Back to blog

Which processes should you automate with AI?

Automate a process with AI when it happens often, follows rules a good employee could write down, and costs real hours or money every month. As a rule of thumb, a process that takes less than about 5 hours of work a month is not worth automating, and an AI agent that makes decisions needs roughly 50 or more decisions a month to pay for itself.

Most companies pick their first AI project by what looks impressive. The better way is to look for work that is frequent, repetitive and dull, where a person spends their day reading, checking and comparing.

Five questions that tell you if a process is worth it

1. Does it happen often? Daily or weekly work is a candidate. A task that comes up twice a year is not.
2. Could your best person write down the rules? If they can explain why each case went the way it did, an agent can learn to apply the same rules.
3. Can you check the result? There has to be a right answer you can recognise, or you will never know whether the agent is working.
4. Does it cost real time or money? Count the hours a month. Below about 5, leave it alone.
5. Does speed matter? A price change or a new regulation rewards whoever sees it first. Work that is late when done by hand gains the most.

Three or more clear yes answers make a good candidate. A no on question 2 or 3 usually means the process needs to be sorted out before anything can be automated.

Automation or an AI agent?

These are different tools. Automation runs a defined process the same way every time: copy the order into the ERP, send the weekly report, file the invoice. An AI agent handles work where the right action varies: which inquiries deserve a reply, which supplier offer is the best one, which lead is worth a call. If the rules can be written down completely, choose the simpler automation. It is cheaper to build and easier to keep running. See how we build these on our AI process automation page.

Processes that usually pass the test

• Sorting an inbox. Separating real customer inquiries from sales pitches, automated mail and spam.
• Checking supplier offers. Reading quotations, comparing them against the specification and asking for new quotes when a contract is old.
• Finding and qualifying leads. Researching companies that are likely to buy, and scoring them against what you sell.
• Watching a market. Following competitors, suppliers, prices and regulations every day, and summarising what changed.
• Moving data between systems. Forms into the CRM, orders into the ERP, documents into the right folder.

Processes that usually fail it

• Rare, one-off decisions, such as choosing a new office or a new product line.
• Work with no clear right answer, where two good employees would disagree and neither could say why.
• High-risk decisions under the EU AI Act, such as hiring or credit decisions, which carry heavy documentation duties. Our EU AI Act guide lists the categories.

An example from our own inbox

Before we build anything for clients, we run it ourselves. Our websites receive a steady flow of email, and most of it is not from buyers. Our inquiry agent reads every site mailbox twice a day. Of the 541 emails its first filter has logged so far, 155 were cold sales pitches and 134 were automated mail. It sets those aside, and only the rest goes on to be judged as a possible buyer. The work is frequent, the rules can be written down, and a wrong answer is easy to spot, so it passes all five questions above.

Once you have picked a process, our checklist of what to prepare before the build covers the next step.

Frequently asked questions

Which business processes are worth automating with AI?

Processes that happen often, follow rules a good employee could write down, and cost real hours or money every month. Typical examples are sorting an inbox, checking supplier offers, qualifying leads and watching competitors or prices.

How small is too small to automate?

As a rule of thumb, a process that takes less than about 5 hours of work a month is not worth automating. An AI agent that makes decisions needs roughly 50 or more decisions a month to pay for itself.

What is the difference between automation and an AI agent?

Automation runs a defined process the same way every time. An AI agent handles work where the right action varies from case to case. If the rules can be written down completely, choose the simpler automation.

Which processes should not be automated with AI?

Rare one-off decisions, work with no clear right answer, and decisions in high-risk areas such as hiring or credit, which carry heavy obligations under the EU AI Act.