What Is Our AI Strategy?
If it hasn’t come up, it will. Regardless of what you think of the ways in which AI is talked about in the media, or in the hallways (virtual or otherwise) where you work, it is going to be a question.
Let’s approach answering the question by assuming that at some point in time there will be a thing we can point to and say ‘that is our AI Strategy’. The form the thing takes can vary and will likely evolve over time, but it’s going to be some sort of knowledge work artifact that is portable and can be shared with different folks in the organization who need to understand it.
So it is going to be tempting then to suggest we can make a document that lists some things we are doing, or are going to do, and call that The Strategy. But that todo list does not quite cover the meaning, otherwise a grocery list could be called strategic. So what else?
In his book On Grand Strategy, John Lewis Gaddis writes about grand strategy as “the alignment of potentially unlimited aspirations with necessarily limited capabilities”. Strategy is about proportioning what you seek to what you can actually accomplish. It is about aligning means (resources, capabilities) with ends (objectives, ambitions).
Rich Horwath says strategy is the how, not the what—the intelligent allocation of limited resources through a unique system of activities to achieve a goal.
So this artifact, the thing we are going to call a strategy has two vital components:
- Ends (What are you trying to achieve?)
- Means (What resources do you have to achieve that? And, how will you use those resources?)
Now we have the specification of sorts for strategy, let’s turn our attention to what we mean when we say ‘AI’.
In Enterprise in 2025 that means generative AI through large language models and the tools and agents built around them that can ‘read’, ‘write’, and ‘reason’ across text, images and code. It could mean retrieval from our data. Maybe some fine tuning.
Typically we are going to be thinking about five topics:
-
Foundation models (text/code/image/audio/video) that generate and interpret content.
-
Retrieval & grounding: fetching our data so outputs are factual and auditable.
-
Tool use/agents: systems that plan and take bounded actions (search, call APIs, fill forms), with guardrails.
-
Orchestration: plugging models into workflows, identity, and data governance.
-
Safety & cost: controlling privacy, latency, spend, and risk.
If strategy is the art of aligning means with ends, then AI belongs firmly in the category of means. This is an important distinction - AI is a capability, not an end in an of itself (unless of course you ‘make AI’, but even then products are designed to solve a problem, to reach a specified end, rather than just for the kicks).
One way of keeping that distinction in mind is to recognize it as a general-purpose technology. The idea being that like steam, electricity and computers generative AI is an enabler that crosses entire economies.
So does it make sense to ask ‘What is our AI Strategy?’ - well sort of, but as much sense as asking ‘What is our Steam Strategy?’ would have made in the early 19th century. It sometimes makes sense in the context of an organization because of an implicit ‘when it comes to doing X’ that could tagged on the end. For example, a farm equipment company that made horse drawn hay machines might have asked ‘What is our Steam Strategy?’ and it could have come with the implicit ‘when it comes to making hay machines’ understood as given within the context of the company.
But, in building our strategy artifact it helps to make things explicit. And this is where an AI strategy becomes something unique and specific to your company. Start with asking: where do we most reliably turn effort into value? Given that, consider the different things AI can help with today. Pick one area where you know generative AI augmentation or automation could translate to revenue, margin, or risk reduction.
Try and express the area you pick in terms of numbers - e.g. “+X% win rate, –Y% cycle time, –Z% errors in N months”.
With a specific End in mind we can begin to construct this strategy artifact:
-
Ends: “+X% win rate, –Y% cycle time, –Z% errors in N months.”
-
Means: Make choices about how you will use AI to meet that goal sketching out answers to questions such as:
- What models?
- What data access?
- How ‘agentic’ do we need to be?
- What and how do we measure?
- What is my budget?
- What other constraints do I need to respect?
That is basically it. Strategy is simply an act of allocating resources (means) to reach a goal (ends). Start there, and the rest will follow. Start small. Pick one domain where AI can create measurable value, something you can express in a single statement, and then make choices about how you will use available resources (now including AI) to get there.
The impact of generative AI as a general-purpose technology