Sometimes the lessons of today can be found in the past.   

Early in the history of the retail brand Trader Joe’s, founder Joe Coulombe was faced with a fundamental dilemma. Trader Joe’s was trying to establish themselves as a differentiated option to both convenience stores as well as generic big-box retailers. The selection and assortment of items that its stores carried needed to tell a story about what the brand was, while also aligning with the operational needs of the business. As the story goes, Joe devised a principle called ‘The Four Tests’, four simple “sniff checks” that every product in inventory had to pass. These were: each item needed high value per cubic inch, a fast rate of consumption, ease of handling, and a reason to exist that was distinct from competitors. These were pragmatic rules that helped shape Trader Joe’s enduring identity as a high-turnover, high-loyalty retail brand. 

Enterprises face a similar challenge today with AI.

The C-suite demands rapid AI enablement across the workforce and AI-led operations transformation. The vendor landscape offers seemingly limitless options: copilots, AI chatbots, solutions and agents in every form, promising transformation. Use case ideas abound, especially those claiming ‘productivity’ benefits that sound immediately realizable yet often prove elusive execution. Meanwhile, AI’s potential extends far beyond these surface-level wins, but it is far less obvious where the real value lies, or which problems deserve attention.  

Every organization on its AI journey ultimately faces the same questions: Where should we focus our scarce bandwidth, and which problems are truly worth solving using AI? 

The Five Tests 

I first heard of Joe Coulombe’s Four Tests on a recent episode of the excellent ‘Acquired’ podcast, hosted by Ben Gilbert and David Rosenthal. This led me to think: could enterprises adopt a similar discipline for AI? Forrester already offers detailed guidance on granular use case prioritization; but what if we had a simple heuristic (a “sniff test”, so to speak) for executives to cut through the noise and focus on the areas of opportunity that matter most? 

I offer the following five tenets as rule-of-thumb filters for executives deciding where to apply AI: 

  1. Does the opportunity offer high business value? Prioritize AI initiatives that directly advance strategic priorities or solve significant business challenges, i.e. where AI enables tangible outcomes such as cost reduction, productivity gains, new revenue streams, or improved customer experience. These are opportunities that offer a clear conversion of a marginal token, hour, or integration into business impact. 
  2. Can we learn fast from this? Or, does the opportunity offer a high turn velocity? Prefer applications or workflows that offer a high turn velocity where results can be observed quickly and iteratively, allowing the organization to adapt and scale what works. Favor processes with frequent cycles and visible outcomes, so learnings can be rapidly captured and applied across the business. A corollary of this is to ensure by design that every AI ‘solution’, whether created by citizen developers or engineering teams or introduced through a vendor offering must be instrumented with success labels, costs, and failure modes to enable continuous evaluation. 
  3. Do we have the right data for this? Focus on opportunities where high-quality, accessible, and well-governed data is available. Ensure data supports compliance, security, and ethical standards. Avoid initiatives that rely on fragmented, low-quality, or uncontrolled data sources. Citizen builds work well when data lives in sanctioned repositories with clean schemas, while engineered products should leverage curated domains with clear ownership and versioned semantics. As a corollary to this, ensure each use case can be safely managed and governed. Ship only what you can operate inside a well-defined control envelope, where governance, risk management, and accountability can be built into every stage, enabling trust in outcomes and resilience to failure. 
  4. Does it build on, or give us, a defensible edge? Select opportunities where proprietary data and context, differentiated processes, expert knowledge or domain-specific insight can be combined with AI to create defensible differentiation. Avoid generic applications that can be easily replicated by competitors or commoditized over time. The most valuable AI use cases blend general models with your unique data, processes, and expertise. This does not preclude the use of commodity AI for commodity work (eg., automate functions like payroll, by all means, if it’s efficient), but don’t confuse operational efficiency with strategic advantage or market effectiveness. Focus your build effort on the opportunities that make your beer taste better. 
  5. Does it make the next opportunity easier? The mental model for AI is dual: emphasizes creating reusable ‘skills’ that are also long-term cognitive ‘product’ assets for your enterprise. Prefer use cases that create such assets, frameworks, or agentic skills which can be applied beyond the initial deployment, or that create building blocks that transform the marginal cost  or value of the next case. This approach creates a flywheel that raises the organization’s AI maturity and lowers barriers for future innovation (rather than resulting in a wasteland of abandoned use cases that seemed like a good idea at the time).  

These five tenets form a compact decision discipline for enterprise AI. High value density ensures each effort earns its cost of complexity. High turn velocity accelerates learning and drives adoption. Data in hand anchors feasibility while operability at scale safeguards trust and compliance. Proprietary edge secures long-term differentiation. When applied together, these principles focus the enterprise AI portfolio on use cases that compound impact and build a solid foundation for AI-driven transformation at scale.