Overview

Vikram Anand joined a consumer goods company as Chief Digital Officer with a mandate to build an AI strategy. Within his first month, he had received twelve separate AI proposals from twelve different business units - a pricing optimiser, a demand forecasting tool, a customer churn predictor, a warehouse automation system, and eight others. Each was reasonable in isolation. "I realised," he told me, "that we did not have an AI strategy. We had twelve shopping lists." Strategy means choosing. Shopping lists just accumulate.

Enterprise AI strategy is the process of deciding which AI opportunities matter most to your organisation, in what order to pursue them, and how to build the capabilities required to execute. It is not a technology plan. It is a business plan that uses AI as the instrument.

Starting with Strategic Context

Before identifying AI opportunities, you must understand the strategic context in which AI will be deployed. This means answering three questions honestly: Where does your organisation compete? Where does it win or lose today? What would need to be true in five years for you to be in a stronger position?

AI strategy should follow from this analysis, not precede it. An organisation competing primarily on price efficiency has a different AI agenda than one competing on product innovation or customer intimacy. A retailer whose primary competitive threat is a faster-moving pure-play e-commerce player has a different AI priority set than one whose threat is private-label penetration from suppliers.

Vikram's company competed on brand strength and retail execution - getting the right products on the right shelves at the right time. That context made the warehouse automation proposal less strategically central than the pricing optimiser and the demand forecasting tool. The competitive battles that mattered most were at the shelf, not in the warehouse.

Identifying and Prioritising AI Opportunities

Every organisation has more AI opportunities than it can pursue. The strategic discipline is prioritisation. A structured prioritisation framework evaluates opportunities across four dimensions.

Strategic impact: how directly does this opportunity advance the organisation's primary competitive objectives? Proposals that address the core competitive challenge get more weight than proposals that optimise peripheral operations.

Value magnitude: how large is the potential financial or operational impact? This should be an honest estimate with explicit assumptions, not a number inflated to win approval. Vikram required all proposals to model three scenarios - conservative, expected, and optimistic - with documented assumptions for each.

Feasibility: does the organisation have, or can it realistically build, the data infrastructure, technical capability, and organisational readiness to execute this? A technically sophisticated opportunity that requires data the organisation does not have is not a near-term priority.

Risk: what could go wrong? Technical risk (the model may not perform as expected), data risk (the necessary data may not be available or clean), adoption risk (the intended users may not use the system), and regulatory risk (the application may attract compliance scrutiny) all affect prioritisation.

Scoring proposals across these four dimensions - even informally, using a simple high/medium/low scale - forces comparison that would otherwise not happen. Vikram's team scored all twelve proposals and found that two clear priorities emerged: the demand forecasting tool (high strategic impact, large value, feasible with existing data) and the pricing optimiser (high strategic impact, large value, moderate feasibility). Eight of the twelve proposals were de-prioritised or deferred without creating significant organisational friction, because the process was transparent and applied consistently.

Developing the AI Vision

A strategy needs a destination - a description of what the organisation will look like when the strategy is successfully executed. For AI strategy, this is the *AI-enabled future state*: what specific capabilities will the organisation have, what will be different about how decisions are made, and what competitive advantages will that enable.

The vision should be concrete enough to be testable. "We will be an AI-driven organisation" is not a vision. "By 2027, our demand planners will use AI-generated forecasts as their primary input, reducing forecast error by 30% and allowing them to redirect their attention from spreadsheet maintenance to supplier collaboration" is a vision. It has a timeframe, a mechanism, a measurable outcome, and a clear description of what changes for the people doing the work.

A concrete vision also serves as a communication tool. When Vikram articulated the demand planning vision to the CFO and the Chief Supply Chain Officer, it generated substantive discussion about what "30% forecast error reduction" would mean for inventory carrying costs and working capital. That discussion was more valuable than any technology briefing - it connected AI to the financial outcomes leadership actually cared about.

Strategic Narratives and Stakeholder Alignment

A strategy that has not been communicated is not a strategy. It is a document. Stakeholder alignment - ensuring that executives, business leaders, technical teams, and front-line employees understand and support the direction - is as important as the analysis that produced it.

Different audiences need different narratives. The board needs to understand the strategic rationale: why AI, why now, what is the competitive risk of not moving, and what investment is required. Business unit leaders need to understand what is changing in their area and what is expected of them. Technical teams need to understand how their work connects to strategic priorities. Front-line employees need to understand what AI means for their role - and have that answered honestly, including the parts that are uncertain.

Vikram learned this the hard way. His initial strategy presentation was designed for the board. When he used the same presentation with warehouse supervisors, they heard "AI will be handling more of your work" and interpreted it as a job threat. The resulting anxiety spread faster than the strategy. He rebuilt his communication in layers - separate, tailored narratives for each audience, developed with input from each group.

Building the Technology Roadmap

Strategy without sequencing is wishful thinking. The technology roadmap translates strategic priorities into a phased plan of capability development - what will be built, in what order, over what timeframe, and with what dependencies.

Good roadmaps are built around the concept of *enabling capabilities*: foundational infrastructure and skills that multiple future AI applications will depend on. Investing in data quality, a centralised data platform, or foundational ML tooling before pursuing specific applications means that each subsequent application is cheaper and faster to deploy. Organisations that skip foundational investment and jump straight to applications typically spend three times as long and two times the budget on each individual initiative, because they build the same infrastructure repeatedly.

Vikram's roadmap had three phases. Phase one (six months): establish a data foundation - clean, centralised, accessible demand data. Phase two (months 6-18): deploy the demand forecasting model and the pricing optimiser. Phase three (months 18-36): expand AI applications across supply chain, using the data foundation and lessons from phase two. Each phase was gated on the previous phase's success criteria, not on a calendar date.

Key Takeaways

• AI strategy follows business strategy, not the other way around. Start by understanding where your organisation competes and wins, then identify which AI opportunities most directly strengthen that position.
  - Prioritisation is the core discipline. Evaluate proposals across strategic impact, value magnitude, feasibility, and risk. Scoring proposals consistently is more important than the precision of any individual score.
  - The AI vision must be concrete and testable. A good vision statement names a timeframe, a mechanism, a measurable outcome, and a clear description of what changes for the people doing the work.
  - Stakeholder alignment requires layered communication. Board members, business leaders, technical teams, and front-line employees each need a different narrative tailored to their concerns. A single presentation will fail someone.
  - Foundational investment before applications saves money and time. Data infrastructure and platform capabilities that enable multiple applications are worth prioritising before any specific AI initiative.
  - Gate roadmap phases on success criteria, not calendar dates. Time-based milestones encourage moving forward before foundations are solid. Capability-based gates ensure each phase is actually ready before the next begins.