I talk to a lot of senior leaders about AI. And almost all of them have the same problem: they're drowning in vendor pitches, analyst reports, and breathless LinkedIn posts — but they can't get a straight answer to the question that actually matters.
Where should we actually focus?
This is the AI strategy question. And it doesn't require a PhD in machine learning to answer. It requires clear thinking about your business.
Here's the framework I use.
Start With the P&L, Not the Technology
The biggest mistake executives make is starting with AI and working backward to business impact. "We should use AI for X" is almost never the right starting point.
The right starting point is: where are the biggest drags on your P&L?
Pick your top three:
- High labor costs in a repeatable process
- Slow time-to-market in a competitive window
- Customer churn driven by a quality or service gap
- Revenue leakage from pricing or forecasting errors
Each of these is an AI opportunity. The technology is almost secondary. What matters is whether AI can close the gap faster and cheaper than alternatives.
The Four Categories of AI Opportunity
Once you've identified your business pain points, AI applications tend to fall into four categories:
Automation — Replace repetitive human tasks with AI. Best for: document processing, data entry, customer support triage, routine reporting.
Augmentation — Give humans AI assistance to do their jobs better and faster. Best for: sales outreach, content creation, research, analysis, code review.
Decision Support — Use AI to improve the quality of decisions. Best for: demand forecasting, pricing optimization, risk assessment, customer segmentation.
New Capabilities — AI enables things that weren't possible before. Best for: personalization at scale, real-time customer intelligence, predictive maintenance.
Most companies should start in the first two categories and work their way right as they build AI maturity.
The Build vs. Buy vs. Integrate Decision
This is where most AI strategies go sideways.
Build (train or fine-tune your own models): Only makes sense if you have proprietary data that creates competitive moat, AND you have the ML engineering team to maintain it. Very few companies should do this.
Buy (off-the-shelf AI products): The fastest path to value. Copilots, AI-enhanced SaaS, industry-specific solutions. Most companies should be here.
Integrate (use foundation model APIs): Middle ground. You customize behavior using prompts and your data, but you're not training models. Good for companies with software engineering capability who need more flexibility than off-the-shelf provides.
The honest answer for 80% of businesses: Buy and integrate. Stop trying to build proprietary models unless you have a very compelling reason.
The Three Metrics That Matter
Forget "AI adoption rate." Measure things tied to business outcomes:
- Time saved per process — How many hours per week is AI returning to your team?
- Quality improvement — Is the output better? Fewer errors, higher conversion, lower churn?
- Speed to value — How long from "we should try AI here" to "we have results"?
If your AI initiatives can't show movement on at least two of these within 90 days, you're probably solving the wrong problem — or solving the right problem with the wrong approach.
What to Do This Week
Stop planning and start experimenting. Pick one process in your business with these characteristics:
- Repetitive
- Data-rich
- Low-stakes if it goes wrong
- Measurable
Run a 30-day AI pilot on that process. Measure the before and after. Show your team what's possible. Build from there.
The companies that win with AI aren't the ones with the best strategy decks. They're the ones who learn fastest.
David Cartolano is an AI Business Strategist. Connect on LinkedIn.