Business process automation: what to automate first and what not

When a founder tells me they want to automate, they have almost always already picked the tool and not the process. The right order costs one afternoon and changes the decision in most cases.

The rule: boring, repetitive and verifiable

A process is a good candidate when it meets all three at once. Boring, because nobody will miss it. Repetitive, because if it happens twice a month the saving will not pay for the setup. And verifiable, because you have to be able to tell in one second whether it went well.

The third is the one almost everyone skips, and it is the one that decides whether you sleep at night. An automated process you cannot check is not automation: it is a black box that will surprise you one day with a client in front of you.

The one-afternoon inventory

Take a sheet of paper and three columns. For every recurring thing done in your company, write down:

1. Frequency. How many times a month it happens. If you do not know, look at last week’s inbox or calendar and multiply.
2. Time. How many minutes it takes each time, start to finish, including the back and forth of waiting for someone to reply.
3. Damage. What happens if it goes wrong. From “nobody would notice” to “we lose the client”.

Multiply frequency by time and sort descending. That ranking almost never matches the feeling of what annoys you most, because what annoys you most tends to be what happens rarely and hurts a lot — and that is not automated, it is redesigned.

The four exits for any process

When NOT to automate, even if it meets the rule

• When the process changes every month. You will spend more time maintaining the automation than running the process by hand. Stabilise it first.
• When the process is wrong. If three steps exist because something happened once in 2019, automating them fossilises them forever. Remove them first.
• When nobody owns the outcome. Every automation needs a person who finds out when it fails. With no owner, the failure is discovered by the client.
• When the volume really is low. Five times a month is a calendar reminder, not a project.

Delegate, automate or AI: not the same thing

All three take work off your hands, but they cost and fail differently. Choosing wrong is what turns an “automation project” into a badly paid new hire.

• Delegating works when the process needs judgement and human contact. It costs salary and training, and it fails when there is no written procedure.
• Automating works when the rules are clear and there are no exceptions. It costs setup once, and it fails silently when something you did not foresee changes.
• AI works when the input is messy — free text, emails, messages — and the output can be reviewed. It costs little per use, and it fails plausibly, which is the dangerous part: it gets things wrong in good prose.

One full example, start to finish

Quotes. Frequency: 40 a month. Time: 25 minutes each, between looking up client details, copying prices and writing the email. Damage if wrong: medium — a wrong price gets caught, but it looks bad.

That is 16 hours a month. The exit is the “high / high” box: AI pulls the client details and drafts the full quote from the price list, and a person reviews and sends. The 25 minutes become 4. You do not save the full 16 hours, you save 13 — and that difference is the one that almost never appears in the proposals people will sell you.

After automating

Every automation needs three things to avoid becoming debt: an owner who gets the alert when it fails, a way to check the result in one second, and a date to review it. Without the third, in a year you will have processes running that nobody remembers building.

And if the automation moves from executing rules to deciding on its own, it is worth naming the level of autonomy you granted it: that is what the Vergara Scale is for.