A plate is pixels. A menu is text: a dish name plus a description that almost always lists the ingredients. That single swap is the whole technical thesis, and it has a number behind it (see 04).
The output is not a calorie count. It's a sentence you say out loud to a server. "Dressing on the side" is correct whether the salad is 400 calories or 550, so the advice survives the uncertainty that sinks everyone else in this category.
Dressing, sauce, aioli, glaze. All of it on the side.
Rice for greens. Fries for fruit. Bun off.
Double the protein. Egg on it. Extra veg.
Ranges, not single numbers, everywhere a number appears. The confidence pip states out loud where the data came from: published chain data, or an estimate off the menu text.
The menu scan is the hook. The fridge scan is the habit. It's what keeps the app open on the five days a week nobody goes to a restaurant, and it is the feature no credible competitor has.
It also runs at a deliberately lower accuracy bar, and that's the point. Identifying food in a photo is 68-86% accurate[3], plenty for "you have chicken, rice, broccoli, eggs." The part that wrecks photo-based calorie apps is portion estimation, which bottoms out near 39%[3]. Here it never applies: you're standing at your own counter, about to portion it yourself.
Not "women's fitness." Not "postpartum." The person who has a number to hit and gets handed a menu. That population is an order of magnitude larger than Andrea's audience and it includes it, which is exactly why this beat the alternatives.
They already pay for adjacent things, and the ladder is well documented. Tracking sits at $80-100 a year. The moment they want a human to tell them what to do, the price jumps 15×. This product sits in the gap: judgment, at software prices.
| What they already buy | Price | What it doesn't do |
|---|---|---|
| MyFitnessPal Premium | $79.99-99.99/yr | Logs what you already ate |
| MacroFactor | $11.99/mo | Adaptive targets, no ordering help |
| Cronometer Gold | $59.99/yr | Micronutrient depth, same gap |
| MenuFit | $9.99/mo | Restaurants only, data "way off"[2] |
| Working Against Gravity | from $99/mo | Human coach, capacity-limited |
| Macros Inc | $139-199/mo | Human coach, capacity-limited |
| Andrea's 12 Week Cut | cohort, sells out | Turns away demand by design |
Prices from vendor pages, collected 12 Aug 2026[5]. The cohort selling out is not a problem to fix. The scarcity is what generates the demand. This is the product that catches the overflow without touching it.
This is the part that is genuinely defensible, and it costs nothing to implement. It is a choice about what to point the camera at.
The FDA requires every chain with 20+ locations to publish calories and provide the full nutrient breakdown on request, compliance date 7 May 2018. That regulation is the only reason a chain database exists. Nutritionix carries 202,000+ menu items across 209,000+ US locations, starting at 200 calls/day free and about $50/mo at hobby tier[3]. This is a licensing decision, not a research project.
The rule doesn't bite under 20 locations, so thousands of independent restaurants have no published nutrition data at all. It cannot be licensed because nobody has it. That is exactly why MenuFit's numbers read as "way off" on local places, and it is the whole opportunity. Independents are a language problem, not a database problem. Menu text in, order recommendation out.
Cole Kosco, a nutrition coach in Palm Beach, launched one app on 6 August 2025 and hit $60,000 MRR in two months, solo[1]. Same shape as Andrea: a coach with an audience about a topic, shipping an app about that topic.
"Products exist and are bad" is the best category to walk into. Demand is proven, price is established, and the incumbent is losing on the exact axis a careful builder is strongest on. The lifetime rating hides it. The recent weekly number is where the truth is.
Every one of those five failures is a design decision we make differently on day one, and three of them are visible in the first 90 seconds of the app. That is the entire competitive plan.
This was not the first candidate. It won on one test the others could not pass: how long the need lasts.
| Candidate | Why it looked good | What killed it | |
|---|---|---|---|
| Scale the 12 Week Cut into an always-on app | Sold out every cohort, real waitlist, obvious unmet demand | The sell-out is the demand engine. Turning a scarce cohort into an always-available product converts "this closes and you'll miss it" into "buy it whenever." Never scale the scarce thing. | Killed |
| Postpartum + breastfeeding nutrition app | The single best-documented hole found: MFP has no lactation setting, decade of unanswered forum requests | Duration of need. The window is 6-18 months, once or twice in a life. Pregnancy apps churn ~70% by postpartum week 4. And a deficit recommendation to a lactating woman is a clinical call, which is why the hole is open. | Killed |
| Perimenopause strength + nutrition | Strongest evidence in the whole research pass. Need lasts 4-10 years. Midi proved $29/mo at $1B | Andrea is in her 30s with a newborn. Zero lived authority, and borrowed authority in a health category is the fastest way to lose a 138k audience. Peloton also shipped Trainwell-powered coaching into it in Feb 2026. | Killed |
| Body scan / 3D progress visualization | Strong single frame, very shareable | Across 5,272 read comments, essentially nobody asked for it. Valuable negative finding. | Killed |
| Menu scan + fridge scan | Eating out never ends. Recurs at every restaurant visit and every night at home. | Unsolved: reach beyond Andrea's audience, and whether the advice is a subscription or a lesson. See 09. | Lead |
Anyone can ship Mifflin-St Jeor in an afternoon. What can't be cloned in a weekend is a documented decision tree reverse-engineered from 160+ real client prescriptions across two 12-Week Cut cohorts, a coaching email archive, and the Mamele Macro Handbook. It already exists, written down, in HQ/skills/mamele-macro-coach/.
