Automation. Start where it pays.

Which processes to automate with AI: the traits of a good candidate

Not every task is a good fit for automation, and the ones that look most impressive are rarely the best place to start. The strongest candidates share a pattern. They are repeated many times a week, they follow rules a colleague could explain in an afternoon, and a mistake can be caught before it does damage.

AI adds one more category to that list: tasks that start with unstructured input. Older automation tools needed tidy, predictable data. Today, models can read an email, a PDF, a form or a chat message and turn it into structured information. That opens up a lot of work that used to need a person only because someone had to read something first.

If a task ticks most of the boxes below, it is worth a closer look. If it ticks only one or two, it may be better left to people for now, or improved with simpler means.

• It happens often: daily or weekly, not a few times a year.
• It follows rules your team can describe, with known exceptions.
• It starts from text, documents or messages that someone has to read.
• The result can be checked against data or by a quick human review.
• A mistake is visible and fixable before it reaches a customer.

Common business processes that suit AI automation

The best opportunities usually hide in ordinary back-office work. Anywhere someone copies information from one place to another, sorts incoming messages, or writes the same kind of update again and again, there is a good chance a dependable automation can take over most of it.

These are starting points to compare against your own work, not a promise that each one fits. The same process can be a perfect candidate in one business and a poor one in another, depending on volume, how varied the inputs are and how costly an error would be.

• Reading orders, invoices or forms from email and entering them into a system.
• Sorting and routing incoming emails or support requests to the right person.
• Keeping data in sync between apps, such as a CRM and an accounting tool.
• Producing recurring reports and status updates from data you already have.
• Checking documents for missing fields or details that break your rules.
• Drafting standard replies for a person to review and send.

Processes that are poor candidates for automation

Some tasks look repetitive but depend on judgement that is hard to write down: negotiating with a supplier, handling an upset customer, or deciding on a hire. AI can help prepare that work, for example by summarizing a case, but the decision itself belongs with a person.

Rare tasks are also weak candidates. If something happens a few times a year, the effort of building and maintaining an automation will outweigh the time it saves. The same goes for processes that are still changing every month. Automating a moving target means rebuilding it every time it moves.

Finally, be careful where a single error would be very expensive and hard to detect. That does not rule automation out, but it means a review step is essential, and sometimes the honest answer is that a checklist or a better form will do more good than AI.

How to identify and prioritize automation opportunities

Start by listing the tasks your team repeats. Ask people where their week goes, and which tasks they would happily never do again. The people doing the work know exactly where the copying, checking and chasing happens.

Then estimate the value of each one. Count how often it happens in a normal week, time how long it takes, and note how often it goes wrong and what a mistake costs. You do not need precise numbers: rough counts are enough to see which tasks stand out.

Finally, weigh how hard each one would be to automate. A task with one input format and a system that accepts data easily is simpler than one with dozens of formats and an old system with no way in. The best first project combines clear value with low complexity, so you see results quickly and learn how automation fits your business.

• Volume: how many times a week does this happen?
• Time: how long does it take each time, including fixes?
• Errors: how often does it go wrong, and what does that cost?
• Inputs: how many different formats and sources are involved?
• Systems: can the tools involved send and receive data easily?

Why your first AI automation should be small

It is tempting to start with the biggest problem. A smaller first project is usually smarter. It shows quickly whether the approach works, it builds trust in the team, and it surfaces practical questions about data, access and review that every later automation will need answered.

Run the automation alongside your team before letting it act alone. Compare its results with theirs, fix the gaps, and only then let it handle the simple cases on its own while people review the rest. Once one process runs dependably, the next one is faster to build because the groundwork is already there.

How Walhallah helps you choose what to automate

We start by looking at your work with you: where the repeated tasks are, what they cost and which systems are involved. Together we pick one process with clear value and a realistic path, and we are honest when a task is better left alone. A focused prototype on your own examples often takes days, and a production version usually follows in weeks rather than months.

For Wegter, the starting point was email-based orders that had to be entered into SAP by hand. We built an automation that turns those emails into structured SAP workflows, removing a manual bottleneck. It is a good example of the pattern: frequent, rule-heavy work that starts with something a person had to read.