AI should not write your fire alarm certificates. Here is what it should do.

We build AI into software for fire and security contractors. So it may be surprising that the first thing we tell new customers is what we will not let it do.

AI will not decide whether a system is compliant. It will not invent a test result. It will not sign a certificate. A certificate is a statement by a competent person that work was done and done correctly. If a machine wrote it, it is not that statement any more.

That still leaves a lot of useful work for AI to do.

The problem with asking a chatbot #

General chatbots are good at writing text that sounds right. Ask one to write a fire alarm maintenance report and it will produce something fluent and confident.

Some of it will be wrong. It may reference the wrong edition of a standard, use clause numbers that changed in 2025, or describe a check that was never carried out. Worse, it will read so well that nobody questions it.

In most industries, a confident mistake is embarrassing. In life safety work, it ends up in a customer's logbook.

What AI should do instead #

In our experience, AI earns its place in a fire and security business in three jobs.

Turning what the engineer said into a clean report. The engineer speaks their notes on site instead of typing on a phone. AI turns that into a structured report in your format, corrects the spelling of the manufacturer and the site name, and keeps every fact the engineer gave. It does not add facts. If the engineer did not mention the panel clock, the report does not say it was checked. It flags that it is missing.

Reading supplier price lists. Suppliers send prices as PDFs, spreadsheets and scanned sheets. AI is good at reading these and matching them to your parts list. A person still approves any price change before it reaches a quote.

Drafting quote wording from approved clauses. AI assembles a quote from your own library of reviewed compliance wording, chosen to match the job. It does not write compliance language from scratch.

The rule behind all three #

In every case, the AI works from something a person has already said or approved, and a person checks the result before it goes to a customer.

That is not a limit on what AI can do. It is what makes the output trustworthy. The engineer still owns the findings. The estimator still owns the price. The business still owns the wording. AI removes the typing, the retyping and the looking up.

What this means in practice #

An engineer who used to spend ten minutes typing a report now spends three minutes talking. The office stops rewriting poor notes. The customer receives a report that reads as if a senior engineer wrote it, because the facts came from the engineer and the structure came from a template that was checked once and used every time.

Nobody had to trust a machine with a compliance decision to get there.

A test for any AI tool you are offered #

Ask the supplier three questions:

1. Can it add a fact the engineer did not give?
2. Where does its compliance wording come from?
3. Who signs off before the customer sees it?

If the answers are not "no", "your approved library" and "a qualified person", keep looking.