Shipping a Personalized Discount Engine: Robinhood's Launch Decision
Objective: Navigate a 3-stage simulated rollout of a personalized subscription-discount engine for Robinhood Gold, choosing at each stage between options that trade off conversion lift against regulatory and PR risk, and land on a launch decision that would survive a Senate letter.
Robinhood's growth team has built a model that predicts which free-tier users are most price-sensitive and can serve them a personalized Robinhood Gold trial discount. Legal, growth, and comms all have a stake in how it ships.
At each stage, pick the option that best balances conversion lift against the lesson's aggregate-vs-individual line and disclosure test, not the option with the highest modeled lift.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, enough to log a 3-stage decision trail
Paid upgrades (optional, faster/deeper)
The free path (Sheets alone) is sufficient to complete every stage; ChatGPT only speeds up writing the leadership summary.
Speeds up turning a raw decision into board-ready language; the decision itself still has to be made by the team
The process
Simulation
Choosing the model's input signals
Week 1Data science has three candidate signal sets ready to feed the price-sensitivity model, and wants a decision before building the pipeline.
Modeled lift estimates: Device+location+browsing history = +14% trial conversion. Loyalty tier + in-app engagement only = +6% trial conversion. Account tenure + open Gold seat inventory = +3% trial conversion.
Which signal set should feed the price-sensitivity model?
Final deliverable
A 3-stage decision log recording the chosen option, verdict, and reasoning at each stage, ending in a specific launch recommendation (scale, percentage, and monitoring window).
See a reference example
Casper Sleep, personalized-offer launch decision log (excerpt) Stage 1: Signal selection -> Chose loyalty tier + engagement (+6% modeled lift), rejected device/location/browsing despite higher lift Stage 2: Disclosure test -> Ship with visible label, accepted lift drop to roughly half Stage 3: Launch call -> 10% rollout, 4-week monitoring window before scaling Recommendation to leadership: Launch limited, disclosed, and monitored. Full-scale undisclosed launch was rejected at every stage it was offered.
Success criteria
You're done when you can:
- Chooses the optimal-verdict option at least twice across the 3 stages
- Final recommendation includes a specific rollout percentage and monitoring window, not just 'launch' or 'don't launch'
- Reasoning at each stage references the lesson's aggregate-vs-individual line or disclosure test by name