When a leadership team asks whether they are ‘ready for AI’, they often mean whether they have chosen a vendor or hired a data scientist. Those are downstream questions. Readiness is whether the organisation can describe a process, measure it, and change it without a six-month programme.
AI amplifies whatever operating system you already have. If handoffs are informal, data is contradictory, and exceptions live in inboxes, a model will produce confident answers that nobody can act on. If you have owners, service levels, and a system of record, even a modest assistant can remove hours of coordination.
We start assessments with four questions. Where does work wait? Which decisions are repeated? Which documents are re-typed? Which metrics leadership does not trust? The answers tell you more about AI potential than a list of use-case workshops.
A practical roadmap then sequences three kinds of work: making the process explicit, making the data usable, and only then introducing models where judgement or unstructured input is the bottleneck. That order is unfashionable. It is also how programmes survive contact with operations.
If you are evaluating AI this year, resist the urge to start with a chatbot. Start with one painful workflow, the data it needs, and the exception path humans will still own. That is readiness you can fund.
