Automation projects usually fail for a simple reason: a messy process gets automated, so the mess now runs faster. Good automation starts with understanding the work, not with choosing a tool.
Pick the right first process
Your first project should be small enough to finish and painful enough to matter. Good candidates share a few traits:
- Frequent: it happens daily or weekly, not twice a year
- Rule-based: most cases follow the same steps
- Digital: the information already exists in a system, form, or email
- Error-prone: mistakes cost time, money, or customer trust
- Well understood: someone can explain exactly how it works today
Typical examples include sending quotes and follow-ups, moving enquiries into a CRM, generating invoices and reports, chasing missing documents, and routing approvals.
Avoid starting with a process that changes every week, or one that only one person understands and nobody has written down.
Map the process as it really runs
Sit with the people who do the work. The official version and the real version are often different, and the differences matter.
For each step, write down:
- What starts it
- Who does it and which system they use
- What information they need
- What decision they make, if any
- What happens when something goes wrong
The last point is where most of the value is. Exceptions take up a large part of the effort in many processes, and they are what a simple automation misses.
Simplify before you automate
Once the map exists, question every step. Why does this need two approvals? Who reads this report? Why is this typed in twice?
Removing a step is cheaper than automating it, and it never breaks.
Choose the right level of automation
Not every problem needs the same tool.
| Situation | Usually a good fit |
|---|---|
| Two popular tools, simple data, low volume | A workflow tool such as n8n |
| Complex rules, high volume, or sensitive data | A custom integration |
| The process itself needs new screens or roles | Custom software |
| Unstructured input such as documents or free text | AI-assisted steps with human review |
Where AI helps, and where it does not
AI is useful when the input is messy and a good-enough first draft saves time. Examples include reading documents, summarising messages, classifying enquiries, and drafting replies.
It is a poor fit when the answer must be exactly right every time, such as tax calculations or payment amounts. Ordinary code is cheaper, faster, and predictable for those jobs.
A practical rule is to let AI prepare the work and let a person approve anything that carries real risk. You can reduce the review step later, once you have seen how it performs on your own data.
Plan for failure
Every automation eventually meets something unexpected. Before launch, decide:
- Who gets notified when a step fails
- How failed items are retried or handled by hand
- Where the activity log lives
- Who owns the automation after it goes live
An automation that fails silently is worse than the manual process it replaced, because nobody knows the work has stopped.
Measure it
Record a simple baseline before you start, such as time spent per week, error count, or how long a request takes from start to finish. Check the same figures a month after launch. This shows whether the work paid off, and helps you choose what to automate next.
Next steps
See how we approach manual processes and repetitive administrative work. If you already know which process is slowing you down, share a project brief and we will suggest a practical first step.
- #automation
- #business process
- #ai
Engineers and consultants at APARK IT Solutions writing about software delivery, web technology, and practical automation.