Decisions, AI and Modernisation
AI in business processes: start with the bottleneck
When AI genuinely produces benefit — and when it merely makes an unclear process faster and more complicated.
The question now comes up in nearly every conversation about business processes that have evolved over time: could AI not solve this? It is a fair question. It is simply asked at the wrong point, as long as it is unclear what the process actually hangs on.
Where the bottleneck really sits
In business processes that have evolved over time, the bottleneck is rarely the speed of processing. It usually sits at one of three points:
At clarification. A case waits because it is unclear who decides or which rule applies. At completeness. Information is missing, contradictory, or spread across several places in different versions. At handover. Time is lost between two participants because the context has to be reassembled.
Not one of these bottlenecks is removed by making a single work step faster.
Where AI creates real value
There are places where the benefit is reliable:
Opening up unstructured input. When information arrives as free text, a document or an email and is transferred manually today, that is a clean use case — the task is well bounded and the result can be checked.
Making scattered knowledge findable. When an answer exists but nobody knows where, search is a real lever.
Suggestions within a defined frame. When a decision rule exists and a person owns the outcome, a suggestion can save time.
All three have something in common: the process is clear at that point. There is a defined task, a checkable result and an accountable role.
Why an unclear process gets worse
Put automation on an unclear point and three effects follow:
Throughput rises at a point that was not the bottleneck — the queue simply moves. Errors are produced faster and discovered later, because no person looks at them in between. And the rule the decision follows now sits in a configuration that even fewer people can follow than before.
Dependence on individuals has not shrunk. It has relocated.
The question to ask before any deployment
Before asking which technology to use, there is a simpler question: at which point in the workflow is most time lost today, and why exactly there?
If the answer is “because it takes a while”, automation is a sensible candidate. If the answer is “because it is unclear who decides”, it is not — then the decision has to be settled first. After that, using AI is a delivery question rather than a bet.