The special cases are the moat, not the base formula. A generic calculator has no opinion about a breastfeeding mother, a PCOS client on Metformin, or someone who's been at 1,380 calories for fourteen weeks and stopped losing. This one does, and the opinions came from outcomes.
138,758 combined followers across @mamelefit and @andreafausett, an email list, and a cohort that sells out. That is distribution funded competitors are spending nine figures to rent. Ladder's General Catalyst deal has GC paying 80% of their sales and marketing spend[5]. The audience is the moat. The software is the delivery mechanism.
This is genuinely undecided and both directions have a real argument. Here they are honestly, then the recommendation and the reason.
"Mamele Macros" or a feature inside the existing Mamele app.
"On the Side." Andrea credited quietly on the macro method, loudly as a user.
Go separate. Call it "On the Side," with the line "macro method by Andrea Fausett" in the About screen and the App Store description, and nowhere else. The deciding argument isn't the credibility point, real as it is. It's the ceiling: the only test this concept has never passed is escape velocity beyond the warm audience, and Mamele branding pre-answers that question with "no." Keep the option open until the data closes it. If it turns out the app only ever sells to her list, folding it into Mamele later is a rename. Un-narrowing a brand later is not.
Base rate for reference: roughly 0.8-1.0% of launched apps reach $50k MRR within two years, and the median app is making $72 a month a year after launch[4]. Everything below should be read against that.
The real advice space is small: sauce on the side, swap the starch, double the protein, skip the glaze, bun off. Once a user has learned twelve asks, they own the product. The research argues duration of need is unlimited because eating out never ends. But the situation recurring is not the same as the need for advice recurring. Andrea's own methodology says the quiet part out loud: tracking is a learning tool, "not something you have to rely on forever." That philosophy is correct, and it is a churn engine.
MenuFit's users complain the numbers are "way off," which tells us they want a number and check it. Answering "510-600" is more truthful and may read as "the app doesn't know." The design has to make uncertainty feel like expertise rather than evasion, and there is no proof yet that it can. This is the core UX bet of the whole product and it is unvalidated.
Chains are a licensing decision. Independents have no published data anywhere and cannot be licensed, which is the opportunity, but it also means every number on a local menu is an inference from a sentence someone wrote to sell a dish. A menu that says "chef's special sauce" gives us nothing. Menu photos in dim restaurants, multi-page menus, chalkboards and QR-only menus all degrade the input before any model sees it.
138k followers with 4-month average retention supplies roughly 500 new customers a month for a while, and then is exhausted. Every other channel is priced against us: median revenue per install across the market is $0.31 against US Apple Search Ads at $4.06 per install[4], and ~560,000 new iOS apps shipped in the first half of 2026 alone while meaningful-usage apps stayed flat. A hard paywall fixes the unit economics on paper ($3.09 vs $0.38 revenue per install at day 60). It does not manufacture an audience.
Don't design around it. The MyFitnessPal API is private and closed to new partners. Their own developer portal states they are not accepting requests[6]. The workable path is Apple Health: MFP writes meal summaries and nutrients to Health, so we can read what's been eaten today and compute what's left. It will not read our meals back into MFP. Plan for a one-way read, treat any partnership as upside, and make the app complete without it.
Phase 0 has no code in it on purpose. Median time to $1k MRR across the market is 58-60 days, so a bet this size should show a pulse inside two months or be dropped.
30-50 people from Andrea's list text a photo of a menu to one number. A human, backed by Claude, texts back the ask within a few minutes. No build, no store listing, no design.
Onboarding on Andrea's documented method, the reveal before the paywall, camera → menu → the ask with ranges and confidence. Nutritionix free tier for chains, LLM inference for independents. No fridge scan, no Android, no integrations.
The retention half. Item detection, remaining-macro matching, saved meals, Apple Health read. Ship only after phase 01 shows people opening the app on days they don't eat out.
Andrea's three content formats, comment-to-DM automation, creator referral codes, the App Store listing built around long-tail phrases with real intent: "what to order," "restaurant macros," "eating out healthy." Paid acquisition only after the paywall economics clear.