.png)
Most large companies already have an enterprise AI strategy, or at least a document with that title. Far fewer can point to AI running in production because of it, changing how a team works and showing up in its results.
This article covers how to build a strategy that gets there: where to start, which use cases to pick, what the roadmap should contain, who owns it, how to govern it and how to tell whether it is working. The examples come from AI Partners engagements with large enterprises.
An enterprise AI strategy decides which business outcomes AI should change, in what order, with which owners, budget and controls, and how progress is measured. The roadmap is its operational form: a dated sequence of use cases over the next two to three years.
An enterprise builds one in five steps: assess readiness, choose a handful of use cases, turn them into a roadmap with named owners, set governance before launch, and measure the result in the P&L. A strategy that lists tools and pilots without saying which workflows change, and who answers for them, is a technology plan. It produces licenses and demos, and very little in the accounts.
Because the roadmap is usually written by people who will not run the use cases, and it counts tools deployed instead of work changed. The result is wide adoption with thin financial impact.
McKinsey's 2026 survey puts numbers on that gap: 44% of respondents say AI is scaling across their enterprise, yet 37% attribute any EBIT impact to it and only 6% qualify as AI high performers. Microsoft hit the same wall in its own program and concluded that access and usage do not equal transformation, since a tool licensed to over 200,000 people does not change how the work gets done.
The other cause is volume. A roadmap with forty use cases has no priorities, and the first quarter goes on debating it instead of building anything.
Check four things: where AI is already used, including tools staff use without approval; whether the data behind your priority workflows is accessible; who owns each workflow; and what your teams can already do. The gaps you find become the first items on the roadmap.
A maturity assessment is only useful if it ends in decisions. In our audits, the diagnosis phase combines field interviews, a review of the tools already in use and a sector benchmark, inside a four-to-six-week engagement, so the findings feed the roadmap directly. Unofficial usage deserves more attention than most leadership teams give it: employees who hide their AI use are showing you where the demand is, and where the risk sits today.
Start with work that happens often, can be checked quickly and already has someone accountable for the outcome. Then cut hard, because a first wave of three use cases beats a portfolio of forty.
Club Med's program shows what cutting looks like. Fifty ambassadors trained over two days identified 102 candidate agents, of which 18 were kept as quick wins ready to deploy and 57 documented as strategic projects. Roughly one idea in six was ready to build. One engagement is not a benchmark, but the ratio is a useful check when a workshop produces a long list.
Who picks matters as much as what gets picked. In a separate one-day workshop, nine Club Med EMEA marketing teams pitched their projects and three were elected by live vote, so the people who would run them chose them. Our guide to identifying high-impact AI use cases covers the scoring criteria.
A short list of use cases sorted into horizons, each with a business owner, a measure of success and a budget line, plus the data or audit work that has to come first. Anything without an owner stays in the backlog.
Vetoquinol's marketing teams built theirs in one working day, after a morning of hands-on training. They finished with 11 use cases scored on four criteria (time saved, error risk, building time and integration) and a 36-month roadmap in three horizons, with a global and local Data & AI audit as the first step of execution. The criteria were fixed before any idea was ranked, so the ranking could not drift toward pet projects. This is the structure we use with clients, built on the same horizons:
Speed in the first 90 days comes from scoping before building. At Trench Group's hackathon, teams built eleven working applications and one concept in three days, and three now run in production, because the use cases had been scoped beforehand.
The CEO and executive committee own the priorities, each use case belongs to the business leader whose team will use it, and the CIO or CDO owns the platform, data and security rules. Without a named business owner, no one defends a use case at budget time.
That split matches where the money already is. BCG's September 2026 research found that more than 80% of AI spending now sits outside enterprise IT, and McKinsey's high performers are twice as likely as others to say their senior leaders demonstrate commitment to AI initiatives. A small central team, often called an AI center of excellence, should set standards and share components without owning the use cases.
Budget follows the same owners. Agent costs scale with usage, and in Futurum's 2H 2026 survey 46.9% of enterprises reported running over budget on AI. Planning spend by function, as in our guide to the 2027 enterprise AI budget, keeps the cost and the owner in one place.
Settle access rights, action logs and human review points for each use case before launch, and scale the controls with the risk. In our experience, governance added after launch is where working pilots get paused.
Few companies have this in place yet. Among organizations running several AI agents on one workflow, Deloitte found that half either have no formal governance or rely only on basic oversight, and just 9% say those workflows are fully governed. BCG reports that 42% of companies expect to grant agents genuine autonomy by 2030, but only 5% have the full set of controls in place today.
European companies also have a regulatory date to plan around. The AI Omnibus, which entered into force on 27 July 2026, moved the high-risk obligations for stand-alone Annex III systems from 2 August 2026 to 2 December 2027, and to 2 August 2028 for AI embedded in regulated products, which puts them inside the structural horizon of most roadmaps. Our guide to agentic AI governance sets out the controls use case by use case.
Count what reached production and what it changed. Three measures are enough: use cases live each quarter, the cycle time of the workflows they changed, and the business results those workflows feed, tracked in the P&L.
The leaders in the research measure the same way. BCG classifies 7.5% of companies as future-built, with a further 41% actively scaling AI, and the leading group measures AI's impact directly in the P&L. In McKinsey's data, nearly three-quarters of high performers report fundamentally redesigning workflows because of their AI use, and they are twice as likely to have defined processes to measure the impact of their initiatives.
Review the roadmap every quarter against those three measures. A use case that misses its target twice comes off the list, and the next candidate from the structural horizon takes its place.
How long does it take to build an enterprise AI strategy?
A first roadmap can be built in a day, as Vetoquinol's was after a morning of training. A full audit with diagnosis, prioritized business cases and governance usually takes four to six weeks, and a first use case in production about a quarter after that.
Who should write the AI strategy, IT or the business?
The business, with IT in the room. Business teams know where the work is wasted, and IT knows what can be built safely on the company's systems.
What is the difference between an AI strategy and an AI roadmap?
The strategy says which outcomes AI should change and why. The roadmap turns that into a dated sequence of use cases, each with an owner, a budget line and a measure.
Should an enterprise AI strategy include AI agents?
Yes, wherever work runs across several steps and systems, and in McKinsey's 2026 survey 40% of respondents from large organizations report scaling AI agents, up from 27% a year earlier. For single drafting or research tasks, assistants are usually enough and cheaper to govern.
Turning a long list of ideas into a roadmap with owners, a budget and a first use case in production is what our AI First Organization engagement does. Book a 30-minute call and we'll map your first three use cases together.
© 2026 AI Partners. All rights reserved. Designed & Developed by DivMonk.
