Skip to content

What to automate with AI, and what still needs a person.

Nearly every company is "using AI". Far fewer are getting much out of it. The difference is usually what they chose to automate.

On this page
  1. The gap between using and benefiting
  2. Good first candidates
  3. Keep a person on these
  4. How we do it ourselves
  5. A simple test before you automate
  6. Sources

The gap between using and benefiting

  • McKinsey's 2025 global survey: 88% of organisations use AI regularly in at least one part of the business. Only 39% report any impact on profit (EBIT), and most of those say it's under 5% [1].
  • The US Census Bureau's business survey puts AI use across all US firms at about 17–20% in late 2025 to mid-2026. It is higher at bigger firms [2].

So a lot of businesses are trying it. Fewer are getting results. Here's the pattern that tends to work.

Good first candidates

Work that is frequent, boring and checkable:

  • Sorting and tagging incoming email or enquiries.
  • Pulling fields out of documents (invoices, forms, PDFs) into a system.
  • Drafting routine replies and follow-ups for a person to approve.
  • Summarising long threads or call notes.
  • Turning one piece of content into several formats.

Keep a person on these

Anything where a wrong answer costs money, trust or legal trouble: prices, refunds, policies, contracts, medical or legal advice, and anything sent to a customer without review.

Here's why this matters. In 2024, Air Canada's website chatbot told a grieving customer he could claim a bereavement discount after booking. The airline's actual policy said otherwise. A Canadian tribunal ruled that the airline was responsible for what its chatbot said. "It makes no difference whether the information comes from a static page or a chatbot" [3].

Your AI speaks for your business. Plan for it being wrong sometimes.

How we do it ourselves

We don't have a client automation case study to show you yet. We do run our own business on these:

  • Gmail automation: routine inbox handling runs on its own. A person still reviews anything that needs a reply.
  • Lead generation: finding and qualifying prospects runs as a workflow. A person decides who we approach.
  • Customer outreach: follow-ups are prepared and scheduled automatically. A person approves what gets sent.
  • Motion graphics from code: edit a line, get a new cut.

The pattern is the same each time. The machine does the repetitive part, and a person makes the call.

A simple test before you automate

  1. Does it happen at least weekly?
  2. Can you describe "done right" in a sentence?
  3. Can a person check the output in seconds?
  4. What happens if it's wrong once? If the answer is "a customer gets hurt", keep a person in the loop.

Sources

  1. McKinsey, "The state of AI in 2025", opens in a new tab
  2. US Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users" (2026), opens in a new tab
  3. Moffatt v. Air Canada, 2024 BCCRT 149, as reported by Law360 Canada, opens in a new tab
WhatsApp ↗