You Cannot AI What You Cannot See

AI cannot improve a process it cannot see: if work is spread across different systems or the data sits in different formats, AI will do nothing to enhance it. The barrier is visibility, not capability, which is why value stream mapping has quietly become an AI readiness exercise.

• John Sansbury illustrates this with a pharmacy stock-out: a doctor cannot tell whether the pharmacy has a medicine because the surgery and the pharmacy run separate systems. A monitoring team and an engineering team using different systems face the same problem.
• When mapping a value stream, ask of each step how it is managed, which team does it and which system is used. The joins then light up: where work crosses tools, where data changes format and where a person rekeys information.
• For a university IT function, the question is not which AI tool to buy but which value streams, such as student onboarding, grant processing, access provisioning or repairs, are observable end to end.
• Follow the sequence: map the value stream as it is, analyse it for non-value steps and broken joins, design an observable to-be flow with consistent data, and only then automate. Otherwise you simply automate the confusion.
• A value stream you can see can also be measured, controlled and improved by any means, so this mapping is foundational service excellence work, not throwaway preparation for a technology project.