How to cultivate AI use cases and avoid magical thinking in enterprise

Executive overview

Most organizations are rushing AI investments without the discipline that uncertainty demands. The result is misaligned use cases, inflated forecasts, and adoption that never materialises. The fix is borrowed from product management: treat AI like early-stage innovation, not execution.

The real risk is not moving too slow — it is making large bets before you know what works.

Magical thinking and the Dunning-Kruger trap

• Magical thinking appears when people leap ahead of what AI can actually do, skip steps, or inflate expected outcomes
• Saying "the AI" as if it is a department you hand work to is a warning sign
• Nobody has more than two or three years of daily ChatGPT experience — everyone is still a beginner
• Dunning-Kruger effect applies: early users overestimate mastery, just as new drivers rate themselves above average
• The antidote is anchoring on the problem: what does "done" or "progress" actually look like?
• Unrealistic organisational timelines compound misaligned expectations into promises that cannot be delivered

How to prioritise use cases

• Align top-down (C-suite strategy) with bottom-up (ground truth from people closest to the work) — neither alone is sufficient
• Assess data maturity: can you actually make data-driven decisions, or does the final call default to gut instinct?
• Do a directional viability check early — order-of-magnitude impact only; spend no more than one hour on a spreadsheet at stage one
• Feasibility matters: great ideas tied to data that will not be ready for three years are not today's use cases
• Over-indexing on financial projection too early produces inflated forecasts and, often, inflated costs to match

Cultivating use cases: the research-and-workshop method

• Research the problem space first: interview willing stakeholders, map the contours, understand depth and scope
• Study competitors and pitfalls before forming a perspective — arrive with a point of view, not conclusions
• Design a workshop that uses well-formed questions to let stakeholders shape the direction while you steer toward a researched destination
• Cross-functional participation builds shared ownership; people who helped shape a direction are far more likely to carry it forward
• Use online collaboration tools (e.g. ideaboardz) to replicate sticky-note facilitation when co-location is not possible
• Identify who could block the initiative and bring them in early — ask them directly what they want this to become

Managing stakeholder inclusion

• Large audiences (80+ people) belong in a readout, not a workshop
• Frame exclusion honestly: "We are doing exploratory work and do not yet know if there is a there there — we do not want to waste your time"
• Offer to share meeting notes as an opt-in alternative to attendance
• In innovation work, reading every executive in on every use case wastes time; roughly 80% of candidates will be abandoned — save senior attention for the 20% that survive

Smaller bets over large transformation programmes

• Prefer a portfolio of smaller, cheaper initiatives over a handful of million-dollar bets — more shots on goal, more learning
• Failed small bets still yield reusable components and build organisational capability
• Large investments in RAG pipelines and LLM-on-vector-database solutions are proliferating before organisations have validated the underlying need
• Cutting workforce simultaneously with AI investment assumes a clean human-to-machine swap — a risky assumption that can destroy institutional knowledge

The evolving role of the corporate AI leader

• Facilitation of human alignment is becoming the core skill — this is not new, but it is newly critical in AI adoption
• The unique human contribution is irreplaceable; augmented intelligence means the partnership produces more than either could alone
• Knowledge architecture — curating data and applying AI to structured knowledge bases — will define the next decade of relevant solutions
• Product management thinking is particularly well positioned to lead this era