Start with the Vision, Not the Tool: A Practical Approach to AI Strategy

We are in a once-in-a-generation moment. Cloud gave us elastic compute, mobile put technology in everyone's pocket, and now generative AI is changing what software can do. Every leader (technology or business) I talk to feels the pressure to "do something with AI." That pressure is exactly why strategy matters. Without one, AI becomes a collection of pilots, tool purchases, and demos that never add up to anything.

Why strategy matters

A good strategy does a handful of things for you:

• - It minimizes waste and maximizes outcomes. Without a strategy, teams chase the same problem with different tools, and money disappears into duplicate efforts.
• - It creates cohesion. Everyone is pulling in the same direction, so the work compounds instead of colliding.
• - It isn't dependent on any one tool. Tools will change, probably faster than any technology cycle we've seen. Decoupling strategy from tools is critical for a long term success.
• - It can scale. Strategy is beyond Clever proof-of-concepts and Experiments that can't grow beyond one team or one use case.
• - It forces prioritization and tradeoffs. This is the point where most organizations struggle, and where a strategy really adds value.

Here's how I think about building an AI strategy that actually works.

Start with the vision.

Here's something that often gets skipped: strategies are built to support the execution of a vision. They don't replace the need for a business vision. For an enterprise, that creates a real problem, because many companies can't just write a formal AI vision. Many are still working out what AI means for them, and that's fine.

I'd suggest two practical paths

- See if AI can be incorporated into your main company vision. If your vision is already about serving customers better, moving faster, or operating more efficiently, AI may simply be one of the ways you get there.

- Create an AI charter. A charter doesn't have to be a grand statement. It defines why you're pursuing AI, what you're trying to achieve, what principles will guide you, and where the boundaries are.

If you're not sure where to begin, begin here: start by defining your AI charter. It gives every later decision something to be measured against.

Develop common language about AI

This is an area leaders should discuss extensively. "AI" is doing a lot of work in most conversations. Are we talking about a tool? A model? A product feature? A way of working? AI strategy is really about capabilities, not tools. It helps to separate three kinds:

- Business capabilities: what the organization needs to be able to do, such as underwriting a risk, resolving a customer issue, or forecasting demand.

- Technical capabilities: what the technology needs to be able to do to support those business capabilities.

- Governance capabilities: how the organization governs AI investments and their use, from business-case and ROI discipline to ethical, safe and compliant operation.

Tools sit beneath all of this. They're how you deliver a technical capability, and AI is one of the tools in that toolbox, a powerful one, but still a tool.

The three types of capability are in constant conversation. Business capabilities define what technical capabilities you need, and technical capabilities can open up business capabilities you hadn't considered. Governance capabilities provide the necessary guardrails to ensure safety and compliance. That said, the question I'd put at the center of every AI conversation is simple: which business capability are we trying to enhance? If you can't answer that, you're not ready to take on AI transformation.

Apply systems thinking to build the strategy

Once you've identified a business capability, resist the urge to automate a single step in a process. Look at the whole business process flow instead. Improving one step in isolation can feel like progress while changing nothing about the outcome. Speeding up a task in the middle of a process doesn't help if the bottleneck is three steps downstream. So define the value of the chain, not just the steps, and use that to decide where to apply AI first. In my opinion, ignoring the big picture in a rush for “quick wins” is why organizations fail to reap the real benefits of AI: the value from automating a task with AI gets lost in the dependencies of the overall workflow. McKinsey's latest survey (reported by The Register) found only 6% of respondents qualify as AI high performers, and what sets them apart is organizational change, not tooling. This is one of the two most critical factors for the strategy (change management, covered below, is the other). This needs business alignment, operations alignment, financial investment, and time commitments.

Choose technical capabilities on merit

With the business capabilities and value chain defined, you can get specific about the technical side.

Knowledge management

AI makes it far easier to find, summarize, and use what an organization already knows. Generative AI in particular, also offers a lot of data manipulation skills and is more tolerant of messy, unstructured data than conventional tools, but it is still highly dependent on access to quality, accurate and complete data. The business capabilities needed should truly decide the extent of investment in a data layer—which can be as cheap as object storage in data lakes or as complicated as RAG (retrieval-augmented generation) applications that rely on intricate ontology of the data.

Agentic automation

This is the area getting the most attention, and for good reason. Agents can take on multi-step work, including in software development. This is where having a solid foundation of infrastructure, common harness patterns (the reusable scaffolding around a model that manages its tools and context), and effective use of MCP (Model Context Protocol) to read and write data reliably through APIs becomes critical.

Two design questions worth asking:

• Is this an agent or an application? When does a workflow need the flexibility of an agent, and when is a conventional application the better choice?
• Where should humans stay in the loop?

This is also where governance capabilities earn their place. Agents take actions, not just give answers, so the organization needs clear guardrails: security and risk controls, human oversight on the decisions that matter, and a named owner accountable for the outcome.

Then layer the deployment complexities.

• How will we deploy this, and how will we scale it? A solution that works for ten users and a solution that works for ten thousand are often very different things.
• What deployment and scaling techniques will we rely on?
• What will it really cost? Cost calculations for AI are tricky. Usage-based pricing, infrastructure, integration, people, and ongoing maintenance all add up.
• Have we modeled what this looks like at scale, not just in the pilot?
• Can we sustain it? A cost that looks small today can look very different once adoption grows.

Notice that these are questions, not answers. I'd rather leaders get comfortable asking them early than discover the answers after the budget is committed.

Lastly, do not underestimate the need for change management

Along with systems thinking, AI strategy leaders should spend the most time here. In my opinion, ignoring this at the strategy phase, for a later phase like (roadmap or implementation phase) is a perilous proposition. The following should all be thought of deeply:

• Communication
• Training
• Identifying champions, pilot users etc.
• Incentive structures
• Success metrics

AI strategies are not simply capabilities; they are the genesis of culture change in an organization. The AI strategies must account for building innovation, speed of execution and a fail-fast mindset, all within the guardrails of governance, and should focus on the long game.

Closing thoughts

The organizations that get the most from AI won't necessarily be the ones with the biggest budgets or the newest tools. They'll be the ones who started with clarity: a vision, a charter, and a clear view of the business capabilities they want to strengthen, backed by the governance and change management to make it stick. AI tools will keep changing. A good strategy keeps things grounded amid all the change.