What Business Processes Should You Automate First?
A scoring method for choosing, and why the most annoying task is usually not the right place to start.
Short answer
Score each candidate on frequency, how clearly the rules can be stated, and how expensive a mistake would be. Automate something frequent, clear and cheap to get wrong. The most irritating task is usually the one with the most exceptions, which makes it the worst first choice.
The instinct is to automate whatever is most annoying. That is usually wrong, because tasks are annoying precisely because they are full of exceptions and judgement — which makes them the hardest thing to automate and the worst place to learn.
Score each candidate
Three questions, scored 1 to 5.
Frequency. How often does it happen? Daily beats monthly. A perfectly automated quarterly task saves four instances a year.
Clarity. How completely can you write down the rules, including the exceptions? If you cannot describe it precisely, you cannot build it reliably.
Cost of error. What happens if it does the wrong thing? Low is good for a first project — you want to learn on something forgiving.
Multiply frequency by clarity. Divide by cost of error. Start at the top.
What typically scores well
| Process | Frequency | Clarity | Cost of error |
|---|---|---|---|
| Enquiry capture and notification | High | High | Low |
| Appointment reminders | High | High | Low |
| Invoice generation | High | High | Medium |
| Report generation | Medium | High | Low |
| Data sync between systems | High | High | Medium |
| Invoice chasing | Medium | High | Low |
What typically scores badly as a first project
| Process | Why |
|---|---|
| Customer complaints | Judgement, high cost of error |
| Pricing and quoting | Exceptions everywhere |
| Anything legally binding | Cost of error too high |
| Hiring decisions | Should not be automated at all |
| Anything you cannot describe | Clarity score of 1 |
The question people skip
Should this process exist at all?
Mapping frequently reveals steps that exist because someone once did it that way. A report nobody reads. An approval nobody has ever declined. A form that gets copied into a system that gets copied into another system.
Deleting a step is cheaper than automating it, faster, and has no ongoing failure mode.
Start small and prove it
One process. Build it. Run it for a month. Confirm it works and that failures are visible.
Then the next one.
Businesses that automate five things simultaneously end up with five half-trusted workflows and go back to doing it manually while leaving the automations running, which is the worst of both.
What "done" means
An automation is finished when:
- it runs without anyone thinking about it
- failures surface somewhere a person actually looks
- someone other than the person who built it can explain what it does
- the manual fallback is documented, for when it breaks
That last one gets skipped and matters more than it sounds. Every automation will fail eventually, and the question is whether anyone remembers how the work used to get done.
The AI automation service starts with this prioritisation, and automating repetitive business processes covers the build sequence.
Worked examples
Related services
Related reading
AI & Automation
What Is AI Business Automation?
What it is, what it is not, and why the unglamorous first step is the one that determines whether any of it works.
AI & Automation
AI Automation vs Traditional Business Automation
Most processes need a rule, not a model. Here is how to tell which one you are looking at.
AI & Automation
How AI Can Automate Lead Follow-Up
Following up reliably without sounding like a sequence. The line between helpful and irritating is narrower than most tools assume.
AI & Automation
n8n vs Make vs Zapier: Choosing an Automation Platform
Three platforms, three different ways of charging you. The billing unit decides more than the feature list does.