How to spot tasks worth automating
Not everything is automated. You pick the repeated task whose mistake or lost hour is already felt, and whose exception can be handled.
Automating what looks good
The task that demos well is rarely the one that takes the hours. A polished notification matters less than the reconciliation someone rebuilds every Friday.
An endless list of things that could be automated does not help either. Without an order, the project starts with the favourite and stalls at the first exception.
If the rule cannot be explained, automating it only hides the mess. First you write what comes in, what goes out, and what happens when a figure is missing.
What to look at in a task
Frequency: how often it is done the same way. Volume: how many cases. Pain: what happens if it is forgotten or copied wrong.
Exceptions: how many cases do not fit the rule. If almost all of them are special, there is no flow. There is human judgement.
Fit with what already exists. Sometimes the current program can do it. Custom development starts when that piece cannot.
When a task is a candidate
- It repeats and the rules fit a few cases.
- The mistake shows up late: an amount, a customer, a status.
- There is a clear source and a clear destination.
- Someone can review the exception without redoing everything by hand.
What is delivered if it goes ahead
The flow
The task runs from a trigger, not from a reminder.
The control
A record of what happened, an alert if it fails, and a queue of exceptions.
The note
What to watch, how to rerun it, and what must not be edited by hand.
How the first one is chosen
Usual means
- The program you already use
- Python
- Queues and logs
- APIs
- Sheets as input, not as the system
A typical situation
The invoice that should come from the order
If the order is already in a system and the invoice is copied, that copy is a candidate. The rare case — a discount that is not on the price list — stays in a queue. It is not automated blindly.
Common mistakes
- Saying that everything will be automated.
- Starting with the hardest integration.
- Leaving no trace when the flow fails.
- Not teaching the team which exception is theirs.
Frequently asked questions
Does every task need a bot?
No. Sometimes it is a rule in the current program or a connection between two tools you already have.
What if the task changes every week?
Then it is not ready. The rule has to settle first.
Does this guarantee fewer mistakes?
No percentage is promised. There is a record of what failed.
Does AI decide alone?
Only when the task allows it, and with a review when the case is unclear.
Where is the automation detail?
On the process automation page. This guide only helps choose where to start.
Related guides
Is there a task that repeats every week?
Describe it in four lines: where the data comes from, where it goes, and what happens when it does not match.