Automation with judgement: what to automate and what not to
Not everything that can be automated should be automated. Technology lets you delegate almost any repetitive task to a machine, but the judgement…
· Automation
Not everything that can be automated should be automated. Technology lets you delegate almost any repetitive task to a machine, but the judgement to decide what to leave to a system and what to keep human is what separates a good automation from a frustrating experience. For the customer, and for you.
Automating is not dehumanising
The goal of automating is not to take people out of the process, but to free people from the mechanical so they can devote themselves to what needs empathy, judgement and creativity. A well-thought-out automation improves the customer experience because it answers them sooner, more consistently and without anyone having to do the same thing twenty times a day. An automation without judgement makes it worse —we have all suffered a bot that does not understand, offers no alternatives and will not let you speak to a human.
The difference between the two is not the technology. It is the judgement of whoever designs it.
What is worth automating
- Repetitive, predictable tasks that follow clear rules and do not vary: a confirmation email, recording an order, a payment reminder.
- High-volume processes where human error is costly and consistency matters: if you do something twenty times a day, the chance of making a mistake rises; the machine does not get tired.
- Tasks nobody enjoys doing and that add no differentiating value: copying data between systems, generating standard reports, sending confirmations.
- Whatever happens outside your working hours: handling queries at midnight, the reminder that goes out on Sunday. Nobody expects you to be available 24/7; the automation can be.
What is worth keeping human
- Judgement calls or exceptions: when the situation falls outside the norm, you need someone who thinks, not a system that applies the usual rule.
- Delicate conversations: serious complaints, sensitive situations, upset customers. What matters there is feeling heard, and an automatic reply —however well written— rarely achieves that.
- Complex sales: when the sale requires understanding the client's specific situation and adapting the proposal, the value is in human judgement.
- The relationship when what the customer needs is a person: there are moments when efficiency is not what matters. Presence is.
- Any process where an automatic error causes damage that is hard to undo: if the cost of the system getting it wrong is high, put a human in the path before the final outcome.
A concrete example: the chatbot's limit
Imagine an online shop receiving a hundred queries a day. 80% are questions about shipping, delivery times and stock availability: answers a bot can give perfectly, with up-to-date data, without anyone having to write them by hand. Automating that 80% makes complete sense. The remaining 20% are complaints about damaged products, questions about returns or particular situations. There, the bot should recognise that it cannot help and route to a person, rather than insisting with generic answers. That recognition of its limit is the difference between a useful bot and one that frustrates.
The golden rule: always leave a human exit
Even if you automate the initial contact, there must always be a clear path to speaking with a real person. Automation should filter and speed things up, never trap the customer in a maze of automatic options. The system resolves the simple things; the human resolves what really matters.
How to decide, case by case
For each process you consider automating, ask yourself three questions: does this need judgement or empathy to be done well? Would an automatic error be costly or hard to undo? Would the customer prefer a person at this moment? If the answer to any of the three is yes, keep human control —or at least put a human validation before the final outcome goes out.
How to improve an automation that works but does not quite convince
The problem is not always whether to automate or not. Sometimes the automation already exists, but customers do not perceive it as useful or the team avoids it because it does not give the right answers. In those cases, the diagnosis is usually found in the original design: the flow covers the common case but has no exit for exceptions, or the tone is too cold, or it simply does not listen to the question properly.
The solution is not to remove the automation; it is to adjust the judgement. Review how often it fails, which queries it does not resolve and where the customer gives up without getting what they needed. That tells you exactly what to improve. A good automation is one that evolves over time, not one that is set up once and never reviewed.
Human judgement does not only matter when deciding what to automate. It also matters when assessing whether what you automated is still doing its job.
In short
Automating with judgement is knowing what to delegate to the system (the repetitive, predictable and high-volume) and what to reserve for people (judgement, empathy, relationships and exceptions). The best automation is invisible for what works well and opens a door to a human when something falls outside the norm. No more, no less.
— AutoFlow · Automation and AI for small businesses and freelancers.