How to understand if a process can be automated (and when it's not advisable)

Edited by Solucom · June 1, 2026

There's a mistake I often see companies make as soon as they discover automation: to automate everything. It's the fastest way to end up with ten fragile streams that no one controls, more problems than before, and the feeling that «automation doesn't work.» The truth is another: automation works just fine, but only on the right things. This article is precisely for you to understand, pen and paper in hand, what the right things are. for you.

Premise: Automating everything is a mistake

Every automation has a hidden cost: it has to be built, tested, documented, and above all kept. A workflow that handles something rare or constantly changing will cost you more time than it saves. The goal isn't «how many things I automate,» but «how much time and how many errors I remove, stably.» Fewer, well-chosen workflows always beat many randomly implemented workflows.

Activity, process, system: three different things

Before deciding what to automate, you need to know What You're watching. There are three levels:

• Activity: a single action. «Copy contact data to CRM».
• Process: a sequence of connected activities towards a goal. «From initial contact to the sent quote.».
• SystemThe set of processes and the tools that make them work together. «How the entire commercial side functions.».

Why does this distinction matter? Why It's better to automate starting with individual tasks within a process, not the «system» as a whole. Those who try to automate «all of sales» get stuck; those who automate «notifications when a lead arrives» achieve results within the same week. Think big about the process, but act small regarding the task.

The Four Questions Test

For every activity you are evaluating, ask yourself these four questions. The more «yes» answers you gather, the better candidate it is.

• Is it repetitive and does it follow clear rules? If you always do it the same way and could explain it to a new colleague in two minutes, it's automatable. If it «depends» every time, it's less so.
• Does it happen often enough? Something that happens every day is worth the effort. Something that happens twice a year, almost never is.
• 3. Is the starting data already digital and sorted? If the information is already in an email, a form, or a management system, you're halfway there. If it's on sticky notes or in people's heads, it first needs to be organized.
• 4. What happens if they make a mistake? If an error is small and recoverable, you can automate with peace of mind. If an error is costly or difficult to fix, more checks are needed—or perhaps it's better to let it go.

Note how these questions measure four specific things: repeatability, frequency, data quality, and risk. Keep them in mind, because they return in the final exercise.

Two concrete examples

Good example — appointment confirmation. A client books, and you send a confirmation email with the location, time, and instructions. It's repetitive, happens often, the data is already in the booking form, and an error is minimal (at most you correct an email address). Four «yeses» out of four: automate it immediately.

Misleading Example — Responding to Important Complaints. It seems automatable («a complaint arrives, I send a response»), but it's not really. Every serious complaint is different, requires tone and judgment, and a wrong response can make you lose the customer. Here, automation can help to sort he prepare, but the answer is written and sent by a person. He seemed like a good candidate: in an honest exam, he is not.

The mini-matrix for deciding

If you don't want to reason case by case, this grid gives you the direction in three lines:

• Frequent + repetitive + low risk → automate. It's the ideal territory. Start from here.
• Rare + variable + high risk → don't start there. So much effort, little return, too much to supervise.
• Frequent + variable → automates only some parts. Assign the mechanical tasks (picking, sorting, preparation) to automation and keep humans on tasks that require judgment.

And when does AI come into play?

AI is useful in automation in a specific case: when it's necessary to interpret something unstructured. Understand what an email is about, summarize a document, classify requests written in a thousand different ways: a fixed rule is not enough here, and AI makes the difference.

When instead avoid itWhere precision and predictability are needed. «If the amount exceeds 1,000 euros, ask for approval» doesn't need AI: it's a clear rule, and a rule is faster, cheaper, and more reliable. Using AI where a simple «if» would suffice is a waste that also adds unpredictability. If you want to better understand the difference between a rule and reasoning system, I explained it separately.

Human review: draft yes, send no

There's one move that makes almost every automation safer: have the machine generate the draft and leave the final action to a person. It is the right compromise when the risk is not negligible. Keep the human in the loop when:

• The communication is directed towards an important or sensitive client.;
• There is an amount, a commitment, or a deadline involved;
• the content becomes public (a post, a review, a visible reply);
• A mistake would be difficult to fix.

In all other cases — the small, recoverable things — you can let automation run on its own. The secret is to decide in advance Where to put the control, don't find out after the first problem.

Common errors

• Start with the most complicated process. «because that's what hurts the most.» Start with the simplest: you need a win, not a feat.
• Automate a confusing process. If something is messy by hand, automating it just makes it messier, faster. Fix it first, then automate.
• Forget the risk. The question «What happens if I make a mistake?» should always be asked, even about trivial things.
• Never question it. A useful automation today can become useless in a year. Review from time to time.

The Exercise: Your Map in 15 Minutes

Here is the real work, the one I really recommend you do now. Take a sheet of paper and write ten activities what do you (or your team) do every week. For each, give a score from 1 to 5 across four columns:

• FrequencyHow often does it happen (5 = every day, 1 = rarely).
• Repeatability: how much it is always the same (5 = identical every time, 1 = always changes).
• Low risk: how harmless is it if it makes a mistake (5 = negligible error, 1 = costly error).
• Little human judgment: how little a person's head is useful (5 = no judgment, 1 = all judgment).

Sum the four grades. The activities with the highest score These are your first candidates for automation; leave the ones at the bottom where they are. In fifteen minutes, you have an organized list and a decision made with sound judgment, not just a hunch.

understand What Automating is more important than knowing. come Do it. Start with individual, frequent, repetitive, and low-risk tasks; use AI only where there's ambiguity; keep a person in the final loop when it counts. Do this, and automation stops being a promise and becomes time back in your pocket.

Frequently Asked Questions

Did you compile your top ten list? If you want, send me the top three activities on the leaderboard: I'll tell you, straight up, which ones are worth automating first and with what tool — sometimes that's all you need a simple flow, other times something else is needed. Write me your three.