On this page
Every second technology pitch now leads with AI, which makes it easy to forget that most of what slows a growing business down does not need intelligence at all. It needs reliability.
Two different tools
Automation follows rules you define. When a form is submitted, create a record. When an invoice is overdue by seven days, send a reminder. When a job is marked complete, generate the paperwork. The behaviour is exactly as predictable as the rules, which is precisely what you want for work that must happen the same way every time.
AI handles judgement and unstructured input. Reading an emailed enquiry and working out what it is about. Answering a policy question from a stack of documents. Drafting a reply for a person to review. It is genuinely useful there — and it is probabilistic, which means it can be confidently wrong. That property never goes away; it gets managed with scoped sources, defined limits, and human review.
The test
Ask one question about the task in front of you: could a competent new hire do this with a checklist on day one?
If yes, it is an automation problem. The checklist is the specification, and encoding it in software gives you the same result every time, with monitoring so you know if it ever stops.
If the honest answer is “no — it takes reading, context, or judgement,” it may be an AI problem. Then the follow-up questions matter: what sources should it rely on? What must it refuse to do? Who reviews the output before it counts?
Could a competent new hire do this with a checklist on day one?
Where businesses get it backwards
The expensive mistake is deploying AI on checklist work. A language model asked to do something a rule could do adds cost, latency, and a new failure mode — being wrong in a way a rule never would be — while removing the thing checklist work values most: sameness.
The quieter mistake is the reverse: trying to write rules for judgement work. The rule set grows edge case by edge case until nobody can maintain it, and it still mishandles the enquiries that matter.
In practice, they stack
The strongest workflows use both, in sequence. Automation moves the data, creates the records, and routes the work — the plumbing. AI sits at the specific points where something has to be read, classified, or drafted — and a person stays at the points where the consequences are real.
A lead-handling flow might look like: form submission captured and logged automatically, AI classifying the enquiry and drafting a first response, a person approving it before anything is sent, and automation handling the follow-up schedule from there. Three technologies, each doing the part it is actually good at.
If you are weighing up which one a workflow needs, start with the checklist test — and if the answer is unclear, that is usually a sign the workflow itself needs mapping before any technology touches it.