News
AI strategy for business leaders: from vision to execution
A working AI strategy aligns artificial intelligence to business value, prioritises a short list of use cases by impact and feasibility, builds the data and governance foundations they need, upskills the people who will run them, and measures the return. Most organisations now use AI somewhere, but far fewer get paid for it. That gap between activity and value is set at the top, by the leaders who sponsor the work, sequence it and hold it to a metric. Closing it is the problem this guide sets out to solve.
The numbers make the gap hard to ignore. Enterprise AI adoption reached roughly 78% by the end of 2025, up from 72% in early 2024 [McKinsey, 2025]. Yet 60% of companies report no material financial value from their AI investments, and only 5% have scaled it to the point of competitive advantage [BCG, The Widening AI Value Gap, 2025]. Buying tools is no longer the differentiator; what separates companies now is knowing which problems to point them at and rebuilding the work around them.
An AI strategy for business is the plan that connects a company's objectives to a prioritised set of AI initiatives, the data and operating-model changes they require, the governance that keeps them safe and compliant, and the metrics that prove they paid off. It is a decision-making discipline rather than a technology shopping list.
This question sits at the centre of how SKEMA Business School works with leaders and organisations, through its executive education programmes in AI, transformation and leadership and its dedicated research centre. What follows sets out the framework, the foundations, and the roadmap a leader needs to move from vision to execution.
What is an AI strategy for business, and why it has become an imperative
An AI strategy answers four questions in order: where will AI create value, what will it take to capture that value, how do we keep it responsible and compliant, and how will we know it worked. Without a clear answer to the first, the other three only help an organisation automate the wrong work.
Two shifts have turned this from a competitive option into a baseline requirement. The first is agentic AI: systems that do not just answer a prompt but carry out multi-step tasks, call tools and act with a degree of autonomy. That raises the upside, the governance stakes, and the question of who in the leadership team owns the result. The second is the widening performance spread. Only 6% of firms qualify as AI high performers, meaning AI contributes more than 5% of their earnings before interest and tax [McKinsey, 2025], and high performers report a return of 10.3 dollars for every dollar invested against an average of 3.7 dollars [McKinsey, Google Cloud, 2025]. The leaders are pulling away while the majority stall in pilots.
The stall is measurable. Gartner estimated that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, weak risk controls and unclear business value [Gartner, 2024]. A strategy exists to keep an organisation out of that 30%, and keeping it out is an executive responsibility before it is a technical one.
The AI strategy framework: from vision to execution
The framework below moves in sequence, from the boardroom question of value down to the operational discipline of measurement. Each stage feeds the next.
Set the vision and frame business value
Start from the business, not the model. The vision states, in plain terms, what AI is meant to change: a cost line, a revenue stream, a decision that is currently too slow or too inconsistent. McKinsey's 70-20-10 observation is useful here: about 70% of the value of an AI programme comes from people and process change, 20% from data, and only 10% from the algorithms themselves [McKinsey, 2025]. A vision that talks only about models is aiming at the smallest tenth of the prize. Framing value against strategic objectives, and turning it into an organisation-wide programme instead of a set of disconnected initiatives, is the work SKEMA's customised executive programmes for companies are built around.
Prioritise use cases by impact and feasibility
A long backlog of AI ideas is not yet a strategy. Score each candidate on two axes: business impact (revenue, cost, risk, decision quality) and feasibility (data availability, technical readiness, organisational appetite). Fund the high-impact, high-feasibility quadrant first and treat the rest as a watch-list. BCG found that roughly 70% of AI value concentrates in core functions such as sales, marketing, supply chain and pricing rather than in scattered experiments across the org chart, so a few concentrated bets tend to beat a wide spread of them [BCG, 2025]. Structured prioritisation and the build-versus-buy call are decision disciplines in their own right, and they are what protects the budget.
