The fastest way to waste time on automation is to start with the tool.
A better starting point is the workflow.
Before reaching for AI, Zapier, scripts, APIs, or custom software, I like to ask a simpler question:
Which tasks are repeated often, consume real time, create errors, and follow predictable rules?
You can get surprisingly far with a one-hour audit.
The goal is not to map the entire company.
The goal is to produce a ranked list of automation candidates.
Why Automation Projects Often Start Backwards
The usual sequence looks like this:
New tool discovered
↓
Team gets excited
↓
"What can we automate?"
↓
Random workflow selected
I prefer this sequence:
Operational friction identified
↓
Workflow understood
↓
Automation candidate ranked
↓
Technology selected
That order matters.
If you automate a bad process, you often get a faster bad process.
If you understand the workflow first, the right technical solution becomes much easier to see.
Minute 0–10: List Repeated Work
Start by listing anything people do repeatedly.
Do not filter the list yet.
Examples:
Copy form data into CRM
Prepare weekly report
Send recurring reminder emails
Check invoice status
Update order status
Move files between systems
Create records from email requests
Manually assign incoming leads
Frequency is the first signal.
A five-minute task performed once a month is probably not urgent.
A three-minute task performed fifty times a day might be.
You can use a simple score:
1 = Monthly or less
2 = A few times per month
3 = Several times per week
4 = Daily
5 = Many times per day
The exact scale is not important.
Consistency is.
Minute 10–20: Estimate Time Spent
Next, estimate the human time required for each task.
Do not only count the visible action.
Include things like:
Open system
Find record
Copy data
Switch app
Paste data
Verify result
Save
Return to original task
Context switching is part of the cost.
A useful rough score:
1 = Under 2 minutes
2 = 2–5 minutes
3 = 5–15 minutes
4 = 15–30 minutes
5 = More than 30 minutes
Now combine frequency and time mentally.
This is where small repetitive tasks often start looking much bigger.
Minute 20–30: Look for Error Risk
Manual work becomes much more expensive when it creates rework.
For each task, ask:
How often does this go wrong?
What happens when it does?
Who has to fix it?
Typical error-prone workflows include:
copying data between systems,
manual calculations,
status updates,
invoice entry,
file naming,
duplicate data entry,
report preparation.
A simple score could be:
1 = Almost no meaningful errors
2 = Occasional minor errors
3 = Errors require correction
4 = Errors regularly create rework
5 = Errors affect customers, money, or operations
The real cost of a task is often:
Original work
- checking
- correction
- interruption
- downstream impact
That changes the priority dramatically.
Minute 30–40: Find Waiting Time
This is one of the most useful parts of the audit.
Some workflows are slow even though the actual human work is tiny.
For example:
Request submitted
↓
Wait 3 hours
↓
Someone copies data
↓
Wait until next day
↓
Another person reviews it
↓
Next team continues
Maybe the total manual effort is eight minutes.
But the process takes more than a day.
That is a strong automation signal.
Ask:
Where does work stop even though nobody is actively working on it?
Score delay roughly:
1 = No meaningful delay
2 = Occasional short delay
3 = Often delayed by hours
4 = Commonly delayed a day
5 = Regularly blocks customers or teams
Sometimes the best automation is not automating the task itself.
It is triggering the next step immediately.
Minute 40–50: Check Predictability
Now ask whether the workflow follows rules.
Automation works well when the process looks like:
IF condition
THEN action
ELSE route exception
For example:
IF invoice overdue > 7 days
THEN send reminder
AND notify account owner
That is a good candidate.
Compare that with:
Customer disputes contract
Determine best commercial response
That requires judgment.
Not everything should be automated.
I generally look for workflows with:
Predictable input
Repeatable steps
Clear rules
Predictable output
Limited exceptions
If every case is different, automation complexity goes up quickly.
Minute 50–60: Add the Tools and Rank the Candidates
Now document which tools are involved.
For example:
Task
Tools
Lead entry
Website form → Email → CRM
Weekly report
CRM → Spreadsheet → Email
Invoice reminder
Accounting system → Email
Support routing
Email → Ticketing tool
Order update
Store → ERP → Customer email
This matters because the technical effort depends heavily on system boundaries.
Two tools with clean APIs may be easy to connect.
A legacy desktop app, PDF workflow, and undocumented approval process may be much harder.
Now create a simple scoring table.
Task
Frequency
Time
Error Risk
Delay
Complexity
Copy leads to CRM
5
2
4
3
1
Weekly report
3
5
2
2
2
Invoice reminders
4
2
2
4
1
Contract negotiation
1
4
5
3
5
You can use a rough priority model:
Priority =
Frequency
- Time
- Error Risk
- Delay
- Complexity
This is not meant to be mathematically perfect.
It is there to create comparison.
Before You Automate, Ask Whether the Step Should Exist
This is important.
Imagine the process is:
System A
↓
Export CSV
↓
Clean in spreadsheet
↓
Email file
↓
Import into System B
You could automate all of that.
But maybe the better solution is:
System A
↓
API integration
↓
System B
The spreadsheet and email may only exist because the systems were never connected properly.
Automation should remove unnecessary work where possible.
Do not preserve every historical workaround just because it is familiar.
A Good First Automation Is Usually Boring
I would not start with the most ambitious workflow.
I would start with something that is:
Frequent
Predictable
Painful
Low-risk
Measurable
Technically manageable
Examples might be:
lead routing,
status updates,
invoice reminders,
report generation,
record creation,
repetitive notifications.
A smaller automation gives you useful feedback.
You learn:
Is the process really understood?
Is the source data clean?
Are there hidden exceptions?
Will users trust the automation?
Did the automation remove work
or just move it somewhere else?
Those lessons matter before tackling a more complex system.
Do Not Use AI Just Because AI Is Available
Some automation problems need AI.
Many do not.
If the workflow is:
When X happens, update Y
you probably do not need a language model.
Traditional rules, APIs, database triggers, queues, scripts, or workflow engines may be more reliable.
AI becomes useful when the task involves things like:
Classification
Summarization
Natural-language extraction
Unstructured documents
Draft generation
Ambiguous input
Even then, I would still define the workflow first.
The model is one component.
It should not become the architecture.
A Developer-Friendly Decision Checklist
Before you automate a task, ask:
[ ] Does it happen frequently?
[ ] Does it consume meaningful human time?
[ ] Does it create avoidable errors?
[ ] Does it create waiting or handoff delays?
[ ] Are the rules reasonably predictable?
[ ] Are the inputs structured enough?
[ ] Are the systems technically accessible?
[ ] Can exceptions be routed to a human?
[ ] Can success be measured?
[ ] Should the task exist at all?
If most of those answers are clear, you probably have a strong candidate.
What I Would Do First
If I had one hour with a small engineering or operations team, I would leave with three things:
A list of repeated workflows
A scored shortlist
One small automation candidate
That is enough to move from vague automation interest to an actual engineering decision.
The point is not to automate everything.
The point is to identify where human attention is being spent on work that does not require human judgment.
That is usually where automation creates the most useful leverage.
I write about this kind of workflow and software decision-making at Ksoft Technologies. This article is adapted from my original guide on finding which tasks to automate first.
Top comments (0)