Which business processes are worth automating?

6 min read·Updated October 5, 2026

Start with the process, not the technology

The question “what app do I need?” almost always comes too early. First you need to understand what happens in the business today, step by step. Automating means having a system carry out steps a person does today; if those steps are messy, the system will just repeat the mess faster.

Pick a process that hurts and answer these questions in writing:

• Who does it? One person, several, or “whoever is free”.
• How often does it happen? Daily, weekly or monthly.
• How long does it take each time? Measure it for a week; don’t estimate from memory.
• Where do the errors happen? And what does it cost to fix them.
• What information does it need? Where it comes from and who provides it.
• What happens next? Who receives the result and what they do with it.

With one page of answers you already have more clarity than many businesses that start asking for software quotes. Sometimes that exercise is enough to uncover a step that isn’t needed or an approval nobody uses.

It isn’t just common sense. Studying how companies adopt AI, McKinsey found that the biggest gains come from redesigning entire workflows, not from automating isolated tasks (McKinsey Global Institute, 2025).

Five signs a process can be automated

1. 01

   It repeats the same way every day

   If someone follows exactly the same steps, without making new decisions each time, a system can follow them too.

2. 02

   It depends on copy and paste

   Moving data from email to a spreadsheet, from the spreadsheet to the accounting system and from there to a WhatsApp message is work that adds nothing and makes mistakes easy.

3. 03

   It causes frequent errors

   A system can check information before letting it move on: no missing fields, totals that add up, a customer that exists. Research on spreadsheets has concluded for years that their errors are common and non-trivial (Panko, 2008).

4. 04

   It needs manual follow-up

   If someone has to remember who should do what and when, and chase it over chat, the process depends on that person’s memory.

5. 05

   It grew faster than the way it’s done

   What worked with 10 customers becomes unmanageable with 500. The typical sign is hiring people just to keep the same process running.

These signs match what studies show. A McKinsey analysis of more than 2,000 work activities in the United States found that collecting and processing data takes about a third of working time, and that both activities have a technical automation potential above 60% (McKinsey, 2016).

With AI, that margin grew. In its 2025 update, McKinsey estimates that already demonstrated technology could, in theory, automate activities that add up to about 57% of US work hours, and it stresses that this is technical potential, not a forecast of job losses (McKinsey Global Institute, 2025). The rule doesn’t change: repetitive, data-based work is still the best candidate.

What you shouldn’t automate

The same studies show the other side. In the 2016 analysis, the hardest activities to automate were managing and developing people (9% potential) and applying expertise to decide, plan or create (18%). The 2025 report reaches a similar conclusion: interpersonal skills, such as negotiating or coaching someone, are the ones that change least. Machines are very good at well-defined tasks; people are still needed to set goals, interpret results and apply common sense.

In a business, that means keeping these manual:

• Decisions that take judgment: approving a doubtful credit, negotiating with a supplier, handling a delicate complaint.
• What rarely happens: if it occurs twice a year, building and maintaining the automation costs more than doing it by hand. Maintenance matters: in an OECD survey of more than 2,000 small and medium businesses in twelve countries, automation was the most cited benefit of digital tools (42%), and maintenance costs one of the main obstacles (39%) (OECD, 2026).
• What is still changing: if the process changes every month, wait for it to settle.
• Moments when the customer expects a person: automating a conversation the customer values can backfire.

Being able to automate something doesn’t mean it’s worth it. McKinsey warns about this: technical feasibility is a necessary condition, but the cost of building the solution and the real benefits it brings also count.

A middle ground that works well: automate the preparation and leave the decision to the person. The system gathers the data, checks it and sends an alert; someone reviews and approves.

Four levels: from tidying the process to owning software

Automating doesn’t mean commissioning an app. There’s a ladder, and it pays to climb it one step at a time:

1. 01

   Improve the process

   Remove steps, name an owner, use one format. It costs nothing in technology.

2. 02

   Automate a task

   A form instead of receiving data over chat, a template, a scheduled reminder in the tools you already use.

3. 03

   Connect your tools

   Link the systems you already have so data is typed only once: a recorded sale that reaches accounting on its own, for example.

4. 04

   Build your own software

   When the process is the core of the business and no standard tool fits your rules.

Franco Cerámicas is an example of the fourth level. It tracked its sales in a spreadsheet and needed to know, for every sale, how much the sales rep earns and how much profit the company keeps. Those commission rules are its own, so the answer was a custom web app connected to its accounting software.

If an earlier step solves the problem, stay there. If you’re between levels 3 and 4, the custom software vs. SaaS guide (linked below) helps you decide.

How to decide what to automate first

List your processes and rate each one in three columns: how often it happens, how much time it takes and how many errors it causes. Use only “high”, “medium” or “low”; you don’t need more precision to prioritize. Here’s what it could look like:

How to read your own table:

• Three “highs”: start there. That’s where the investment pays back fastest.
• High frequency, but little time and few errors: usually solved with a simple automation (level 2).
• Low frequency and a lot of time: it probably needs human judgment. Before automating, check whether the process can be simplified.

Start with a single process, measure for a week or two how it turned out, and only then move on to the next. A small automation people actually use is worth more than a big system nobody adopts.