Build the data, technology and operating-model foundations
This is where most pilots quietly die. AI runs on accessible, governed, reasonably clean data and on an operating model that lets a model's output actually change how work gets done. The detail that separates winners is sequencing: BCG reports that high performers are about three times more likely to have redesigned their workflows before choosing a model, rather than bolting a model onto an unchanged process [BCG, 2025]. The foundation is organisational as much as technical.
An AI operating model is the set of roles, decision rights, processes and platforms that lets an organisation build, deploy and run AI repeatedly rather than as one-off projects. It covers who owns a use case, who signs off on risk, how a model moves to production, and how it is monitored once live.
Govern for responsible and compliant AI
Governance is not a brake bolted on at the end. It runs in parallel with delivery, deciding which use cases are acceptable, what data they may touch, and how their outputs are checked. Treated early, it speeds deployment by removing the late-stage surprises that kill projects. This is detailed in its own section below.
Develop talent, leadership and change management
The 70% of value tied to people and process has to be funded as such. McKinsey reports that high performers direct around 70% of their transformation budget to people, process redesign and change management [McKinsey, 2025], not to software licences. A strategy that under-invests here is funding tools the organisation has trained no one to use. The leadership and culture work this depends on is the subject of its own section below.
Measure ROI and iterate
A use case with no metric attached has no way to prove it worked. Define the indicator before launch, baseline it, and review on a fixed cadence so that funding follows evidence. The measurement section below sets out which KPIs hold up.
An AI maturity model for business
Knowing where you stand keeps the strategy honest. Five stages describe the path from curiosity to advantage, and most organisations can place themselves on it in a few minutes.
-
Scattered pilots, individual enthusiasm, no shared data foundation or governance. Useful for learning, not for value.
-
A first prioritised use-case portfolio, an owner for AI, and basic data access. Value starts to appear in one or two functions.
-
Reusable data and platform foundations, governance in place, several use cases live in core functions. AI begins to show in the numbers.
-
AI embedded in standard workflows, monitored in production, measured against business KPIs. This is roughly where the 6% of high performers sit [McKinsey, 2025].
-
AI reshapes the operating model and the offer itself, with continuous iteration. This is the 5% BCG calls future-built [BCG, 2025].
How to prioritise AI use cases
Prioritisation is the part of the strategy that protects the budget. Score candidates against a small set of criteria and let the scores, not the loudest stakeholder, decide.
Business impact
The size of the cost, revenue, risk or decision-quality effect, expressed in numbers wherever possible.
Feasibility
Is the data available and usable, is the technology mature, will the affected team adopt it.
Time to value
How long until the use case produces a measurable result.
Risk and reversibility
What happens if the model is wrong, and how easily the decision can be caught and corrected.
On build versus buy, the test is differentiation. Buy or adopt a vendor tool for capabilities that are common across your industry, such as a coding assistant or a customer-service agent. Build, or heavily customise, only where the use case touches proprietary data or a process that is itself a source of competitive advantage.
Governance, risk and responsible AI
Responsible AI has moved from a reputational nicety to an operational requirement. The number of documented AI incidents rose to 362 in 2025, up from 233 the year before [AI Incident Database, Stanford HAI, 2026], and the regulatory frameworks are now concrete: among organisations with formal AI governance, 60% map their controls to GDPR, 36% to ISO/IEC 42001 and 33% to the NIST AI Risk Management Framework [Stanford HAI, 2025].
Responsible AI is the practice of designing, deploying and monitoring AI systems so that they are accurate, fair, secure, explainable and compliant with the law, with clear human accountability for their decisions. It covers data quality and provenance, bias testing, security, and a documented decision on who is answerable when a model is wrong.
