When is it not worth automating?

Whoever discovers automation is tempted to apply it to everything. But a badly chosen automation creates more work than it saves, and can drive customers away. A good rule: automate what repeats, has clear rules and costs little when it fails. For the rest, think twice.

Eight situations where it is better not to automate

1. The task hardly repeats. If it happens a few times a year, the time to set up and maintain the automation is greater than the time saved.
2. The rules are not clear. If each case needs a different decision, there is nothing to automate yet. Define the rules first.
3. The conversation is delicate. Complaints, difficult news, situations where the person needs to feel heard. Here, a person makes the difference.
4. The data is not reliable. If the starting information is incomplete or disorganised, automation multiplies the errors instead of solving them.
5. A mistake is costly. If a failure can cause serious loss, a legal consequence or a loss of trust, it helps to always have a person confirming.
6. Nobody owns the process. An automation without an owner stops working when something changes, and nobody notices.
7. The process itself is bad. Automating a confusing process makes it faster and just as confusing. Simplify first.
8. The customer prefers a person. In some services and for some audiences, the human relationship is part of the value.

Questions to ask before automating

• If this fails with nobody seeing, what happens? If the answer worries you, a confirmation step helps.
• How much time do I really save? Count the time to set up, test and maintain too.
• Who is responsible for this? One person, by name.
• Is the data passing through here personal or sensitive? If so, data protection rules apply, and extra care helps.
• Will the customer realise it is automatic? And if they do, are they better or worse served?

Alternatives to automation

1. Eliminate the task. Often the best automation is to stop doing something nobody uses.
2. Simplify. Fewer steps, fewer fields, fewer approvals.
3. Create a checklist. A list of steps anyone follows avoids forgetting without technology.
4. Create templates. Pre-prepared texts and documents save time and keep human control.
5. Automate only a part. For example, the acknowledgement yes, the reply to the request no.
6. Keep a person in the loop, confirming before sending.

A sign to stop

If an automation takes more work to watch than the original task did, or if customers begin to complain, stop and review. Switching off an automation is not failing: it is learning what it should do. In process automation, what is left out is as important as what goes in, and each automation is tested before being switched on. To choose what to automate, read which tasks to automate first, and for the case of customer service, whether you have to tell customers it is an AI.

The "what if it fails?" test

Choose a task you are thinking of automating and write three failure scenarios: the automation does not run, sends what it should not, or sends to whom it should not. For each, write what would happen, who would notice and how soon. If in any case the answer is "nobody would notice for weeks" or "we would lose a customer", you need a human confirmation step, or not to automate. Repeat the exercise in six months with the automations you already have: some deserve a review.

In short

• Do not automate what hardly repeats, has fuzzy rules or delicate conversations.
• Unreliable data, costly mistakes and ownerless processes call for care.
• Before automating: eliminate, simplify, create a checklist or templates.
• If watching the automation takes more work than the task, stop and review.

Quick questions

So automation is not worth it?

It is very much worth it when well chosen: tasks that repeat, with clear rules and low risk. The article serves to avoid the others.

Can I automate only part of a task?

You can, and it is often the best solution. Automate the repetitive part and keep a person on the part that needs a decision.

Is AI good for everything?

No. AI helps with language tasks, but it can make mistakes and invent. For tasks with consequences, always keep a person reviewing.