Every AI strategy deck starts the same way: “Let’s identify our use cases.”
Cue the Post-its, the buzzwords, the endless ranking exercises: fraud detection! personalized marketing! supply-chain optimization! You end up with a laundry list that looks impressive and is supposed to guarantee inertia.  Except instead you just end up with a giant backlog that  paralyzes decision-making.

Turns out, more use cases don’t equal more progress. They equal more noise.

Broad reaching use case roadmaps feel rigorous, but most are theater. They create the illusion of progress while locking executives into endless debates: Which use case is bigger? Faster? Cheaper? The boardroom fights drag on while competitors move.

Take a bank I advised. They had 143 documented AI use cases—a monument to analysis paralysis. Ninety percent were dead on arrival: unscalable, unmeasurable, or just shiny distractions. Meanwhile, a smaller fintech ate their lunch by doing the opposite: focusing their energy on just three high-impact AI initiatives and executing with ferocity.

Lesson: Multiplying use cases dilutes energy. Subtracting and prioritizing concentrates it.

The Case for Subtraction Without Suffocation

Let’s be clear: subtraction doesn’t mean strangling bottom-up creativity. Some of the best AI breakthroughs—fraud detection algorithms, demand forecasting models, even ChatGPT itself—started as skunkworks experiments that blossomed unexpectedly.

The goal is not to kill creativity, but to channel it. To create a framework where ideas can emerge, but only the ones that can truly scale and deliver business outcomes survive to maturity.

The Culling Framework: From Chaos to Concentration

Most companies either (1) hoard every idea in a massive backlog, or (2) overcorrect by cutting everything except one or two “pet” projects from the C-suite. Both extremes fail. What’s needed is a thoughtful framework for subtraction—one that prunes without poisoning.

The real discipline of AI strategy isn’t piling up possibilities. It’s subtracting ruthlessly, through a framework that keeps creativity alive but forces hard choices about where to bet deep.

The 6-Lens Framework for Creating Focus

This isn’t about slashing for the sake of it. It’s about channeling finite resources—data talent, compute, executive attention—into the few plays that will compound over time.

1. Strategic Relevance

• Is this use case tethered to a core business priority (revenue, efficiency, risk mitigation) within 18–24 months?

• Or is it a curiosity play with no executive owner?

• Example: For a logistics company, AI-driven route optimization = strategic relevance. AI-generated social copy = sandbox curiosity.

2. Build vs. Buy Discipline

• Buy when it’s a commodity: things like transcription, customer support bots, or generic analytics. Don’t waste cycles reinventing what hyperscalers or SaaS players already do at scale.

• Build when it creates durable differentiation—a moat that competitors can’t easily replicate.

Example: A retailer can buy off-the-shelf AI for workforce scheduling. But building an AI demand forecasting engine tied to their unique sales + inventory data creates proprietary insight that compounds.

3. Moat Potential

• Does this use case add to your data compounding advantage—collecting, generating, or structuring unique data over time?

• Or is it a one-off “wow demo” that doesn’t deepen your strategic position?

Example: Training a one-time custom LLM on last year’s marketing copy is a dead end. Building a recommendation system that continuously improves with every customer interaction strengthens your moat with every transaction.

4. Partner Fit (especially for non–tech-native companies)

• If you don’t have deep AI/ML bench strength in-house, the question isn’t if you should partner—it’s with whom.

• Look for partners who bring:

  • Data Access: Can they unlock external datasets you can’t generate internally?

  • Integration Capability: Do they fit into your workflows, not force you into theirs?

  • IP Clarity: Who owns the derivative models, the outputs, the training data exhaust? (This is where most “partnerships” quietly rob you of your moat.)

Example: A healthcare system that partnered with a genomic analytics startup to co-develop predictive diagnostics. The startup provided cutting-edge models; the health system provided proprietary patient data. Both gained—but only because IP ownership and long-term data rights were explicit from day one.

5. Scalability Path

• Can this move from pilot to production inside of 12 months with real users?

• If not, is the longer horizon justified by regulatory, scientific, or strategic payoff?

• Avoid zombie pilots that linger in “demo mode” for years.

6. Outcome Magnitude

• If this works, will it move a needle the board actually cares about?

• Use the “CFO Test”: would your CFO put this line item in her quarterly update to investors?

How It Plays in Practice

When you run ideas through these six lenses, subtraction stops feeling like suffocation and starts feeling like focus.

At a healthcare company I worked with, teams had generated dozens of use case ideas—from AI for marketing copy to AI for clinical trial recruitment. Rather than cut indiscriminately, leadership ran each idea through this five-lens filter.

• The marketing copy AI? Fun, but low strategic relevance → sandbox.

• The clinical trial recruitment AI? Aligned with enterprise growth, had clear data pathways, and a defined operational owner in the trials division → greenlit.

The result wasn’t fewer ideas. It was a concentrated portfolio where every surviving initiative was tied to impact, owned by a business leader, and resourced to scale.

Remember: AI strategy isn’t about asking “what else could we do with AI?”
It’s about asking:

• Where should we buy, so we don’t burn cycles?

• Where should we build, because it deepens our moat?

• Which partners accelerate us without stealing our data crown jewels?

• Which experiments deserve a sandbox, and which deserve all-in commitment?

Think of it as gardening: if you never prune, the weeds choke out the fruit. If you cut too aggressively, nothing grows. But with the right framework for subtraction, you create the conditions for both focus and surprise breakthroughs.

Because in the AI era, traction doesn’t come from adding more. It comes from having the discipline to subtract wisely.