AI is not observing your organisation
AI analyses representations of organisational work, so evidence, interpretation, hypotheses and recommendations must remain distinct.
Give an AI system your procedures, spreadsheets, meeting notes and process maps and it can do something genuinely useful with them. It can compare accounts, identify inconsistencies, structure information and generate questions. But we need to be careful about what has actually happened. The AI has not observed your organisation. It has examined representations of it. That distinction becomes important very quickly. Imagine a process containing three approval stages. AI might reasonably suggest that there is duplication and ask whether one or more approvals could be removed. Perhaps it is right.
But perhaps the first approval exists because of regulation. The second was introduced after a fraud. The third compensates for poor information earlier in the process. Or perhaps nobody remembers why any of them exist. The process diagram cannot necessarily tell us. This is why I find it useful to separate four things: Evidence. Interpretation. Hypothesis. Recommendation. A system record showing that an activity occurred is evidence. Our explanation of why it occurred is an interpretation. A possible underlying cause is a hypothesis. What we think should change is a recommendation.
Generative AI can help with all four, but they are not the same thing. The danger comes when a plausible hypothesis quietly becomes treated as evidence. Language models are particularly good at producing coherent explanations. Unfortunately, organisational life does not have to conform to the most coherent explanation available. People have histories, incentives, informal relationships and knowledge that may never appear in the documentation. So use AI to interrogate organisational information. Use it to challenge assumptions. Ask it to identify contradictions. Ask what evidence would distinguish between competing explanations. But when the evidence runs out, do not simply ask AI to complete the story. Go and find out. The objective is not to produce the most sophisticated explanation of the organisation. It is to develop enough understanding to make a better decision about what to do next.