How to Choose a Process Worth Automating: A Checklist

The most common question we hear from clients isn’t “can this be automated?” — it’s “what should we actually automate first?” And that’s the right question, because this is exactly where success or a wasted budget is decided. Most automations that don’t work out don’t fail on technology — they fail because the wrong process was chosen. In this article we pull together six criteria and a simple checklist that tells you whether you have a strong candidate before you write a single line of code or buy a license.

Why the process matters more than the tool

According to McKinsey, current technology could in theory automate roughly half of the activities people are paid to do, yet fewer than 5% of occupations can be fully automated. In other words: you almost never automate someone’s whole job — you automate specific repetitive tasks inside it. The skill isn’t finding the technology; there’s plenty of that today. The skill is picking the right slice of the process.

In practice, companies tend to make one of two mistakes. Either they want to automate a process that runs twice a month and will never pay back, or they reach for something so complex and volatile that the automation turns into permanent maintenance. The criteria below help you avoid both traps.

Repeatability and volume: where the investment pays back

Automation has fixed costs — analysis, design, integration and ongoing maintenance. Those costs are spread across the number of repetitions, so the more often a process runs, the sooner it pays for itself. Sources consistently show that the highest ROI comes from high-volume processes — a rough threshold often cited is on the order of thousands of runs per month. A process that runs a few times a year may be annoying, but automating it rarely pays off.

What to do: Before you start, calculate two numbers — how many times a month the task repeats, and how many minutes one run takes. Their product gives you the monthly time load, the basis of every payback calculation.

Structured input data drives accuracy

An automation is only as good as the data it receives. Processes working with structured or at least semi-structured data — forms with consistent fields, tables, standardized documents — have a much higher accuracy ceiling than processes that depend on free text and contextual judgment. The messier and less uniform the input, the more exceptions arise and the sooner a human has to step in.

The good news is that with large language models, the line of what’s “manageable” has shifted — today we can reliably handle even semi-structured documents such as invoices or orders in various formats. Still, the rule holds: the cleaner and more uniform the input, the cheaper and more stable the automation.

What to do: Look at where the data comes from. If it’s emails with attachments in ten different formats, expect a more expensive solution than a clean export from a single system.

Error rate and cost of rework: where manual work hurts most

Manual data handling carries an inherent error rate. Studies and practice report that manual transcription error rates typically hover around 1%, rising to 1–4% for more complex tasks. Across a few dozen records that’s negligible; across tens of thousands it means hundreds of errors — and every correction, complaint or mis-issued invoice has a price.

Processes with a high error rate are therefore attractive candidates, because automation here saves not just time but also the cost of chasing down and fixing mistakes. But watch one thing: the automation has to reach an accuracy threshold — the point below which it doesn’t displace the manual effort it was meant to replace. If a human still has to check every output, the savings evaporate.

What to do: Estimate how many errors the process makes today and what fixing them costs. Where errors are expensive, the value of automation climbs fastest.

Rule stability: automate what rarely changes

An automation replicates the rules that were valid when it was built and relies on systems and procedures not changing fundamentally. That’s why UiPath and other sources advise choosing tasks that are clearly defined and happen the same way every time. A process that changes its rules, forms or systems every quarter easily turns from automation into endless maintenance — and the cost of updates eats the savings.

This doesn’t mean automating only trivial things. It means looking for rule-based processes: decisions driven by clear rules, a standardized flow and only a limited number of exceptions. According to Blueprint, clarity and standardization of rules is the first filter every candidate should pass.

What to do: Ask the people who run the process: “How often does the procedure change?” If they say “almost never,” you have a good candidate. If “almost constantly,” shelve it.

Payback: a simple calculation to point you in the right direction

All the criteria ultimately converge on one number — payback. You don’t need to build a complex model. Just compare the annual saving (hours saved × cost per hour + cost of errors) with the cost of development and maintenance. This is precisely why volume and error rate matter so much: the fixed costs of automation are spread across the number of repetitions, so two processes of similar complexity can have completely different payback purely because of volume.

And the motivation isn’t only financial. UiPath reports that 74% of organizations spend more than a quarter of their time on repetitive manual tasks — the time freed by automation can be redirected to work a machine can’t do.

A quick checklist for choosing a process

Before you send a process into automation, run through six questions. Add a point for every “yes”:

1. Repeatability — does the task run essentially the same way every time?
2. Volume — does it repeat often enough (on the order of hundreds to thousands per month)?
3. Structured data — do inputs arrive in a consistent, readable format?
4. Error rate — is it done manually today, and are the errors costly?
5. Rule stability — does the procedure change only rarely?
6. Payback — will the investment return within a reasonable time (typically under a year)?

Five points or more means a strong candidate worth starting with. Three to four points call for a closer look — the process can often be “straightened out” or narrowed so it makes sense. Two or fewer is a signal to leave the process alone for now and come back when the conditions change.

Summary

Choosing a good process isn’t about hunting for the most complex technology — it’s about honestly answering six questions: repeatability, volume, data structure, error rate, rule stability and payback. Processes that pass this filter give automation its fastest and most certain effect — and they keep you away from projects that look tempting but never pay for themselves.

If you suspect you have processes ripe for automation but aren’t sure which ones, we’re happy to walk through them with you. At DIGITAL WOLF this is exactly where we start — before we build anything, we assess together where automation has the fastest payback and where it isn’t worth it yet. Get in touch and we’ll go through it on concrete numbers.