AI training for product teams

How should a product manager use AI to synthesise interviews?

Use the assistant to organise the notes, then check every theme against the notes themselves, counting how many people actually said it. Here is a worked example from a fictional scheduling app, Tandem. A product manager has notes from six customer interviews, with names removed. One participant asked for a dark mode. Four described trouble rescheduling a recurring meeting. Two mentioned the price.

The prompt: "Group these interview notes into themes. For each theme, give the number of participants who raised it and one short quote. Do not generalise beyond what the notes say."

A weak synthesis, from a prompt that just asked for key insights, opens with "Users strongly want dark mode" and "Pricing is a major concern for most customers." One mention became a strong want, two became most. Read by a busy stakeholder, that summary could move the roadmap.

The reviewed synthesis reads: "Rescheduling recurring meetings (4 of 6), with a quote; price (2 of 6); dark mode (1 of 6)." It is less exciting and much more useful. The product manager has also kept the original notes alongside, so anyone who doubts a theme can read the source.

What goes wrong when product teams rely on AI?

Three things go wrong most often: the summary overstates the evidence, a competitor or market claim is invented, and confidential material ends up in an unapproved tool. NIST's Generative AI Profile describes confabulation as confidently stated but false content, and that is the specific danger in market scans that cite numbers or features the assistant cannot verify.

The habits that help are concrete. Count mentions instead of accepting adjectives. Open the source behind any market or competitor claim before it goes into a deck. Remove customer names and confidential roadmap details unless the tool and plan are approved for them.

What should a product team practise first?

Begin with synthesis, because it teaches the evidence habit every other task depends on. This sequence is a suggestion, not a measured result.

When is this not the right training for a product team?

It is not right if your goal is to learn to build AI features, evaluate models or write code; hands-on engineering is out of scope, although machine learning foundations are planned. It is not right for a single product manager learning alone, because the plans are built for teams with a five-seat minimum.

It does not include SSO, LMS integrations or a browser extension at launch, and lessons are English only. Because the product is pre-build, a team that needs material now should start with the free product management role guide.