A workable governance layer assigns an owner to every live use case, defines which data each may use, sets a risk tier that determines the level of human review, and monitors models in production for drift. The hard questions it settles are as much organisational and ethical as technical, which is the ground SKEMA's Centre for Artificial Intelligence works on. Directed by Professor Margherita Pagani, the Centre pursues what it calls AI with purpose: human-centric AI examined across four levels of value, algorithmic, business, societal and ethical, by an interdisciplinary faculty of social scientists, AI researchers and engineers. Its work is anchored in international scientific collaboration, with academic partners that include UC Berkeley, City, University of London and LUISS in Rome, and a scientific committee drawing on senior scholars in the field such as Thomas Davenport. That research feeds straight into the questions a governance layer has to settle, including explainability, bias and who is accountable when a model is wrong. SKEMA faculty set out the underlying principles in their work on the ethics of AI and human-machine cooperation. Governance handled this way clears use cases faster, because the hard questions are answered before, not after, a model reaches production.
The leadership challenge behind AI transformation
The 70-20-10 split has a corollary. If most of the value of AI sits in people and process, then most of the work of an AI strategy is more a leadership job than a technical one, and it falls to the people at the top. This is the part that separates the organisations that scale AI from the ones that stall.
It starts with a vision the rest of the organisation can act on. A leader's job is to state, in business terms, what AI is meant to change and why it matters, then to hold that line as the programme meets resistance. Without it, AI fragments into disconnected pilots that each optimise a corner and move no number the board recognises.
Sponsorship turns the vision into budget and cover. High performers direct around 70% of their transformation budget to people, process redesign and change management rather than to software licences [McKinsey, 2025], and that allocation is an executive decision before it is anything else. A sponsor also clears the path: they settle the cross-functional disputes, fund the unglamorous data and governance foundations, and protect the programme through the trough between the first pilot and the first measurable result.
Cultural transformation is the harder half. AI changes how decisions are made and who makes them, and an organisation that treats it as an IT rollout will get IT-rollout results. Leaders set the tone for whether the workforce experiences AI as a threat or as leverage. BCG reports that 64% of executives favour a side-by-side model in which AI augments people rather than replaces them, and only 7% anticipate a net reduction in headcount from AI [BCG, 2025]. That framing matters for adoption: a team that expects to be replaced will not surface the friction a rollout needs to hear.
Human-AI collaboration is where the framing becomes operational. Agentic systems now carry out multi-step tasks with a degree of autonomy. That turns the human-machine boundary into a management question: who decides, who checks, and who is accountable when a model acts on its own. SKEMA puts that question at the centre of its applied work, including the AI Innovation Centre on its Montreal campus, which develops applied AI tools and connects managers to AI work in one of the world's leading AI hubs.
Talent development funds the rest. The scarce resource is rarely the model; it is people who can frame a problem for AI, judge its output and redesign the work around it. Demand for these capabilities is climbing fast, with the agentic AI skill cluster growing by about 280% in a single year [Lightcast, Stanford HAI, 2026], and supply has not caught up. Building this judgement mid-career is the focus of SKEMA Executive Education, from its Global Executive MBA, positioned around leading with purpose in the age of AI and ranked 5th worldwide by the Financial Times in 2025, to its executive certificates in leadership, management and innovation for managers taking on transformation in mid-career. These programmes are built to develop that judgement, which decides more about an AI transformation than the tools do.
Measuring success: ROI and KPIs
The KPIs that survive scrutiny tie back to a business outcome, not to model activity. Track four families and baseline each before launch.
- Productivity: time saved or output gained per task. Controlled trials of coding assistants, for example, have measured productivity gains around 19.3% [Stanford HAI, 2026].
- Cost and revenue: the cost removed or revenue added by a live use case. In customer service, autonomous agents now handle up to 80% of interactions in some deployments, with a median payback of about 5.1 months [Gartner, Stanford HAI].
- Decision quality and speed: error rates, cycle times and the consistency of decisions a model supports.
- Adoption: the share of the target population actually using the tool, the leading indicator that predicts the rest.
Vanity metrics, such as the number of pilots launched or prompts run, tell you nothing about value. Report the four families above on a fixed cadence and move funding towards what works.
An AI strategy roadmap: 30, 60, 90 days and 12 months
A strategy earns trust by producing an early, visible result. The sequence below front-loads the foundations and a first win.
- Days 1 to 30, frame and prioritise: Agree the value thesis with the leadership team, run a use-case scoring workshop, and pick two or three initiatives in core functions. Set the metric for each before any build starts.
