Why MealScout is priced near compute cost

MealScout, my AI nutrition app, charges $5.99 a month with a real free tier, in a category where AI features ship at $9.99 a month or $99 a year. That is not a promotional price. It came out of a spreadsheet, and the spreadsheet is more interesting than the price.

The category prices on category norms

Nutrition apps price where the category lets them. Among the apps MealScout gets compared to: Cal AI starts around $9.99 a month, MyFitnessPal Premium+ runs $99.99 a year, MacroFactor $71.99 a year. When AI photo analysis arrived, most of the category bolted it on as a premium upsell. The feature was new; the pricing logic was inherited.

AI apps break the SaaS pricing habit

Classic software has marginal costs near zero, so pricing is pure positioning: charge whatever the value story supports. An AI app is different. Every meal photo, every text log, every coaching question is an inference call with a real cost. Cost scales with engagement, which inverts the usual logic: your most loyal users are your most expensive users.

That leaves two lazy strategies. Price at brand level and treat inference as overhead, which invites churn and undercutting the moment users notice the markup. Or subsidize usage and hope scale fixes it, which is how you lose money on exactly the customers you most want to keep.

The spreadsheet

The alternative is boring: know what an active user costs to serve, and price from cost up instead of category down. The model is one row per behavior: logs per day times cost per log, coaching messages times cost per message, times thirty. The price has to clear that with room for the free tier, the app store's cut, and the heavy users who log ten times the median.

I am not publishing my per-user costs, and the exact figures change every time model prices drop. The point is the direction of the derivation. $5.99 is what came out when the inputs went in; it is not what a pricing survey said the category would tolerate.

Pricing as a promise

The MealScout site says the quiet part out loud: AI nutrition should be priced near compute cost, not premium-brand markup. Writing that down is a constraint I chose. As inference gets cheaper, the honest moves are better models per dollar, more included usage, or a lower price. Markup pricing carries no such obligation, which is why it quietly erodes trust as underlying costs fall.

The free tier follows the same logic. Basic logging is genuinely cheap to serve, so giving it away is not charity; it is customer acquisition priced at what it costs.

The CFA part

Subscription economics are decided by churn more than by price. A $5.99 subscriber who stays two years is worth far more than a $9.99 subscriber who cancels in four months, and low prices anchored to visible logic churn less, because there is nothing to feel cheated about. Price is a retention feature. The goal is not the maximum extractable dollar this month; it is the largest area under the curve.