Automation makes sense when it makes the work better. Whether that requires AI remains to be seen.
We see it regularly: someone wants to automate something and already assumes AI is the answer. Yet we still need to work out exactly what needs to happen.
An article from MIT Sloan got me thinking about this again. The researchers stress that successful applications start with a business problem. First show that AI works better than the alternative, then scale up.
That alternative deserves more attention.
If an order above a certain amount needs approval, you can simply make that a rule. An AI model does not need to think it through every time. In a case like that, fixed rules are clearer, testable and predictable.
We do use AI ourselves to help discover and put together those rules. By examining data, looking for patterns and bringing exceptions to light. We check those suggestions with the people who know the work. Just because something happens often in a dataset does not mean it is the right way to work.
Once the agreements are clear, ordinary software can carry them out. AI can play an important role in building a solution without that solution needing AI for every decision.
For me, a project therefore starts with questions. Where are employees losing time? Why is a customer still waiting? Which mistakes keep coming back? Only when we understand that can we choose a useful solution.
That is why tailoring a solution matters. Two companies can use the same software and still work very differently. Their agreements, exceptions and responsibilities differ. Sometimes adapting an existing package is enough. Sometimes integrations or new software are needed.
Anyone who decides in advance that it has to be AI skips an important part of that investigation. I first want to know what we can improve for customers and employees. Then we choose how to do it.