The Annual Planning Trap
It is late August. Enterprise planning cycles are kicking off. And every business unit is adding "AI" to their strategic priorities because the board expects it, the CEO mentioned it at the all-hands, and nobody wants to look like they are behind.
This is how bad AI investments get made. Not through malice or ignorance, but through the well-intentioned belief that AI should be everywhere. It should not.
The AI Applicability Test
Before putting AI on any initiative for the coming year, run it through a simple test:
- Is the problem actually a prediction, generation, or classification problem? AI excels at these. If the problem is a workflow problem, a coordination problem, or a policy problem, AI is probably not the answer. Better software, better processes, or better management might be.
- Do you have the data? Not "will you have the data eventually" but "do you have it now, in a usable format." If the data does not exist, the first initiative is data collection, not AI deployment.
- Is the current process understood and documented? You cannot automate what you do not understand. If the team cannot describe the current process in detail, they are not ready for AI. They are ready for process mapping.
- Is the expected value large enough to justify the investment? AI projects are expensive when you account for engineering time, integration effort, ongoing maintenance, and change management. A $200,000 AI project that saves $50,000 annually is a bad investment no matter how cool it is.
What Belongs in the Plan
Focus AI investments on a small number of high-impact opportunities. We recommend no more than three AI initiatives per business unit, with staggered timelines and clear kill criteria. For each one, document:
The specific business metric that will improve. The magnitude of expected improvement. The data requirements and current data readiness. The integration requirements. The total cost including ongoing operations. The 90-day checkpoint criteria.
A plan with three well-scoped AI initiatives will outperform a plan with twelve underfunded ones every time.
What Does Not Belong in the Plan
"Explore AI for department X" is not an initiative. It is a vacation. Kill it. "Build an AI center of excellence" is not a deliverable unless it has a client list of internal teams it will serve and metrics those teams care about. "Implement AI copilots across the organization" sounds bold and is usually code for "buy licenses we won't use."
Be ruthless. The companies that will lead in AI by 2027 are not the ones doing the most AI in 2026. They are the ones doing the right AI, well.