Automation or AI: Choosing the Right Support for a Task

"Automation" and "AI" are often used as synonyms. They aren't, and mixing them up leads to poor choices: handing an AI model a task that a three-line rule would settle, or, the other way round, trying to write rules for work that requires reading and interpreting text.

This article offers a way to choose based on the nature of the task, with the errors to expect and the place to put a human check. It promises no specific gain: the right choice depends on your work, and AI doesn't solve everything.

Two different ways of working

Rule-based automation carries out instructions you have written: "when a form is received, create a record and send an acknowledgement." With the same data, the result is the same every time. If a case doesn't fit the rules, the automation stops or sets it aside, depending on how it was built.

Generative AI produces an answer from statistical models: summarizing a message, suggesting a draft, categorizing a request written in free text. For the same input, the output can vary, and it can be plausible while being wrong. It is useful where rules would be impossible to write, but its output needs checking.

The two can be combined: an automation triggers the work and routes the information, AI prepares a suggestion, and a person approves. That is often the most reasonable combination, rather than an either-or choice.

Three types of tasks

To decide, place the task in one of these three categories. The boundaries are blurry, but the starting question stays the same: can the right answer be described by a rule?

Fixed rules

Copying information from one tool to another, creating a record, sending a reminder, filing a document according to a clear condition. The right answer is known in advance. Rule-based automation fits, and AI mostly adds unpredictability.

Judgment on free text

Deciding whether a message is a quote request, a complaint or a billing question; extracting details from a freely written email. Wording varies too much for exhaustive rules. AI can suggest a classification, but the result should be treated as a suggestion.

Writing

Preparing a draft reply, rewording a description, summarizing an exchange. AI can speed up a first version. Tone, factual accuracy and any commitments made in the text remain a person's responsibility.

The decision table

The table below is illustrative. It shows a line of reasoning, not measurements about your business; redo it with your own tasks.

In the first two rows, the check can focus on exceptions and a sample. Once a text goes out to a customer, it covers every message.

A decision tree in four questions

1. Can you write the rule? If so, start with rule-based automation. It is simpler to test and to explain.
2. Is the content free text? If so, AI can help read it or suggest a reply, provided you plan a check.
3. What happens if the tool gets it wrong? An internal error is fixed in minutes; an error a customer can see, or one involving personal information, calls for tighter control, or no automation at all.
4. Who checks, and when? Decide before building, not after the first incident.

To work out which task to tackle first, our article What Should a Small Business Automate First? offers a prioritization grid. The two complement each other: one helps choose the task, the other the type of support.

Keeping a person accountable

Common automation tools build this idea in as approval steps: a workflow that waits for a person's response before it finishes, as described in Microsoft's documentation on Power Automate approvals. It is an example of a feature category, not a tool recommendation.

On the privacy side, the Office of the Privacy Commissioner of Canada and its provincial counterparts state in their principles for generative AI that accountability for decisions rests with the organization, not with an automated system, and that people affected by a significant decision should have an effective way to challenge it. The NIST AI Risk Management Framework, voluntary to use, is organized around four functions: govern, map, measure and manage. For a small business, its spirit can be summed up simply: know where AI is used, what could go wrong and who is watching.

If you add AI, also limit the information passed to the tool to what is necessary. This article is not legal advice; for obligations that affect your data, confirm your situation with a qualified source.

Frequently asked questions

Is simple automation less "modern" than AI?

That is not a criterion. A clear, well-tested rule is more predictable, easier to explain and simpler to monitor. Choose AI when the task requires it, not for the sake of appearances.

Can AI reply to customers on its own?

We don't recommend it for personalized replies, prices or sensitive cases. AI can prepare a draft; a person reads and sends it. A standard message with no commitment is better handled by rule-based automation.

Do we need to change tools first?

Not necessarily. Look at what you already use. If the difficulty lies in passing information between a form, an inbox and a CRM, our CRM and integrations page describes how we approach that need.

Next step

Pick one task, place it in one of the three categories and fill in the table with your own examples. To scope it with help, AETHER can review the task with you and clarify, before anything is built, what belongs to a rule, what belongs to AI and where the human check goes. See our Automation and AI page.

Describe the task to simplify