Artificial intelligence is usually associated with automation, speed, and efficiency. Organisations invest in AI expecting faster reporting, better forecasting, and more efficient operations. While AI certainly delivers on many of those promises, I believe its most valuable contribution is something far less obvious: it reveals relationships between operational events that people simply cannot see, no matter how experienced they are. Most enterprise decisions are still based on formal processes, documented workflows, and KPI dashboards. Yet anyone who has spent time observing frontline operations knows that people rarely follow processes exactly as they are documented – they adapt, build workarounds, make small calls on the ground that never make it into an official report.
I recall the story: I worked with a large retail organisation operating more than 1,800 stores. Every day, the business generated roughly 45,000 operational events covering product availability, inventory movements, replenishment activities, store execution, and audit findings. By the time that information reached senior management, it had been condensed into a handful of dashboards and KPI reports. Those reports gave a good insight into what was happening, yet they couldn’t explain why it kept happening.
The prevailing belief was that recurring out-of-stock situations were primarily a supply chain issue. The obvious explanation was a supply chain problem — delayed deliveries, bad forecasting, inventory shortages. But when AI analysed the operational data across systems at once, a different pattern emerged. The products were usually already in the store. The problem was execution: stock sitting in back rooms, shelves not refilled on time, the same breakdowns showing up again and again in different locations.
On their own, those events looked insignificant. Together, they formed a clear operational pattern. AI connected thousands of seemingly unrelated events into a single cause-and-effect chain and showed that management had been trying to optimise the wrong part of the business.
What impressed me most was not the speed of the analysis. It was the change in decision-making. Instead of asking managers to sift through tens of thousands of operational signals, AI connected seemingly unrelated events into a small number of prioritised root causes, so we could stop spending time searching for problems and start actually solving them.
That experience fundamentally changed the way I think about AI. Its greatest value is not that it automates reporting or produces better dashboards. It reveals how an organisation actually operates rather than how it believes it operates. And once leaders can see those hidden operational patterns, they stop treating symptoms and begin improving the system itself.
However, there is an important misconception that often follows this discussion. Reading this, you could think that AI can easily discover problems on its own. It doesn’t. What it does is reveal recurring patterns that are simply too large, too subtle, or too fragmented for people to notice manually. It helps identify symptoms; operational observation reveals the reason.
Sometimes, something similar can be observed on the product (or software) side as well: there are situations where users consistently follow a completely different navigation path than the one the team had originally designed.
The first instinct is to call it an adoption problem. I had such an instinct once, but further examination made things clearer: in that case, users were solving a different operational problem than the product team had originally assumed. They didn’t use it incorrectly; they were adapting it to fit the reality of their work. Once this gap was clear, the conclusion became more and more evident: instead of forcing users back into the original design, we redesigned the workflow around the behaviour that had already emerged naturally.
AI helps reveal such patterns, but it cannot reliably determine whether those patterns represent poor execution, poor process design, changing operational realities, or simply rational human adaptation. AI can effectively point out inconsistencies between locations (that can look identical in reports), recurring workarounds hidden inside normal activity, subtle shifts in behaviour (developing over time), or weak operational signals that are individually insignificant. But interpretation still requires human judgement and operational expertise.
Quite often I see another common mistake most organisations make: they assume AI will automatically make operations more efficient and sometimes even treat it as the starting point of operational improvement. If the system is well designed, AI implementation will definitely bring positive results. If it isn’t, AI tends to make the problem harder to see, because it now runs faster and looks more authoritative on a dashboard, and poor assumptions may result in quite negative consequences, in the long run specifically. To avoid this, understanding how the organisation actually works must be the first priority – not on paper, but in reality. If the answer is unclear, AI won’t bring any positive improvements.
Before introducing AI into operational environments, I would encourage product, engineering, and operations leaders to ask five questions:
- Do we understand how work is actually performed—or only how it is documented?
- Are we measuring completion or execution?
- Are different teams achieving similar results for the same reasons?
- Where does decision-making happen?
- What assumptions about user behaviour have we never tested?
None of these questions are really about AI. They are about whether the organisation has an accurate understanding of its own operations. AI doesn’t understand the organisation; it understands data. If operational data lacks actual information about performance, AI will optimise only the documented version of reality. It has no way of knowing what happens outside the system unless humans capture it.
Best organisations don’t have the most sophisticated AI models or the largest datasets. They work with data; they remain curious enough to question their own assumptions. Best teams keep checking whether their data still matches what’s happening on the ground. The lack of data is rarely the biggest risk in enterprise operations. Believing that the data tells the whole story is.
