Dogfooding Growthtyping at Rewardful: 10 prototypes for $2.38

Growthtyping was built to be dogfooded before it was built to be sold. The first real test was Rewardful — the affiliate and referral platform where Tomas runs growth. The goal was simple: take real experiments from a real backlog and turn each one into a prototype the team could actually evaluate.

Across two sprints we produced ten on-brand Rewardful prototypes. Total model-generation cost was about $2.38 — $1.53 for the first five and $0.85 for the next five. That was the entire artifact cost of two experiment sprints.

The first sprint covered a live social-proof ticker, an interactive ROI calculator, a purchasing-power-parity pricing banner, a plan recommender quiz, and an in-app onboarding checklist. Each one went from pattern to rendered prototype through the normal product flow — compile the prompt against the Rewardful brand profile, render, register the experiment, export the files.

The strongest result was the ROI calculator: clear sliders, a large projected-value output, believable Rewardful copy, and the most direct tie to pricing-page conversion. The onboarding checklist was the strongest activation concept, with a credible app shell. Not everything was meeting-ready as generated — the plan recommender quiz came out clean but generic, and one render shipped a placeholder company name that needed cleanup. That is useful signal, not failure: the point of a prototype is to find those objections in minutes instead of after a sprint.

The second sprint stayed focused on pricing, activation, and referral growth: a usage-based price slider, a pricing objections FAQ, a customer logo wall, a personalized welcome survey, and a waitlist with a referral ladder. Each prototype carried a specific validation question — does showing usage math reduce plan confusion, which objections deserve above-the-fold treatment, can we learn intent without adding onboarding friction.

The second sprint also exercised the render ledger: every successful render now logs a cache key, version id, file count, and cost estimate. Nothing re-generates on page load, and every cent is accounted for.

Tomas showed the prototypes to the Rewardful team, and the team loved the output. That validates the first half of the promise: Growthtyping can produce credible, on-brand artifacts good enough to bring into a real team discussion — five prototypes at a weekly all-hands instead of one rough idea in a doc.

The half still open is the harder one, and we are honest about it: the success condition is a prototype that gets greenlit, shipped, or killed with a sharper objection because it existed. That is the proof we are capturing next.