Kennisbank

Insight: you see through AI, but are not yet steering on it everywhere

What insight looks like

At the insight level, something has changed that goes beyond isolated experiments. The organization not only understands that AI does something, but also why it works in one place and not in another. A language has emerged to have that conversation: people talk about data quality, about responsibility for errors, about which processes are and are not suited to automation. That is a different conversation than a year earlier, when the question was mainly whether a pilot delivered anything.

What is characteristic of this level: insight is present, but it is not yet anchored everywhere in how decisions are made. One department has a sharp view of which data it needs and which is missing. Another department is still following a pilot in good faith without anyone having established whether the underlying infrastructure can support it. Insight is therefore not an even layer across the organization. It is more a collection of places where the light has come on, with sections still standing in the dark in between.

What is still missing

The step to the next level, intelligence, requires that this insight translate into steering. Not incidentally — as a reaction to a problem that arises — but structurally, as part of how the organization normally works. This touches the same fundamental dimensions that were already the determining factor at earlier levels: is data management set up in such a way that decisions based on AI outcomes can be accounted for? Is the IT infrastructure set up so that new applications do not each time start from scratch? Is the organization itself shaped so that responsibility for AI-driven processes lies somewhere, rather than being spread thinly everywhere?

At the insight level you often see that these questions have indeed been asked, but that the answer differs per dimension. One dimension is already far enough along that steering is possible; another is still trailing behind isolated initiatives. That difference in pace is precisely where the spread in a plotting round becomes visible: if several people score separately, it becomes clear that one respondent considers the organization already far advanced on data management, while another places that same dimension at foundation. That spread is not noise. It shows which dimension can bear attention first, and which dimension still has to follow.

The sequence that determines it

The temptation at this level is to focus attention on the dimensions that are most visible — often the dependent dimensions, close to daily work. But the fundamental dimensions take precedence. If data management is not in order, any steering based on AI outcomes remains shaky, however much insight has been built up. If the IT infrastructure is fragmented, every application remains an island, even if the organization can already talk about it well internally.

This is not a different sequence from the one that applied at earlier levels. It is the same sequence, but at a point where the cost of skipping it becomes more visible. At baseline and foundation, the absence of a foundation could still be hidden behind the fact that little was happening. At insight, more is already in motion, and that makes a weak fundamental dimension felt sooner: steering that does not hold, decisions that cannot be traced back afterward, applications that stall as soon as they cross the boundary of a single department.

What the step to intelligence requires

The transition to intelligence is not a matter of applying more AI. It is a matter of bringing the fundamental dimensions far enough in order that the dependent dimensions can structurally build on them. This differs greatly per organization: one organization mainly needs to consolidate its data management, another has an IT infrastructure that first needs to be brought together before broader steering becomes possible. What it will be depends on where the spread in the plotting round is greatest — that is the place where the next level is waiting.

Anyone wanting to look beyond intelligence will find on the page about the level after intelligence a description of what emerges when steering is no longer merely structural, but self-learning. Anyone seeking recognition earlier in the process can look back at activation to see what the phase looked like before insight emerged, or find out what it means once agents that carry out work begin to play a role alongside the steering layer being built at this level.

The next question

Insight tells you where the organization stands in its capacity to carry AI: which dimensions are solid and which are still loose. That is a different question from which part of the work itself can actually be transferred to AI. Once the load-bearing side is more clearly in view, that second question becomes relevant: which part of the tasks within a process can actually be taken over. That is precisely what the work scan from FTE TO AI was made for — a calculation method that determines per task which part of the work qualifies for takeover, so that the conversation about intelligence is not only about readiness, but also about what, once that readiness is there, can actually be gained.