Turn Company Priorities Into an AI Opportunity Portfolio

Most organizations do not suffer from a shortage of AI ideas. They suffer from a list in which every idea is “high priority.” This decision connects to Build an AI Strategy That Starts With Business Value and Rank AI Opportunities by Value, Feasibility, and Risk, which provide the strategic direction and portfolio context. A portfolio forces comparison. It shows where AI may improve current operations, reshape an experience, or create a new source of value—and what the organization must be capable of before it proceeds. Create three portfolio lanes Sort opportunities into: Run better: reduce cost, delay, rework, or capacity constraints. Serve differently: improve how customers or employees access help, decisions, or information. Create new value: enable a new product, service, channel, or business model. Do not require the lanes to be equal. Require each idea to explain which strategic priority it supports. Use a one-page opportunity card For every idea, capture: business problem; affected group; current baseline; proposed AI role; data required; human decision owner; expected benefit; major risk; estimated effort; first test; and stop condition. The NIST AI RMF Playbook is useful when identifying context, affected parties, risk tolerances, and measurement. The GAO AI Accountability Framework provides additional questions about governance, data, performance, and monitoring. Score without pretending uncertainty disappeared Use separate 1–5 scores for: strategic value; measurable business benefit; process and data readiness; employee and customer adoption; implementation effort; risk and reversibility; strength of ownership. Keep the individual scores visible. A single total can hide a fatal weakness. A high-value idea with unapproved data is not “almost ready.” When AI output will influence important decisions, the NIST Generative AI Profile helps teams consider risks specific to generative systems. The OECD AI Principles provide a broader test of accountability, transparency, robustness, and human-centered values. Balance the portfolio A sensible first portfolio usually contains: one low-risk proof point that builds confidence; one capability-building initiative that improves data or workflow foundations; one strategically important experiment with explicit limits; a short “not now” list with reasons and review dates. Do not call a pilot successful because the demo worked. Define business evidence, quality thresholds, user adoption, operating cost, and incidents before launch. OpenAI’s evaluation guidance explains how repeatable evaluations can replace impressionistic testing. Protect the downside Ask where information travels, who can access it, how long it is retained, and how a customer or employee can challenge an outcome. The FTC’s data-protection guidance is a practical starting point. The OWASP Top 10 for LLM Applications helps technical and business teams discuss common application risks in shared language. Run the quarterly decision Review each initiative as keep, fix, scale, or stop. Record the evidence and owner. A strong portfolio is not a trophy case of pilots; it is a changing set of investments disciplined by strategy, capability, and results. Continue with Rank AI Opportunities by Value, Feasibility, and Risk. Use Run a Quarterly AI Portfolio Review: Keep, Fix, Scale, or Stop to revisit the decision with current evidence.