Is this process worth automating? How to tell before you spend
Not every annoying job is worth automating. A few minutes with a calculator and five honest questions will tell you whether a process deserves the money, or a quick fix instead.
By Gary Kellett5 min read
For directors and operations leads deciding which job to automate first.
Everyone has a job they would love to see automated. Usually it is the one that annoyed them most recently. That is a poor way to choose. The right first automation is the one with the biggest gap between what it costs you today and what it would cost to run automatically, and you can estimate that gap in an afternoon.
Start with what the job costs today
You do not need a consultant for this part. You need four numbers, and a rough answer for each is fine.
Step 1: How many people do it?
Count everyone who touches the job, including the person who checks it.
Step 2: How long does it take each time?
Ask the people who do it, not their manager. Their estimate is usually closer.
Step 3: How often does it happen?
Daily, weekly, per order, per invoice. Turn it into a yearly count.
Step 4: What does an hour cost you?
Use a loaded hourly cost: salary plus employer costs, divided by working hours.
552 hours
a year spent on one weekly job, in a simple worked example
Multiply hours by your loaded hourly cost and you have the yearly cost of the job. Then add what errors cost: credit notes, re-deliveries, late payment penalties, the customer who did not come back. Errors are often the larger number, and they are the one people forget.
Five questions that separate good targets from bad
Once you know what the job costs, these questions tell you whether automation is the right fix.
- Is the input predictable? A web form or a supplier's standard invoice is easy. Free-text emails in any layout are harder, though modern AI handles far more variety than it used to.
- How many exceptions are there? If one case in five needs a judgement call, plan for a person to handle those. If most cases are exceptions, automation will disappoint.
- Can a system reach every piece of data the job needs? If a step relies on someone's memory, capture that knowledge first.
- Is the process stable? Automating something that changes every month means paying to change the automation every month.
- Who owns it? A named person who does the job today should help design the new version and judge whether it works.
Watch out: The exception trap
Most automations that disappoint do so because of exceptions nobody counted. Before you commit, ask the team to keep a simple tally for two weeks: how many items went through cleanly, and how many needed someone to think. That tally is worth more than any vendor demo.
When the answer is 'fix it by hand'
Sometimes the honest answer is that the job is not worth automating. If the yearly cost is small, or the process is about to change, a better spreadsheet template, a standard feature in your existing system, or simply removing a step will give you most of the benefit for almost no money.
I would rather tell a client that than build something they do not need. A surprising number of 'automation' projects turn out to be a ten-minute configuration change in a system they already pay for.
Comparing the cost to automate
On the other side of the scales sit the build cost and the running cost. Running cost is the one people miss: hosting, AI model usage, monitoring, and someone fixing it when a supplier changes their invoice layout. A good automation earns back its build cost within months and keeps earning after that. If the sums only work over several years, look again.
This is also why I offer to prove an automation on a client's real process before they pay for it. If the numbers do not stack up on real data, it is better to find out in a week than after a three-month project.
A quick way to rank your candidates
List three to five jobs. For each one, write the yearly cost today, the rough exception rate and whether the data is reachable. Rank by yearly cost, then strike out anything with a high exception rate or unreachable data. The job at the top of what is left is your first target.
Share the list with the people who do the work before you decide. They will often know about a cost you missed, such as the hour every Friday spent fixing what went wrong during the week, or a customer who complains every time an order is late. They will also tell you which jobs they would be glad to lose, which matters: an automation the team wants is adopted in days, while one imposed on them is quietly worked around. A short conversation now saves a disappointing project later.
If you want to skip the spreadsheet, the process cost calculator does the arithmetic for you and shows the yearly figure in a couple of minutes.
Questions directors ask
How much does a process need to cost before it is worth automating?
There is no fixed threshold. Compare the yearly cost of the job today, including errors, with the build and running cost of the automation. If the automation would not pay back within the first year, look for a cheaper fix first.
What makes a process hard to automate?
Unpredictable inputs, a high rate of exceptions that need judgement, data the system cannot reach, and a process that keeps changing. Count exceptions for two weeks before you commit.