Which processes can you automate with AI?

The first question isn't "can AI do this?", but "what does this cost now?". Give every step of a process three numbers: how long it takes, with how many people, and how often a month.

An example. A task of forty minutes, with two people, eighty times a month. That's 1,280 hours a year. At an hourly rate of € 45, that's € 57,600 a year. Now you know what's at stake, and you can judge whether building pays off.

That way you avoid the classic mistake: starting with the process that irritates you most, instead of the process that delivers most. You can run the numbers straight away with the AI savings calculator.

This is the AI fit check from our APO method. Answer each question with no (0 points), partly (1) or yes (2). If a question doesn't apply to the task, it doesn't count.

1. 01Is it clear what goes in and what should come out? Can the person doing the work explain: I get this, I deliver that?
2. 02Is the information it needs digital? In a system, a mailbox or a folder. Not in someone's head.
3. 03Are there at least ten examples of good work? Input and good output, digital.
4. 04Is it done in a few steps? No long chain of judgements building on each other.
5. 05Does it go roughly the same way every time? A fixed pattern, not an exception every time.
6. 06Is a mistake cheap to fix? A mistake shows and is quickly corrected.
7. 07Can the outcome do without a detailed justification? If it needs a legal or in-depth explanation, the answer is no.

Add it up, 14 points at most. 11 or more is a strong opportunity, 8 or more a good one, 5 or more doubtful. Below that, the task isn't suitable (for now). Rather not count yourself? The check below does it for you.

A high score doesn't mean AI may do it alone. Three questions decide whether the human stays at the helm:

• Is original creativity the core? A campaign idea or a new concept stays human work. AI can make variants and first drafts.
• Does someone need to be physically present? A technician on site, a conversation at the table.
• Is a person personally accountable for the outcome? With advice, a diagnosis or a legal signature, AI does the groundwork and the human decides.

That leads to one of four roles: human (AI doesn't take part), AI assists (the human decides, AI does the groundwork), human approves (AI does the work, a person checks before it goes out) or fully AI. Most opportunities land on the middle two. Fully AI is the rarest role. More on that in human in the loop.

Most AI projects don't stall on the model, but on the data. It isn't there, it isn't digital, or there are no examples of good work to learn from.

That's why the method has a brake: if the data or the examples are missing, we calculate with 40 percent confidence at most, however well the task scores otherwise. And we never calculate with 100 percent saving: per step 90 percent at most, because exceptions, checks and management always remain.

A low score isn't a rejection, but an assignment. First collect the examples or make the information digital, then build. Also read why AI projects fail.

Processes that score high in practice, with what they delivered for clients:

• Customer service: answering customer questions. An agent prepares the answer with the right order information, an employee checks and sends it. At one client, the share of questions with a prepared answer went from 30 to 80 percent. See AI in customer service.
• Back office: processing requests from email and forms into their own system. At Van Berkel Professionals that saved 192 hours a month, with a person who approves.
• E-commerce: turning supplier product information into product copy in six languages. At EasyToys 132 hours a month.
• Sales: drafting proposals and quotes from fixed building blocks. In an ongoing engagement that went from 4 hours to 1 per proposal.
• Finance: booking and checking invoices.
• HR: screening CVs and summarizing candidates, with the decision staying with the recruiter.

All departments at a glance: AI by department.

• A process nobody can explain. If three colleagues do it three different ways, you're automating the chaos. Make the agreement first, then the automation.
• Work that hardly ever happens. An hour of work twice a year delivers too little to build for. A good prompt is enough then.
• Exceptions. Let AI do the ordinary work and leave the exceptions to people. That's exactly what people are good at.
• Decisions with big consequences, without a check. Money, health or someone's legal position always need a person who approves.

The seven questions tell you whether a task is suitable. They don't tell you which task should come first, what building costs and whether it pays off.

That's what the AI Quick Scan is for: one day at your site, with the people who do the work. We map the processes in measurable steps, test them as above and calculate per opportunity what it delivers and costs. Within a week there's a business case and a roadmap. The Quick Scan costs € 6,000 and is also available on its own.

You can also explore on your own first: with the free AI self-assessment you see in two minutes where your organization stands.