By Décio Dalke Jr. · Oct 2026 · Leia em português

Automating a task is not automating a process

The company buys the software, the task gets faster and the gain never shows up. The difference between automating a task and automating a process, with a case from an industrial project.

Three automations and zero processes. — pixel-art row of three complete control boxes whose wire ends never connect

Décio Dalke Jr. — Arquiteto de negócios. Sócio-gerente no ecossistema MitUP (mitup.pt).

The business owner buys the software. Signs the contract, pays for the licence, books the training. And waits for the productivity gain the salesperson promised.

Six months later, the numbers don’t add up. The task did get faster, that much is true. It’s just that nobody can point to where the company gained.

The cause is a mix-up. Automating a task and automating a process are two different things, and almost everyone buys the first thinking they’ve got the second.

Automating a task means making a single step faster: the buyer raises the order in two clicks instead of ten. Automating a process means making the whole flow run from one end to the other, from order to payment, without anyone re-keying anything along the way. The first makes one person’s life easier, but the second delivers results for the company.

What’s the difference, in practice?

In 1913, Henry Ford changed the way the car was assembled. Until then, the car stayed in one spot on the factory floor, built up on trestles, and the teams moved from one car to the next. Ford did the opposite: the car started moving through the factory, and each station always did the same thing, in the same order, on the part that arrived. The labour hours for final assembly of each car fell from more than 12 to fewer than 3 [1].

Notice what he touched first: the order in which things happen. Only after that were the tools at each station and the components redesigned to serve that order.

Anyone buying software without doing that part is buying the parts of a Model T without the assembly line. They end up with better parts and keep building the car standing still in a corner of the shed.

Where does it break?

The most common case happens in practically every small and medium-sized company that decides to “go digital”.

The buyer gets a new quoting and ordering system. An excellent system: side-by-side quotes, supplier history, approval by authority level. The purchasing team’s life improves that same week.

Except nobody checked whether that system talks to payment management. Or to the check on what was actually received.

The result: the order is born in the purchasing system and dies there. Someone in finance types the same order again, into their own system. Someone in the stores writes down the receipt in a third place. Three very functional controls, each perfect in its own area, completely blind to one another. And more manual entry than before, because now there are three places to get it wrong instead of one.

Why does this need saying again?

Because I was already writing about it in 2021, before the word “automation” became a synonym for artificial intelligence. The sentence was in a piece about process management:

“The purchasing task was automated. The ‘Purchasing and payments’ process was not.”

Back then the villain was off-the-shelf software, and today it’s the AI agent. The mistake is the same, it has just become faster and more expensive: one agent is set up to draft the email, another to classify the order, a third to post it to the spreadsheet, and nobody has designed the flow the three of them live in. The company ends up with three automations and zero processes.

How do you know whether your process is modelled?

You don’t need a consultant or software to answer. Take one process in your company and try to say, out loud, four things:

  1. What goes in. What information or material starts the process, and where it comes from.
  2. What comes out. What has to be ready at the end, and for whom.
  3. Who does what, and in what order. Think about what each person delivers.
  4. How you know it went wrong. What signal appears, and where.

If you get stuck on any of the four, the process lives only in people’s heads. And what lives in people’s heads can’t be automated: it gets copied with errors, or it’s lost when the person leaves.

I saw this happen at COMPERJ, on an industrial project worth R$1.8 billion. Engineering was focused on one part, procurement on another, and production was making items without finishing them and without generating measurement. Each area followed its own process, without keeping track of the order of the other two. Nobody had designed the flow between them. Each one optimised its own part, and the whole got worse.

What fixed it was proper planning: productivity indices, scope survey, schedule, histogram, and an execution plan focused on measuring the work done, which aligned the three areas to the same order. There was also a bloated electrical team. Cutting it brought costs down, and revenue stayed where it was. What made revenue go up was the alignment between engineering, procurement and execution. Two results, two causes, and only the second has to do with a modelled process.

What should you do before buying?

Before the licence, before the pilot, before the agent: design the flow. Even if it’s on paper. Then choose the tool that serves that flow. The market tends to sell it the other way round.

Process first. Tool second. You can’t automate what hasn’t been modelled.

References

[1] Ford Motor Company, “Game-changer: 100th anniversary of the moving assembly line”, 2013. https://media.ford.com/content/fordmedia/fna/us/en/features/game-changer--100th-anniversary-of-the-moving-assembly-line.html


Originally published in 2021, on the MitUP blog and on LinkedIn. Rewritten in 2026.