The mistake almost everybody makes on the first choice
When a company decides to automate something, instinct pulls it towards the biggest process, the one people have been complaining about in meetings for years
It's the wrong choice, because the biggest process is almost always the most tangled one as well, the one with twenty exceptions, three departments involved and two people who do it differently without ever having noticed
Starting there means spending months arguing about special cases before you even see something working, and in the meantime the idea spreads around the company that automation is a promise that never arrives
The first process shouldn't be the one that hurts most, it should be the one that shows you quickly that the mechanism holds
The real test, four conditions that have to hold together
A process is a good candidate for the first automation when it has four features together, and if one of them is missing look elsewhere before pushing on with it
- high volume, meaning the very same thing happens many times a month, because the saving is born of repetition and not of difficulty
- rules that can be written down, meaning somebody in the company can say how it's done and can say it in sentences, not with an it depends followed by half an hour of anecdotes
- low damage if it goes wrong, meaning that if the system does something crooked you notice soon and put it right with no embarrassing phone calls and no money lost
- a person inside who really knows it and has a bit of time to give it, because without them the rules never come out and the project drags on
The last condition is the one that gets underrated most, a process that's perfect on paper but described by nobody stays a process that doesn't get automated
Why the best candidate is almost always boring
The automations that get off to the best start are the ones on jobs nobody would ever talk about over dinner, downloading the same files from the same portals every month, renaming them according to the office rule, retyping a document that already arrived in a tidy format, checking that two lists say the same thing
They're boring, and boredom is exactly the signal you're looking for, because where there are no decisions to take the rules are already clear
There's a practical advantage too, a boring, frequent process gives a result you can see, the person who used to do it stops doing it and notices on their own, and that counts more than any presentation, because when the time comes to automate the difficult process you'll have somebody in-house who has already seen the thing work
How you build the list of candidates
You don't need a long analysis to make the list, you need half a day done properly
Ask the people who work in the office which part of their day they'd gladly redo with a single button, because whoever does the job knows it better than anyone watching from outside
Write down every task that comes up with three pieces of information beside it, how many times a month it happens, how many minutes it takes each time, and what happens if it's done badly
Then look at the list and drop everything that happens only a few times a month, even if it annoys you, because you never win back the time it takes to explain those rules
Of what's left, keep the tasks a person can describe from beginning to end without stopping, and set aside the ones where the answer is it depends
In the end three or four names are left, and at that point the choice makes itself, take the one with the highest volume among those with low damage
The sum, with numbers I'm making up here
Say two candidates come out of the list, the first is recording a document that arrives four hundred times a month and takes three minutes each time, the second is a complicated case that comes up fifteen times a month and takes forty minutes each time
The first makes one thousand two hundred minutes a month, that is twenty hours, the second makes six hundred, that is ten hours
The first is worth twice the second in hours, but above all it's worth far more in ease, because four hundred identical cases mean rules you can see with the naked eye, while fifteen complicated cases mean fifteen different stories
If you want to redo the sum with your own company's numbers, there's a calculator on the work out the savings page
The first process teaches you the second
Once the first automation is running, the list you made at the start may no longer be in the same order, and that's not a problem
It happens because writing the rules of the first process forces you to look closely at how your office really works, and from there you see things that weren't visible before, the bottleneck that was two steps further along, the figure that gets checked three times because nobody trusts the first check
So choose the first process with your head, but don't settle all the others now, because you'll make that decision better in a few months
If your case is an old program, that's no problem
It may be that the best process to automate runs on a program with no direct link to the outside, no API, no serious exports, just screens
That's no reason to rule it out, because an automation can work directly on the screens, opening them and filling them in the way an operator would, and this works even on programs nobody maintains any more
The test stays the one from before, and the kind of program is a technical matter that comes afterwards