Article
Why automating a messy process makes it worse
Published on · José María de Juan
Many automation projects start the same way: a process takes too long, and someone proposes putting software on top of it. The request that travels by email today will become a form, the spreadsheet will become a screen, and approvals will arrive with a notification. Months later the process takes just as long, and now it is also hard to change.
What automation fixes and what it does not
Automating a step makes it run faster, with nobody typing it in. It does not decide whether the step was needed. If a purchase request goes through four approvals because there was a problem with a supplier years ago, the new software will ask for all four approvals, on time, for ever.
A messy manual process has one advantage that is lost when it is automated: people correct it as they go. Whoever receives the email knows that this approval is a formality and gives it without looking, or picks up the phone when something is urgent. The mess is visible and can be argued about. Once the same route sits inside a program, it stops being seen: the unnecessary steps become screens, and nobody asks any more why they exist.
Where the mess shows
Some signs come up in very different companies:
- The same piece of data is entered in two places, and nobody knows which one wins when they disagree.
- An approval is always given by the same person, who never turns it down.
- To find out where a request stands, you have to ask someone.
- There is a "control" spreadsheet that only its author understands.
None of them is fixed by technology. Behind each one is a decision nobody took: who owns each piece of data, who decides at each step, and which steps exist only out of habit.
Three questions before automating anything
Before a line of code is written, it is worth drawing the process as it runs today, which is rarely what the procedure says, and asking three questions about each step.
The first: if this step disappeared tomorrow, who would notice? If the answer is "nobody", the step goes. Automating it would be a way of keeping it.
The second: who decides here, and with what information? A step where nobody decides anything is a transfer of data, and a system can do that. A step where someone decides should stay with a person, with the information in front of them.
The third: where does each piece of data come from, and who answers for it? Until that is clear, any integration will copy the doubt from one system to another.
With the answers, the process usually ends up shorter. What is left is what deserves to be automated, and then the software is smaller, is built sooner and costs less to maintain.
Where to start
There is no need to review the whole company. It is enough to pick the process that draws the most complaints and walk through it with the people who do it every day. At LATIDE we do this in a first review of about 60 minutes, free of charge, and hand over a one-page summary of what we saw. Sometimes the conclusion is that the process is fine and what is missing is a tool. Other times, that half the steps can go. Either way, you know before spending on software.