While AI excitement continues to revolve around bigger models and better data, many enterprises get caught up in the complexity of bigger, faster, better. Precision, the quiet backbone of an effective strategy, often remains overlooked as a consequence. Yet without it, even the most powerful AI can accelerate the wrong outcomes.
Cassie Kozyrkov, first and former Chief Decision Scientist at Google and now founder and CEO of Kozyr, a company dedicated to advancing decision intelligence, joined The AI Forecast to explore how this often-discounted discipline can transform AI from a technical novelty into a strategic asset by simply remembering to ask ‘why’.
Here are some key takeaways from that conversation.
Choosing the right model isn’t enough. Ask better questions first.
Paul: There’s growing concern about the dark side of AI hype, particularly its impact on thinking and decision-making. Are we genuinely improving how we think, or just outsourcing it?
Cassie: The biggest misconception is that AI’s value lies in prediction. It doesn’t. The true power is in choosing better actions, but only if we start with the right question. Enterprises often overprioritize modeling and predictions while neglecting the decision-making process that gives those models value. The model itself, no matter how advanced, is secondary to the clarity of the problem it’s solving.
Decision intelligence is about starting with the right question, not just optimizing the algorithm. We’ve invested heavily in tools that generate answers, not frameworks that help us ask the right questions.
I caution against the cognitive cost of overreliance. When shortcuts replace engagement, organizations lose their capacity for deep thinking. Think of, “Should I be surprised I don’t get big biceps if I use a forklift?”
The unicorn myth is stalling enterprise AI.
Paul: If the real advantage lies in asking better questions, how should enterprises rethink the way they structure their AI efforts? What does it take to succeed across modeling, data management, and translating insights into decisions?
Cassie: Enterprises often fall into the trap of hiring a so-called “unicorn,” someone expected to do it all, but believing that myth holds companies back. It assumes technical skills alone deliver value, when in reality, data science is inherently interdisciplinary. Real success doesn’t come from individual brilliance; it comes from well-orchestrated teams with distinct, complementary roles.
Enterprises must shift from “finding the right person” to “designing the right process,” since AI is a team sport. Success depends on collaboration between decision-makers, domain experts, data scientists, and engineers—all aligned around the decisions that matter.
Start with the decision, not the tool. Define what you’re trying to decide, what information supports it, who owns the outcome, and what action it should drive. That’s how you make AI useful, but decision intelligence ties it together. It creates clarity across disciplines, helps you reason through uncertainty, and ensures your AI efforts drive real, actionable outcomes.
AI is a magic lamp. The danger lies in the wisher.
Paul: Hearing AI success depends on collaboration and well-designed processes. How should organizations be thinking about their responsibility in using these powerful tools? What concerns you most about how they’re approaching generative and agentic AI today?
Cassie: When answers are cheap, the real power lies in clarity of intent. We talk too much about the genie and not enough about the lamp, or the person making the wish.
Enterprises that focus on the wisher, not just the genie, are the ones thinking in terms of decision intelligence. The danger isn’t in how powerful the AI is; it’s in how well the human formulates the request. You can make the genie as big and strong as you like, but how skilled is the wisher? How responsible? How thoughtful?
The lamp, which contains and governs the AI, is just as critical. In the world of agentic AI, that structure matters as much as the model itself.
Too many enterprises are fixated on what AI can do, neglecting how humans frame the problem. That’s where misalignment happens, not because the AI is flawed, but because the ask was misguided. The real skill enterprise leaders need isn’t prompt engineering, it’s decision design. We need more skilled wishers who know what to ask, why it matters, and what to do with the answer.
Catch the full conversation with Cassie Kozyrkov on The AI Forecast on Spotify , Apple Podcasts, and YouTube.
