Stanford Online: AI Can Estimate Impact and Cost, but a Human Sets the Priority

Stanford Online has published a video in which Anand Subramani, a lecturer at Stanford University, and Jiaona "JZ" Zhang, Chief Product Officer at Laurel, walk through how to prioritize a product roadmap. The framework has three steps: estimate impact, estimate cost, then make a decision that a spreadsheet cannot make. According to the speakers, AI is good for stress-testing the first two steps. It does not replace the third.

A caveat up front: we only have the video's description, not a full transcript. What follows is a summary of the stated points, and any practical takeaways are marked as our interpretation.

Three Steps, Two of Which a Model Can Check

The first two steps will look familiar to any product manager. First comes impact: how many users a feature will affect, and what it will change in retention, revenue, or support load. Then comes cost: engineering weeks, dependencies on other teams, and the price of maintenance after release. Both estimates can go into a table and be sorted by the ratio of benefit to cost.

This is where, according to the speakers, AI helps. The key word in the video description is pressure-test, meaning a check for robustness. You can ask a model to find weak assumptions in an impact forecast, remind you of hidden dependencies, compare an estimate with similar projects, or ask uncomfortable questions about the numbers. It acts like a meticulous opponent in a meeting, not as the author of the final list.

Where the Spreadsheet Ends

The description calls the third step a judgment call. Two columns of estimates know nothing about company strategy, the risk of losing a key customer, team morale, or what a competitor will ship in a quarter. These factors are hard to encode as numbers, and a human is responsible for them.

The speakers offer a test question: "Why is this number one?" If the answer is "the AI decided," that is, in their words, outsourcing your own thinking. A product manager who cannot explain the logic of a priority without referring to the model does not own their roadmap.

• Format: Stanford Online video, published on 15 October 2026.
• Participants: Anand Subramani (Stanford lecturer), Jiaona "JZ" Zhang (Chief Product Officer, Laurel).
• Steps in the framework: three — impact, cost, judgment.
• What is delegated to AI: stress-testing the impact and cost estimates.
• What stays with the human: the final decision and its justification.
• Control question: "Why is this at the top?"

How to Apply This

Below is our interpretation of the speakers' points, not quotes from the video.

First: use the model after your own estimates are already written down. If you let it produce the first draft, you risk anchoring on its numbers instead of your own assumptions. Start with your table, then ask the model to find where you are wrong.

Second: ask for criticism, not a ranking. A prompt like "list five reasons this forecast is overstated" is more useful than "sort the tasks by importance."

Third: write down the rationale for a priority in your own words, in one paragraph. If you can't write the paragraph without the phrase "the model calculated," the decision has not been made yet.

Fourth: separate inputs from the verdict. AI can broaden the input, for example by adding risks and alternatives. The verdict belongs to the person who will then explain it to the team, to management, and to customers.

The material is more methodological than newsworthy. It does not announce new products or metrics, but it describes a working boundary between automation and accountability that many product teams are now drawing in practice.