Product playground
Things I think about when I'm not inside a roadmap.
Short product memos. Each one states the bet and how I'd know it was wrong, because a product opinion is only worth reading if it can be tested.
These are analyses of products I don't work on — hypotheses, not results.
Onboarding critique
Bundled streaming subscriptions ask for the decision too early
The bet
Moving plan choice to the first playback boundary — rather than the first screen — converts better because value has already been demonstrated.
Success would mean
Trial-to-paid conversion within seven days, and drop-off at the plan screen.
Open teardown →↑
- Observation
- Telecom-bundled streaming apps typically surface plan selection before the viewer has seen anything worth paying for.
- Friction
- A first-time viewer is asked to compare tiers with no reference point for what the tiers actually change about their evening.
- Proposal
- Let the first session play, then interrupt at a natural stop with a two-option choice framed in content terms, not feature-matrix terms.
- Risk
- Revenue recognition shifts later; support needs a clear story for viewers who expect an upfront price.
AI product analysis
Most AI assistants communicate certainty they haven't earned
The bet
Showing the basis of an answer — sources, coverage, or an explicit 'not covered' state — improves the quality of user decisions more than improving the model does.
Success would mean
Rate of users who verify or escalate when the answer is ungrounded.
Open teardown →↑
- Observation
- Assistant interfaces present a confident answer and a weakly-grounded answer in exactly the same visual language.
- Friction
- Users calibrate trust on tone, then over-trust in the cases where the model is least reliable.
- Proposal
- Design a grounded-answer state and an out-of-scope state as first-class UI, with the out-of-scope state routed to a human path.
- Risk
- Too much hedging reads as a weak product; the states have to be quiet, not apologetic.
GTM experiment
Clinician-facing AI should be rolled out by workflow, not by feature
The bet
Sequencing rollout around a single workflow — intake, say — produces higher sustained usage than launching capability-by-capability.
Success would mean
Sustained weekly usage at week eight, not activation at week one.
Open teardown →↑
- Observation
- Research on AI adoption in mental healthcare points to providers weighing clinical-workflow fit over raw usefulness.
- Friction
- Feature-based rollouts land tools in the middle of a workflow that nobody adjusted, so usage decays after the pilot.
- Proposal
- Pick one workflow, instrument it before launch, ship the AI step inside it, and only expand once the workflow metric holds.
- Risk
- Slower surface-area growth, and a narrower story for stakeholders who want breadth early.
Contact
Let's build something people actually want to use.
I'm exploring full-time Product Management opportunities where I can work across AI/data, thoughtful design, and go-to-market execution.