Which processes are worth automating — and which are not

The question is almost never "should we automate". It is what to automate first — because budget and team attention are finite, and not every process returns the investment.

In practice the common mistake is not automating too slowly. It is automating the wrong thing. An automated bad process becomes a bad process that happens faster and can no longer be corrected by hand.

Below are the signs that let you judge a process before spending money on it.

Which processes are good candidates?

Short answer: the ones that repeat, follow a clear rule, touch money, and are done manually today.

Four signs worth looking for together — not individually:

• It repeats often. Daily, or several times a week. Work done once a quarter rarely pays back, however unpleasant it is.
• It has a rule. You can write it as a sentence starting "if… then…". If it is decided "case by case" every time, what you need first is an agreement, not automation.
• Mistakes cost money. A wrong margin, a forgotten invoice, a late document — measured in euros, not in irritation.
• The data already exists. It may be scattered, but it has to exist. Automation does not create information nobody ever recorded.

When all four line up, the process is almost always worth the work. When two do, it is a candidate. When one does, leave it alone.

When does automation just cement the mess?

Short answer: when the process has not settled, when it exists only in one person's head, or when it is about to change.

Three cases where waiting is the right call:

1. The process is still finding its shape. If the procedure changed three times in the last six months, automating it will freeze an arbitrary version.
2. The rule is not a rule. "We calculate differently for this client" is not an exception if it is true of fifteen clients. That is undocumented pricing, and it has to be written down first.
3. The process is about to disappear. Automating something that a different way of working will replace in six months is the most expensive way to gain nothing.

What these three share: the problem is not in the technology. Which is why technology will not solve it.

Where do you start when there are several candidates?

Short answer: with the one that touches money daily and whose result you will see within weeks.

A useful rule: pick the process whose improvement you can measure without a separate study. If proving the result requires analysis, there probably was not one.

In practice the first process is almost always one of three: the order journey from enquiry to invoice, inventory and margin tracking, or a repeating client communication. What the first of those looks like is covered separately in where digitising actually starts.

What does an automated process actually look like?

Short answer: not a robot doing a person's job, but a place where the data reconciles itself.

A concrete example from a production system: in inventory tracking, every part carries a cost, a selling price and a margin multiplier. The selling price calculates itself, a purchase in one currency converts into the selling currency, and profit is visible per part, per sale and per client — in real time, with no manual entry. That is not artificial intelligence; it is simply one correct record instead of five approximate ones. What such a system looks like is on the parts inventory and sales platform page.

Where a decision genuinely requires judgement rather than a rule, that is where AI integration begins — but it is the second step, not the first. The first step is almost always order in the data.

What does it cost, and is funding available?

Short answer: cost depends on scope, and in many cases part of it can be covered by EU digitalisation funding.

We quote no figures here, because they depend on how many processes are included and what state the data is in at the start. Funding options and conditions are covered separately: EU funding. Materials and templates you can start from on your own are collected here: resources.

If you would like the candidates assessed together — including which ones are not worth automating — get in touch. The first conversation is about the process, not the tools.