Which processes are suitable for AI?

The question "where can AI help us?" usually produces two kinds of answer. A long list of everything, or a paralysed "I don't really know". Both lead to standstill: a list without an order never gets finished, and someone who sees no candidate never starts.

There is a simple way in between. A few traits that make a process suitable, and a way to choose from the suitable processes where to begin. We walk through both below.

Not every process is suitable

AI can do a lot, and that is exactly why "it can do anything" is not a useful answer. It can write a text, summarise a document, answer a question, recognise a pattern. What it does well in your organisation depends on the process around it: how often it comes back, how clear the input is, and what goes wrong if the answer is off for once.

So the choice starts with your own work. Which of the processes you already run have the traits that let AI work. That is what you look at.

Four traits of a suitable process

The processes where AI already makes a difference share four traits. The more of them a process has, the more suitable it is.

• It comes back often. Work that repeats every day or week in the same form pays off faster than an exception you handle once a year. The gain adds up with every repetition.

• It costs noticeable time now. A process that eats up hours every week makes the gain visible and worth the effort. For a job of a few minutes there is little to win, even if AI can do it.

• It runs on information you already have. Reports, emails, files, data from your own systems. The more the input is already sitting somewhere, the easier AI can work with it. Work that leans on knowledge living only in someone's head is harder.

• A slip-up is not a disaster. Start with work where a person still checks it before it goes out: a draft, a summary, a first pass. Not with a process that lands unseen in front of the customer or in a legal document.

A process that has all four, say a weekly report you compile from your own data, is a strong first candidate. If a process is missing a couple, it can still work, the start just gets harder.

Which processes to leave for now

Just as useful is knowing what to skip for the time being. Four kinds of work are better left to rest:

• One-off or different every time. Work without a fixed pattern takes more effort to set up than it returns.

• High risk when it goes wrong. Processes where a wrong answer does direct harm, financial, legal or to a customer, are not a good first choice. There you want experience built up first.

• Messy or missing data. If the information is scattered, outdated or unreliable, AI only takes it the wrong way faster. Get the data in order first, then automate.

• Calls for judgement, empathy or responsibility. Making an assessment, breaking bad news, making a final call. AI can prepare it, the work itself belongs with a person.

These are not processes that can never work. They are processes you pick up later, once your organisation has more experience with it.

Weigh on impact and feasibility

Say you are left with a handful of suitable processes. How do you choose from those where to begin? With two questions per process.

• Impact: how much does it return if this works? Saved hours, shorter turnaround, fewer mistakes, more calm for your people.

• Feasibility: how easily do you get it working? Is the data ready, is the process clear, and is the margin for error manageable?

Set those two against each other. A process with high impact and high feasibility is your best starting point: a quick win that returns something fast and gives the proof for the steps that follow. High impact but low feasibility you keep for later, once you have more experience. Low impact you leave alone, however easy it is, because the effort does not weigh up against the gain.

You do not need to build a complicated scoring model for this. A short estimate per process, high or low on both, is enough to set the order.

From a list to a choice

The practical outcome is a short list. Write down the processes that have the traits, add a rough estimate of impact and feasibility per process, and pick the two or three that score high on both. That is where you begin.

More than two or three at once is rarely wise. Your attention scatters and no single process really gets finished. The rest of the list is your stock for the next round. Which processes you do first belongs in your AI direction, where you also record who leads them and where you are working towards. How to set out that direction is covered in Building an AI strategy: from experiments to direction. If you would rather start small with one process, Where do you start with AI in your organization? shows you how to go about it.

How we can help

The traits are simple. Applying them to your own organisation is harder, because you are so used to your own processes that you no longer see the opportunities in them. We are happy to take a look with you.

In the AI Strategy for Managers workshop we map out the processes with your management team and weigh them up together, so you go home with a few concrete candidates to start with. If you then want it built as well, from choice to working application, we pick that up in AI in Business.

Book a no-obligation call and we will look together at where the first gain is for your organisation.

Frequently asked questions

Sources

• Dutch Data Protection Authority (Autoriteit Persoonsgegevens), GDPR preconditions for generative AI (https://autoriteitpersoonsgegevens.nl/en/documents/gdpr-preconditions-for-generative-ai): relevant as soon as you choose a process that handles personal data.