- Days 31 to 60, prepare the ground: Stand up the data access and basic governance the chosen use cases need. Assign an owner and a risk tier to each. Brief the affected teams early, while the plan can still absorb their feedback.
- Days 61 to 90, ship the first use case: Put one prioritised use case into production, measured against its baseline. Capture what the workflow redesign actually required, because that is the reusable lesson.
- Months 4 to 12, scale on evidence: Roll out the use cases that proved their metric, retire the ones that did not, and reinvest the saved capacity. Formalise the operating model so the next use case is faster than the last.
Common mistakes that undermine an AI strategy
The failure patterns are consistent across industries, which makes them avoidable.
- Technology in search of a problem: Buying a model and then hunting for somewhere to use it. The fix is to start from a prioritised business problem.
- No executive owner: AI treated as an IT programme, with no leader who owns the value thesis. The work then fragments into pilots no one is accountable for, the pattern that strands most organisations in the maturity model.
- Bolting AI onto an unchanged process: The reason high performers redesign workflows before choosing a model [BCG, 2025]. A model dropped into an old process inherits the old process's limits.
- Weak or ungoverned data: The most cited cause of abandoned projects [Gartner, 2024]. Foundations are unglamorous and decisive.
- Governance treated as an afterthought: Late-stage risk reviews kill projects that early governance would have shaped to pass.
- Funding the algorithm, starving the people: Ignoring the 70% of value that sits in people and process change [McKinsey, 2025].
- No metric, no review: A use case that no one measures cannot be defended when budgets tighten.
FAQ
-
It is the plan that connects a company's objectives to a prioritised set of AI use cases, the data and operating-model changes they require, the governance that keeps them compliant, and the metrics that prove the return. It is a way of deciding where AI is worth the effort, not a list of tools.It is the plan that connects a company's objectives to a prioritised set of AI use cases, the data and operating-model changes they require, the governance that keeps them compliant, and the metrics that prove the return. It is a way of deciding where AI is worth the effort, not a list of tools.
-
Set a value-led vision, prioritise use cases by impact and feasibility, build the data and operating-model foundations, govern for responsible use, develop the talent to run it, and measure ROI against a baseline. Ship one use case early to prove the approach before scaling.
-
Leadership decides whether the strategy works. Executives own the value thesis, sponsor the budget (high performers put around 70% of it into people and process change), set the cultural tone that drives adoption, and design the human-AI boundary that governance relies on. The technology is rarely the constraint; executive ownership usually is.
-
A repeatable sequence that moves from business vision to measurable execution: vision and value, use-case prioritisation, data and operating-model foundations, governance, talent, and measurement. Each stage feeds the next, so a weak foundation caps everything above it.
-
Because organisations invest in models while neglecting data, governance and process change. BCG finds 60% of companies see no material value from AI [BCG, 2025], and Gartner expected at least 30% of generative AI projects to be dropped after proof of concept [Gartner, 2024], largely for those reasons.
-
Baseline a business metric before launch, then track productivity, cost and revenue, decision quality and adoption. High performers report about 10.3 dollars returned per dollar invested, against a 3.7 dollar average [McKinsey, Google Cloud, 2025], which is the spread a disciplined strategy is trying to close.
From vision to execution starts now
The companies pulling ahead are not the ones with the most AI tools, but the ones that picked a few problems worth solving, rebuilt the work around the answer, governed it properly, and measured the result. With 60% of organisations still extracting no material value from their AI spend [BCG, 2025], the advantage is going to the minority that treat AI as a strategic discipline rather than a procurement exercise.
The first execution step is usually a single, well-chosen use case with a metric attached. For organisations setting out to turn AI ambition into measurable value, and for the leaders preparing to drive the transformation themselves, the strategic, technological and leadership capabilities this demands are no longer optional. SKEMA Executive Education supports that work through executive programmes focused on AI, transformation and leadership that turn the frameworks above into practice.
