What Tasks Should Never Be Automated, Even When You Can

A customer complaint, a difficult negotiation, or anything involving genuine emotional weight deserves a real person, since a fully automated response in that moment usually makes things worse.

Someone upset about a bill, grieving, or scared about a health or legal issue needs empathy and flexibility. Automation can route them to a person quickly, but it shouldn't be the one responding.

Decisions that require weighing context a system can't fully see, like whether to make an exception for a specific customer, should stay with a person who can actually make that call.

Refunds, credit decisions, pricing exceptions, and hiring decisions all fall here. AI can gather information and suggest options, but a person should decide and be accountable.

Anything where a wrong answer could hurt someone, like medical guidance, electrical or gas safety, or legal advice, should not be left to an automated system. At most, automation should direct people to emergency services or a qualified professional.

If nobody will check what an automation is doing, it shouldn't be doing anything important. Automated messages can break, send the wrong information, or keep firing at a customer who asked to stop. Every automation needs an owner.

A thank you call after a big project, a personal note to a long time customer, or a check in after a problem was fixed are relationship moments. Automating them saves a few minutes and loses what made them valuable.

If a customer is talking to an automated system, don't disguise it. People are generally fine with automation when it's honest and helpful. They feel deceived when they discover the "person" they were confiding in was a bot.

Ask: if this goes wrong, how bad is it? And would the customer be upset to learn it was automated? If either answer worries you, keep a person involved.

If you'd like help deciding what to automate and what to protect, our AI automation team starts with exactly that conversation.