• Develop a shared understanding of AI capabilities.


• Match AI functionalities to user needs and service gaps.


• Generate realistic AI-enhanced service concepts based on current tech.


• Make informed decisions about AI implementation value and complexity.

• Explored AI functionalities.


• Identified AI opportunities.


• Established a foundation for AI integration.

Not all AI functionalities apply to all domains. Tailor the card set to your context. For manufacturing, emphasize computer vision and predictive maintenance. For content services, focus on NLP and recommendation systems. Don't use too many cards; 8-12 relevant ones are enough. Consider custom cards for domain-specific AI. Teams often want to use AI everywhere. Challenge this: "What problem does this solve that simpler solutions don't?" AI adds complexity, cost, and failure points. It should deliver clear value. The best AI solves real problems; the worst seeks problems to solve. Non-technical participants may propose impossible ideas. That's fine for ideation, but check reality early. "Real-time surgery video analysis" sounds great but needs edge computing, training data, medical certification, and new hardware. "Flagging unusual transactions" is standard. Distinguish between "possible someday" and "practical now." Every AI capability needs data, often labeled data. Without it, you can't build it quickly or cheaply. When evaluating concepts, ask: "Where does the training data come from? How much do we need? How do we label it? Do we have it?" Most AI projects fail on data. Cards should indicate typical data requirements. AI isn't perfect. It's probabilistic. For each concept, ask: "What's the acceptable error rate? What happens when it's wrong?" A 70% accurate recommendation engine is fine; a 70% accurate diagnostic AI is dangerous. Design for failure. Users need to know when they're interacting with AI. "Black box" AI erodes trust. Plan: How do we explain this to users? How do they give feedback? How do they override it? Good AI augments human judgment; bad AI replaces it without recourse. Cards should indicate implementation difficulty. Some AI capabilities are commoditized (sentiment analysis, image recognition, chatbots); others require custom development. Distinguish between "integrate an API" and "build from scratch." Start with the former. AI introduces bias, privacy, and fairness concerns. Ask: Whose data are we using? Could this discriminate? Is this transparent? Are we respecting privacy? What are the downstream consequences? Unethical AI fails.