Most enterprises can now access the same foundation models as their competitors, so the model itself is rarely the differentiator. What separates winners from the pack is strategy: choosing the right problems, having the data and workflows to feed them, and being able to measure whether the effort paid off.

A large share of enterprise AI initiatives stall not because the technology fails, but because they were never tied to a concrete business outcome. Executives who can articulate where AI creates value and where it does not protect their organisations from expensive experiments that never reach production.

An enterprise AI strategy answers four questions before any model is trained. First, value alignment: which measurable business outcome (revenue, cost, risk, cycle time, customer experience) does this serve? Second, use-case selection: is the problem high-value AND feasible given today's data and tooling? Third, data readiness: do we have accessible, governed, sufficiently clean data for this use case? Fourth, build vs buy: do we consume a vendor product, fine-tune or orchestrate an existing model, or build something proprietary?

Build-vs-buy is a spectrum, not a binary. Buying (or using an off-the-shelf API) is fastest and cheapest for commodity capabilities; building is justified only when the capability is a genuine source of competitive advantage or involves proprietary data that cannot leave the organisation. ROI must account for total cost of ownership: not just licences and compute, but integration, change management, monitoring, and ongoing model maintenance.

A mid-size insurer wants to "use AI." A disciplined strategy team scores candidate use cases on value and feasibility. Fully-automated underwriting scores high on value but low on feasibility (data readiness and regulatory risk), so it is deferred. Claims-document summarisation scores high on both: the data exists, the workflow is well understood, and a vendor tool can be bought and integrated in weeks. The team ships the summarisation copilot first, measures cycle-time reduction against a baseline, and reinvests the proven savings into the harder underwriting problem later.

The lesson: sequence use cases so early, feasible wins fund and de-risk the ambitious ones, rather than betting the programme on a single moonshot.

Strategy sets the direction that every other discipline serves. Maturity models describe how far an organisation can execute this strategy today. The operating model decides who owns and runs the initiatives. Governance constrains which use cases are acceptable and how they must be controlled. And the agentic landscape expands the menu of use cases from single-shot predictions to multi-step autonomous work. A strategy that ignores any of these downstream disciplines tends to produce pilots that never scale.