When a business shouldn't use AI
We build AI systems for a living, and they're the wrong tool for some jobs.

When nobody checks the output
The clearest case is where nobody is checking the output and being wrong costs someone. In 2024 a Canadian tribunal ruled against Air Canada after a chatbot on its website gave a customer wrong advice about bereavement fares. The company argued the chatbot was responsible for its own statements, which the tribunal called remarkable. The award was only about C$812, and it's a small claims body, but the point is simple: if it's on your website, it's your advice.
When a made-up answer looks real
The next is where the answer has to be true and a made-up one looks the same as a real one. New Zealand's courts have dealt with this directly. In February 2026, the Supreme Court said a self-represented applicant had cited authorities that appeared to be invented by AI, and that false citations may in serious cases amount to obstruction of justice or contempt. RNZ reported in 2025 that New Zealand's incidents had come from people representing themselves, not lawyers. The courts' guidelines say chatbots may invent cases and may wrongly confirm them as accurate, and the Law Society says fake cases can even carry a real citation from a different case.
When information shouldn't leave the building
Don't use AI where the information shouldn't leave the building. The courts tell lawyers not to enter private, confidential, suppressed or privileged information into these tools, and the Privacy Commissioner's office says that if you're in doubt, don't use them on personal information. Most businesses hold something in that category.
When the person wants a person
It's also the wrong tool when the person wants a person. If a customer has had a death in the family, or a long-standing client is unhappy, or a tradesman needs to stand on a deck and look at a rotten post, a tool can help prepare for that conversation but it shouldn't have it.
When a simple rule will do
Sometimes you don't need AI at all. If the job is the same every time, a simple rule, a spreadsheet formula or a standard form will be cheaper and more reliable. A process that doesn't work properly should be fixed first, because AI just makes it run faster.
When you can't tell if it's working
It's also a poor fit where you can't tell if it's working. Microsoft itself says generative AI responses aren't guaranteed to be fully factual. If you don't have someone who knows enough to spot a wrong answer, you won't find out until it costs you. Business.govt.nz says a real person should have the final say on anything important.
Where the line sits
Used on the right job, with someone checking it, AI saves a lot of time. Our own LandSure reports state what was measured and leave the why to an engineer, because that's where we think the line sits for that job. It's worth working out where it sits for yours before you start.
Sources
Sources: Moffatt v Air Canada, 2024 BCCRT 149, as summarised by DWW and Daily Hive. Jones v Family Court at Whangārei [2026] NZSC 1. RNZ, 29 August 2025. Courts of New Zealand, generative AI guidelines, 7 December 2023. New Zealand Law Society, 14 August 2025. Office of the Privacy Commissioner, "AI". Microsoft, Microsoft 365 Copilot privacy page. business.govt.nz, "Safe and smart AI use".




Comments