AI is everywhere in business — but are companies making better decisions because of it? Affiliate Professor Philipp Eisenhauer's course, Impact-Driven Business Decisions, teaches students to answer that question, from first measurement to final investment decision.
Setting the Stage
The course is grounded in a concrete setting: product data quality in large-scale online retail — where generative AI is already rewriting titles, descriptions, and images at scale and every change hits revenue directly. But how do you know if the changes are just different — or actually drive business outcomes? It's a question Eisenhauer tackled in his framework published in Harvard Business Review. The fundamental challenge is that you can never observe both outcomes: what happened with the AI change and what would have happened without it. This is the core problem of causal inference. To deal with it, students learn methods like matching, synthetic controls, and time-series modeling — each a different strategy for constructing the missing counterfactual. But measuring impact is only the first step — what matters is turning that evidence into decisions: which changes to roll back, which to scale across the catalog, and where the evidence isn't strong enough to act yet.
Connecting the Pipeline
So how do you get from measurement to decision? Most courses teach causal inference or decision theory in isolation — this one connects them into a single end-to-end pipeline. Organized around the decision loop — Measure, Evaluate, Allocate — the course walks students through three questions: What happened? — measuring causal effects. What did we learn? — scoring evidence reliability. What should we do? - allocating resources in light of uncertainty about its impact. The entire pipeline is powered by the Impact Engine, an open-source Python ecosystem Eisenhauer developed, and every lecture runs as executable code.
"Most courses teach you about methods. This one has you operating the system that turns measurement into decisions," Eisenhauer says.
Embedding AI
AI shows up at every layer of the course. The product changes students measure — rewritten titles, updated descriptions, new images — are themselves generated by large language models. A separate LLM then evaluates each causal estimate against structured diagnostics, and that confidence feeds directly into allocation: where to trust the evidence, where to stay cautious, and where better measurements are needed. Students work through the entire pipeline using AI-assisted programming.
"AI is what we're measuring. AI is evaluating what we learned. We then use both inputs to decide which AI initiatives to fund. And AI is building the whole system so we can do so automatically and at scale," he says.
Looking Ahead
Eisenhauer is already planning the next iteration. Guest practitioners will bring the decision loop into new industries — showing students that measuring impact, evaluating evidence, and allocating resources are skills that transfer. Deeper evidence diagnostics will push the boundary on what an AI system can handle and where human judgment is still needed. And cloud-native infrastructure will give students hands-on experience with the tech stack employers actually use.
"AI is changing what companies do and how they do it," Eisenhauer says. "I want to close the gap between knowing a method and shipping a decision system — and do it in the classroom, with AI."
More about Eisenhauer's work is available on his personal homepage..