What Not to Automate: Tasks That Still Need a Person

Do not automate tasks that depend on judgement, carry high stakes, involve strong emotion, rarely repeat, are undocumented, sit in an unstable process or need legal sign-off. Automate the preparation around them instead, so a person makes the final decision with better information in front of them.

Why is it worth knowing what not to automate?

Most advice about automation is about what to switch on. Fewer people say what to leave alone, yet bad automation of the wrong task is more expensive than no automation. It can send the wrong message to a customer, make a decision nobody reviewed or lock in a process that was already broken.

The aim is not to avoid anything difficult. It is to put the machine where rules are clear and keep a person where responsibility, empathy or discretion is needed.

What are the warning signs that a task should stay with a person?

If a task shows one or more of these traits, treat it with caution. The more that apply, the stronger the case for keeping it manual.

Judgement-heavy. The right answer depends on context that is hard to write down, such as whether to make an exception for a long-standing customer.

High-stakes. A mistake could cause significant financial, safety, legal or reputational harm.

Emotional. The other person is upset, grieving, frightened or angry, and how you respond matters as much as what you say.

Rarely repeated. If a task happens a handful of times a year, the effort of building and maintaining automation may never pay back.

Undocumented. If the steps live in one person's head and nobody can describe them, there is nothing reliable to automate yet.

Unstable process. If the way of working changes every few weeks, the automation will keep breaking or keep doing the old thing.

Needs legal sign-off. Anything where a qualified person must review or approve before it takes effect should keep that person in the loop.

Task types, why to be careful, and a safer option

The pattern that works is to automate the prep, not the decision. The table shows what that looks like for common task types.

In every row, the safer option keeps a named person accountable while taking the repetitive work off them. That is usually where most of the value is anyway.

What does automating the prep, not the decision look like?

Take a refund request. An automation can pull the order, check the policy, note the customer's history and draft a recommended reply. A person then reads one screen instead of searching four, and makes the call. The same pattern works for contract drafts, quotes and complaints: the system assembles, the person decides.

This also gives you a gradual path. If a task proves predictable over time and the person approves nearly every suggestion unchanged, you can later consider loosening the review for the simplest cases. That decision should rest on records you have kept, not on optimism.

How do you decide when you are unsure?

Ask three questions. What is the worst realistic outcome if this goes wrong, and can it be undone? Could a new colleague follow written rules to do this correctly? Would I be comfortable explaining to the customer that software handled it? If the answers are uncomfortable, keep a person in charge and automate only the surrounding steps. Where rules apply, such as data protection or sector regulation, check them with a qualified adviser rather than assuming a tool is compliant.

Erudience services related to this guide

Use cases related to this guide

Frequently asked questions

What business tasks should never be automated?

There is no universal list, but tasks involving high-stakes judgement, distressed people, legal sign-off or unclear processes should keep a person in charge. You can still automate the preparation, such as gathering information or drafting, so the person decides faster.

Is it bad to automate customer service?

Not if you choose the right parts. Routine, rule-based requests can suit automation, while complaints and emotional situations usually need a person. A good setup routes difficult cases to a human quickly, with the details already summarised.

Can AI make decisions for my business?

It can suggest, rank and draft, but responsibility stays with you. For decisions that affect money, safety, employment or legal rights, keep a person reviewing and approving the output.

What if a task is only rarely done?

Probably leave it manual and write a checklist instead. Automation takes effort to build and maintain, and a task that seldom happens may never repay that effort.

Can a task become automatable later?

Yes. Once a process is documented, stable and well understood, it may become a good candidate. Mapping it first is a sensible way to find out.