Most companies don't need more pilot programs. They need better questions. Here are five conversations senior management should have before approving the next AI initiative.
There's a big difference between "doing something with AI" and building an AI strategy that truly moves the business forward. The former often begins with competitive pressure, flashy demos, or a sense of urgency to avoid falling behind. The latter starts with uncomfortable questions: What problem do we want to solve? What outcome do we want to drive? What data makes it possible? What risks are we willing to take? Who is responsible if this works, fails, or scales too quickly? This shift in focus matters because many organizations continue to fall into the same trap: treating AI as an answer before they've precisely defined their business question.
The first conversation is about value and strategic fitAI shouldn't be treated as a sideshow for the technology department or a trendy project for the board. It should reinforce the existing value proposition or open up a clear new source of growth. If an initiative can't explain which KPIs it will drive in the next 12 to 24 months, it's probably not yet a strategy; it's corporate curiosity. That's why it's essential to require every initiative to answer four questions from the outset: what is the problem, who is the customer, what outcome does the business expect, and how will time-to-value be measured? The best teams don't start with the tool; they start with the outcome.
The second conversation is less flashy, but much more decisive: data and infrastructureMany companies overestimate their AI readiness and underestimate their data problems. When information is fragmented, poorly governed, or trapped in legacy systems, the conversation about advanced models comes too soon. Before discussing agents, copilots, or automation at scale, management should ask where critical data resides, who governs it, how reliable it is, and which use cases would be blocked if data projects were frozen for six months. That question, while it may sound extreme, often exposes the true extent of the data gap. It also forces a very concrete discussion: what should remain on-premises, what should move to the cloud, and what hybrid integration is needed to avoid disrupting the business while modernizing.
The third conversation is about risk, governance and complianceAs AI moves beyond internal processes and begins to impact sensitive decisions, customers, and regulated processes, the question is no longer just “Does it work?” but “Under what conditions do we allow it to operate?” The European Union, for example, classifies certain uses as high-risk, including tools for employment or access to essential services like credit. This approach offers a useful signal even for companies outside Europe: not all use cases deserve the same level of freedom. An internal assistant for summarizing meetings doesn't need the same level of control as a model that influences pricing, credit approval, or recruitment. Therefore, each relevant use case should have its own control sheet: purpose, data sources, known limitations, monitoring metrics, escalation points, and moments of human intervention.
The fourth conversation has to do with operating model, talent and changeAI doesn't fail solely due to a flawed model; it also fails when no one knows who decides, who approves, who trains, who monitors, and who absorbs the operational change. Many companies continue to manage AI as if it were a traditional IT project, when in reality it intersects strategy, operations, legal, security, data, and business. This demands visible leadership, clear incentives, and practical literacy that goes beyond simply hiring a couple of experts. Furthermore, the organization needs to redefine roles: what tasks are delegated to systems, what decisions remain human, and how performance will be measured in an AI-augmented work environment. Without this redesign, adoption becomes ambiguous, and internal trust quickly erodes.
The fifth conversation is about portfolio, measurement and sequencingOne of the most frequent mistakes is lumping together quick wins, transformational bets, and exploratory experiments. When everything competes with everything else, the most eye-catching wins, not the most valuable. Management needs a shared record of initiatives with the business owner, quantified benefits, dependencies, risk level, and maturity stage. It also needs the discipline to kill, pivot, or scale projects based on evidence, not enthusiasm. AI isn't consolidated by accumulating pilot projects; it's consolidated by better prioritization, quarterly reviews, and allocating sufficient investment to the foundational elements: data, platforms, governance, and training.
The conclusion isn't technological. It's managerial. The companies that will capture the most value won't be those that buy the most tools or announce the most experiments. They'll be the ones that learn to ask better questions before committing budget, reputation, and executive time. AI doesn't need more fireworks in the committee room. It needs more sound judgment at the table.
Does your company already have AI initiatives underway, but still lack a clear framework for prioritizing them? In the next newsletter, I'll share an executive summary with the 5 questions no committee should skip.
