Let’s kill the phrase “AI strategy.”
It sounds smart. It checks a box. It looks great in a slide deck. But it’s backwards.
Your business needs a strategy. AI is one way to execute it. No model, no matter how well trained, can fix a lack of clarity about what you’re trying to achieve.
If your goal is to reduce claims cycle time, then your strategy is about automation and accuracy. AI might help. If you want to detect fraud faster, then your strategy is risk mitigation. AI might help. But in both cases, it’s just a tool, not the point.
I had this conversation three times last month. Different companies, same setup. Executive calls and says they need help building an AI strategy. I ask what business problem they’re trying to solve. Long pause. Then something vague about “staying competitive” or “leveraging our data assets.”
That’s not a strategy. That’s a fear response.
Here’s what’s actually happening: competitors are announcing AI initiatives. The board is asking questions. Someone read an article about how AI is going to transform insurance or banking or healthcare. And now there’s pressure to “do something with AI” even if nobody can articulate what that something should accomplish.
So teams launch pilots. Lots of them. A chatbot here, a recommendation engine there, maybe a fraud detection model if you’re feeling ambitious. Six months later, you’ve got half a dozen experiments running in parallel, no clear path to production, and a finance team asking uncomfortable questions about ROI.
This is what happens when the tool becomes the strategy.
The organizations getting this right are doing something completely different. They’re starting with use cases. Not “AI use cases.” Business use cases that might benefit from AI.
A property and casualty insurer we worked with last year had a clear problem: claims processing was slow and error-prone. Adjusters were spending hours on routine cases that should have taken minutes. Customer satisfaction scores were suffering. Operational costs were climbing.
That’s a business problem. The strategy was to automate routine claims triage and route complex cases to experienced adjusters faster. AI was one component of that strategy, alongside process redesign and better data integration.
They didn’t build an “AI strategy.” They built a claims transformation strategy that happened to use machine learning for classification. Took four months from concept to production. Cut routine claims processing time by 60%. Improved adjuster productivity by 35%.
That’s what success looks like. The technology serves the outcome.
But here’s where most organizations go wrong: they confuse activity with progress. They spend 18 months building an AI governance framework before they’ve deployed a single model in production. They form an AI Center of Excellence that produces beautiful documentation but no working systems. They hire data scientists and then wonder why business results haven’t improved.
It’s governance theater. And it’s expensive.
Real AI governance emerges from doing the work, not from predicting every possible scenario in a conference room. You need some guardrails upfront, absolutely. But you learn what governance actually requires by putting models into production and watching what breaks.
One financial services client spent a year building their AI governance framework. Roles, responsibilities, approval processes, risk tiers, the works. Very thorough. Then they deployed their first model and discovered the framework didn’t account for model drift. Nobody has defined who monitors deployed models or what triggers a review. The framework was comprehensive but fictional.
They would have been better off starting with one model, defining minimal governance around it, learning from production, and building the framework from real experience instead of imagined scenarios.
The best clients I work with stay tight and focused. They pick one use case with clear success metrics. They define the outcome they’re trying to create. They build the minimum viable governance needed to deploy safely. Then they ship fast.
They don’t waste cycles debating whether to use TensorFlow or PyTorch. They don’t spend three months tuning a model from 87% accuracy to 89% when 87% is already better than the current manual process. They don’t blow up operations with six unmonitored pilots running in parallel.
They stay disciplined about what matters: Does this create value? Can we measure it? Do we know who’s accountable? Can we deploy it safely?
If the answer to all four is yes, they ship. If the answer to any of them is no, they don’t.
That’s strategy. Not the label. The discipline.
Here’s the test: If I asked you to describe your AI strategy without using the words “artificial intelligence” or “machine learning,” could you do it? Can you tell me what business outcomes you’re driving, what constraints you’re removing, what customer problems you’re solving?
If you can, you probably have a real strategy. If you can’t, you have a technology in search of a purpose.
AI is powerful. It’s going to reshape industries. But it’s not a strategy any more than “using databases” is a strategy or “having a website” is a strategy. It’s infrastructure. It’s capability. It’s a means to an end.
Your business defines the end. AI is one way to get there.
So the next time someone asks about your AI strategy, flip the question. Ask what business strategy AI is meant to serve. If that answer is clear, you’re in good shape. If it’s not, you’ve got bigger problems than picking the right model architecture.
Strategy is what you do, not what you label.
