A patch on a wooden leg: when does AI automation make sense?
There's no shortage of agencies that come to solve your problem with AI. A patch here, a patch there. A year later you have many patches, a lot of maintenance, and still no overview.
Short answer. Only once customers, orders, stock and invoices exist once, in one system. Automation on top of a messy foundation only makes the mess faster. After that: rules for what is fixed, AI only where interpretation is needed, safely, with a policy and human control.
Automating on top of a mess
Automation does exactly what you tell it, faster. If your customer data lives in three systems, a smart integration copies the errors three times as fast. If your quoting process is wrong, you get wrong quotes with less manual work. AI doesn't make a messy base orderly. It makes it efficiently messy.
First one central system
That's why we never start with automation. First one central system in which customers, orders, stock and invoices exist once. Then the data in it clean. Only then do you look at which tasks your team still repeats every day, and which of those a script, a rule or a virtual colleague can take over. Usually it turns out half the patches are no longer needed: the wound they were covering no longer exists.
What automation does do well
Retyping data between systems that stay side by side for a good reason. Sorting emails and tickets to the right person. Generating documents and reports from what's already in the system. Reminders and approvals that today live in someone's head. All tasks with a clear rule and a measurable result.
When AI, and how
Not everything that's automation has to be AI. A task with a fixed rule, where the same input always gives the same outcome, you solve with a script or a rule in your system. Cheaper, predictable, explainable. AI you use where interpretation is needed: reading an email and routing it to the right person, extracting a supplier invoice from a PDF, summarising a long file. It's good at that. But it sometimes guesses, and it doesn't say so.
That's why three things are non-negotiable. One: safe. Which data may go to which model, and where does it run? Customer data, personnel data and contracts don't just go to a free chatbot. Two: a one-page policy that everyone knows. What's allowed, what isn't, and how you check a result before you use it. Three: human control built in. Every AI step has a point where someone looks, certainly for money, customers and decisions that can't be undone. We set up those three with you, before the first automation runs.
People keep deciding
The repetitive work goes to software. The decisions stay with your people. A good automation always has a place where a human looks: the exception, the large amount, the customer who wants it differently. That's not a limitation of the technology, but a choice.
The order
First one central system, then clean data, then automation. Fewer tools instead of more. And every automation measured on the whole: how many hours a week come back, and what changes for the departments before and after? Because a patch helps too, in one place. It just lacks the overview and thinks locally. What saves an hour in one department sometimes costs two in the next.
What that looks like in euros, with three real proposals: baling wire between the applications →
Whoever sells you an AI solution without first looking at your whole company is selling you patches.
One insight from the floor a month.
No product news, no ten emails. Once a month, what we saw on the floor and what you do with it. Unsubscribe anytime.