How to automate business processes without breaking anything
The right order to automate: which processes are worth it, which you should never touch, and why most projects fail during the mapping stage.
We run thirteen brands with a small team. That is not talent: it is that almost everything repetitive is automated. This is the order we did it in, and the mistakes we paid for on the way.
First: what NOT to automate
This is the part everyone skips, and the part that decides the outcome.
Do not automate a process you do not understand. If nobody can explain why step four exists, automating it only makes the error happen faster and more often.
Do not automate a process that is about to change. If the department is reorganising in two months, wait. Automating on sand is throwing away the work.
Do not automate judgement. Deciding whether a client is a fit, whether a text is good, whether a price is fair: that can be assisted, not delegated. Automation that replaces judgement produces uniform, mediocre results.
The order that works
1. Map with a stopwatch, not with meetings
This is where most projects die. Someone runs a workshop, people describe how they think they work, and that idealised version gets automated. The one that does not exist.
What works is measuring. For two weeks, record how often each task happens and how long it takes. Two surprises almost always show up: a task that looked minor eats 30% of the week, and one everyone called critical happens three times a month.
2. Rank by frequency × friction
Not by what looks impressive. The formula is simple:
Times per month × minutes per time × how much people hate doing it
That third factor is not sentimental. Tasks people dislike get postponed, and postponed tasks generate the expensive errors.
3. Start with the boring one
The first automated process should not be the most ambitious, it should be the one with the least risk if it fails. Moving data from one place to another. Generating a report someone copies by hand. Flagging a deadline before it arrives.
It does two things: it delivers value in days, and it teaches the team how the system behaves before you trust it with anything important.
4. Always leave the manual door open
Every automated process needs a way to do it by hand when it fails. No exceptions. The day the integration goes down — and it will — operations cannot stop.
Where artificial intelligence fits
With a distinction that saves a lot of money:
Classic automation for anything with rules. If this, then that. Cheap, predictable, and it does not hallucinate. Most of what you want to automate falls here, and using a language model for it means paying more for less reliability.
Language models for things that have language and no rules. Classifying inbound email, pulling data from invoices in different formats, summarising a call. That is where they earn their cost.
Agents for chains of decisions that depend on context. It is the frontier and also the most fragile: reach for it when the above is not enough, not before.
What it actually costs
The line nobody budgets is maintenance. An automation is not a building you hand over, it is an installation you maintain: APIs change, formats change, the company changes.
Budget 15-20% of the build cost annually just to keep it running. If a supplier never mentions this, they have never maintained what they build.
The sign you are getting it right
It is not how many processes you have automated. It is that the team stopped asking to hire someone for tasks nobody does any more.
At The Market Club the artificial intelligence department was born the day two people could not keep up with four mastheads. Today it is thirteen brands with the same team. We do not use artificial intelligence: we build it, for ourselves first.
If you want to see how it would apply to your operation, let’s talk for thirty minutes.