Begin with strategic choices
An AI strategy should make choices about where the organization wants AI to create value, which work needs stronger control, what should remain human-led, and what it is willing to spend to learn. A list of tools cannot answer those questions. A useful North Star connects business outcomes, employee capability, responsible use, and an explicit tolerance for cost and risk.
Build the baseline from more than one source
The current state is rarely visible in a single report. Evidence may sit across paid-license inventories, identity and usage data, approved-tool records, employee research, support demand, workflow observation, use-case submissions, and risk reviews. Bringing those sources together reveals the difference between access and adoption—and between what leadership thinks is being used and what work actually depends on.
Study the variance, not only the total
Paid licenses can be inactive while free or unmanaged tools carry meaningful work. That variance matters. Compare licensed versus active use, paid versus free use, approved versus observed platforms, and intended versus actual tasks. The goal is not to eliminate every difference; it is to understand whether cost, capability, convenience, awareness, or policy is driving behavior.
Categorize tools and work separately
Platform categories help clarify the portfolio: general productivity, specialist creation, research and analysis, engineering, automation, or capability embedded in an enterprise system. Work needs another lens: data sensitivity, consequence, user role, frequency, reversibility, and required quality. Mapping the two prevents a convenient general-purpose tool from becoming the default answer for every task.
Match control to sensitivity and consequence
Not every use case needs the same boundary. Low-sensitivity drafting may need simple guidance, while work involving confidential data, customer information, financial decisions, external publishing, or system actions requires progressively stronger access, grounding, review, logging, and approval. Governance becomes more usable when people can see why the boundary changes.
Let demand shape the roadmap
A structured use-case backlog turns scattered enthusiasm into portfolio evidence. Group opportunities by problem, beneficiary, reach, value, feasibility, data readiness, risk, and repeatability. An innovation programme can be especially useful here: it creates a governed path for employees to surface friction, test ideas, and produce comparable evidence rather than submitting an unstructured wish list.
Optimize cost against value, not seats
The cheapest portfolio is not automatically the best, and maximum access is not automatically adoption. Cost decisions should consider active use, task fit, time or quality improvement, avoided risk, capability overlap, support effort, and the value of the use-case backlog. This makes it possible to expand, consolidate, redesign, or retire access using evidence instead of enthusiasm.
Treat strategy as a review loop
The first portfolio is a hypothesis. Usage, new use cases, incidents, employee feedback, model changes, and measurable outcomes should continuously update it. The operating model therefore needs owners, controls, training, adoption support, performance measures, and recurring portfolio reviews. Strategy becomes real when evidence can change the next decision.