: thirty minutes, free and with no obligation. If we’re a fit, we tell you how we’d start; if not, who to turn to.

Or email me@joansanfeliu.com

Prefer to write?

Tell us about your case in a few lines and we’ll reply by email.