Safety and privacy for AI at work

Lesson 4 of 6

Keeping customer data safe when you automate

Specific habits for when AI is wired into real workflows, not just a one-off prompt, so the safety work happens once, at setup.

Everything so far in this course has been about a single moment: you, a chat window, and a decision about what to type. Automation is different. It is where AI stops waiting for you and starts quietly doing work on its own: a missed call turns into a text back, a new enquiry lands in a spreadsheet, a review request goes out three days after a job finishes. It is genuinely useful, and it also means customer data is now moving between tools while you are not watching. That is not a reason to avoid it. It is a reason to set it up carefully once.

The mindset: slow down at setup, then relax

When you do a task by hand, you are in the loop every time, deciding what happens as you go. When you automate it, you make those decisions once, at setup, and the automation repeats them without you. That concentrates the safety work into a single moment instead of spreading it across every job. Get the setup right and it stays safe every time it runs after that.

Only move the data each step needs

The most protective habit here is the same one from earlier in this course: minimise. If a roofing firm’s automation sends a review request three days after a job, it needs a name and a way to reach the customer. It does not need their full job history, their payment details, or the site notes. When you connect two tools, you usually get to choose which fields flow across. Choose the minimum. Every field you leave out is a field that cannot go wrong.

Use reputable tools and keep logins locked down

Automations usually run through a connector such as Zapier or Make, linking your booking system, your inbox and your invoicing together.

  • Stick to established, reputable tools with proper privacy terms, rather than an obscure free tool you found in a forum post.
  • Protect the login that controls it all with a strong, unique password and two-factor authentication. That one account is now the key to several others.
  • Grant only the access an automation actually needs. Fewer permissions means less exposure if anything ever goes wrong.
  • Review access every few months, especially after anyone leaves who had a hand in setting automations up.

Test with fake data first

This is the habit that separates a safe automation from a nervous one. Before an automation touches real customers, run it with a made-up contact: a test name, an email you control, a fake phone number. Watch what actually happens. Does the right message go out? Does the data land where you expected? Only once it behaves perfectly with fake data should it run on the real thing. It costs ten minutes and saves you an automation quietly emailing the wrong thing to a hundred real customers before anyone notices.

Keep a human at the important moments

Automation is brilliant for the routine and low-stakes: an automated “thanks, we’ll be in touch” is fine to send unattended. It is riskier for anything final or sensitive: a message that quotes a price or makes a promise is worth a human glance first. Decide deliberately which steps are safe to run alone and which deserve a pair of eyes before they go out.

Keep a simple map

Because automations run unattended, it is easy to lose track of what you have built. A one-page note, for each automation, of what starts it, what data moves, and which tools it touches, is not bureaucracy. It means that if a customer ever asks what happens to their details, you can answer in minutes.

Next, we turn all of this into something you can actually hand your team: a one-page AI policy.

For the full walkthrough, see our guide Keeping customer data safe when you automate.