Pick a product you actually use
Teardowns fail when you analyse a product you have only read about. Choose something you have used in the last month so you can reason from behaviour, not marketing pages.
Guide
Product teardowns: what they are and how to do one properly
A product teardown is a structured analysis of why a product is built the way it is — the job it serves, the trade-offs behind each decision, and what would break it. It is the single highest-leverage practice for building product sense, and the format most PM case interviews quietly test.
What is a product teardown?
In hardware, a teardown means physically disassembling a device to see what is inside and what it cost to build. In software product management, the idea is the same but the components are decisions: the problem chosen, the users prioritised, the flows simplified, the pricing set, the quality bar accepted. You are reconstructing the reasoning of the team that shipped it.
A good teardown is not a critique or a redesign. It is an explanation. If your write-up could have been produced without ever opening the product, it is a review — not a teardown.
How to do a product teardown, step by step
State the job it is hired for
Write one sentence: who has the problem, when it bites, and what they did before this product existed. Everything downstream is judged against this sentence.
Walk the core flow and note every decision
Go through the primary flow slowly. At each screen, ask what the team chose to show, hide, default, or delay — and what that choice costs them.
Analyse across fixed lenses
Unstructured teardowns drift into opinion. Fixed lenses force coverage — you cannot skip evals just because pricing is more fun to argue about.
Form a falsifiable point of view
End with a claim someone could disagree with: the bet this team is making, the risk it carries, and what you would ship next. A teardown without a POV is a feature list.
Write it up in one page
One page, per-lens verdict, one recommendation. Length is not rigour — this is exactly the artefact hiring managers ask for in a case round.
The 8 lenses for an AI product teardown
AI products need lenses that classic teardown templates never had — a chat wrapper and a retrieval-heavy workflow tool look identical from the outside and are completely different products underneath. These are the eight lenses TeardownIQ runs every session against, and the question each one opens with.
| # | Lens | The question it forces you to answer |
|---|---|---|
| 1 | Problem Definition | What job is this product hired for? Who experiences the pain most acutely? |
| 2 | Model Selection | Which foundation models are likely used? Why this model over alternatives? |
| 3 | RAG Architecture | Is RAG being used? What's retrieved, how, and how is context managed? |
| 4 | Evals & Quality | How does the team likely measure output quality? What failure modes exist? |
| 5 | UX & AI Interaction | How is uncertainty communicated? Where does AI augment vs. replace? |
| 6 | Data Strategy | What data flywheel exists? Is there proprietary training data? |
| 7 | Monetization | What's the pricing model and why? How does it align with the AI cost curve? |
| 8 | Moat & Defensibility | What would make this hard to replicate in 12 months? |
A worked example: Duolingo
Run the first three lenses on a product most people have used. Problem definition: Duolingo is hired to keep a motivated-but-inconsistent learner returning daily — the competitor is not another app, it is quitting. Model selection: generative models power explanations and roleplay conversation, while the core exercise loop stays deterministic, because a wrong answer graded by a hallucinating model destroys trust faster than a boring exercise does. UX and AI interaction: AI is scoped to places where being wrong is cheap — practice conversation, hints — and kept away from streaks and scoring, where being wrong is expensive.
Notice the shape of each verdict: a decision, the constraint behind it, and the cost of the alternative. Do that eight times and you have a teardown.
Practise a teardown instead of reading about one
TeardownIQ walks you through all eight lenses on real AI products with Shuri, a Socratic AI tutor that pushes back on thin answers and scores each lens. Sessions take 10–15 minutes and produce a shareable scorecard.