Reads the runner's message with Claude (claude-opus-5). Your key goes straight from this browser to api.anthropic.com. It never reaches this site's server, and it is kept only for this tab.
A runner tells the app what happened, the way they'd text a coach. The plan adjusts using your coaching rules, explains every change, and never lets a cold or a work trip turn into a catch-up week before race day.
“Calf sore for two days, hurts to run, about 6/10.” The AI only turns the message into structured facts: what, how bad, how long.
Illness above or below the neck, injury by pain level, travel, no catch-up miles, taper protection. Plain rules you and Clayton own and can change.
Every changed day is highlighted with the rule behind it, in the weekly table and in a 3D view of the whole training block.
Orange edge = workout, dark edge = long run, yellow = changed by a rule. Final 3 weeks (taper) in orange.
Based only on public information (App Store listing, runaccomplice.com) and the numbers you shared. Every point is a hypothesis to verify in the app and in PostHog once I'm on TestFlight.
Because half of trials already convert, the product isn't the problem: people who try it, buy it. Every extra trial is worth a lot, so the job is to get more of the right people into the trial. Per 1,000 downloads: 170 trials → ~85 members today. At 30% trial start, even if trial→paid drops to 45%, that's ~135 members.
A serious runner downloads to see their plan. If the paywall arrives before they see it, many leave. Ask three things (race date, goal time, current weekly miles), then show the first two weeks of their personalised plan and their race-day pace, then offer the trial to unlock the rest.
Measure: % of downloads reaching the paywall and trial start among them; A/B against the current order.
The website says $297/year; the in-app purchase is $349. Anyone who compares feels overcharged inside the app. Options: same price in both places, or (in the US, where apps may now link to web checkout; worth confirming the current Apple rules) a “join on the web for $297” link that also saves Apple's fee.
The site sells seasonal builds (“Join the Fall Marathon Build”). Someone whose race is in spring may not want a trial clock running now. Offer “Start my trial when my build starts” (schedule + reminder), or a free base-building phase that ends in the trial.
Sign in with Apple; only the three plan questions up front. Watch connection (Garmin, Coros next), notifications and profile come after the trial starts, when people are motivated to finish setup.
14 ratings at 5.0 is excellent quality, but few. Ask for a rating at a happy moment (first long run completed, first weekly check-in) with the system prompt. More ratings lift store conversion, which feeds the top of this funnel.
PostHog funnel: app_open → onboarding_q1 → onboarding_complete → plan_preview_seen → paywall_seen → trial_started → day5_reminder → converted. One change at a time, each long enough to reach a few hundred paywall views, so the effect isn't noise.
Plan adjustment: the Plan adjuster tab shows how I'd build it: the AI reads the runner's message, and explicit coaching rules change the plan and explain each change. Your v1.2.1 already has weekly check-ins and plan adjustments; this could slot in behind them.