mycyclesafe.ai
Ten million Indian women are diagnosed with a condition their period app was never designed for.
Problem
She downloads a period app. It assumes a 28 day cycle, then tells her she is 47 days late, every day, until she deletes it. She is on her third app.
Seven months pass, on average, between her first symptom and a diagnosis in India. Each clinic visit restarts from nothing, because a seven minute consultation cannot reconstruct eighteen months.
She takes inositol for six months because a stranger on Reddit said it worked. She never learns whether it helped. Neither does her doctor.
Source: diagnostic delay, ICMR national survey (n = 8,993, 8 states). Definition: "Priya" is a composite drawn from recurring accounts in PMOS patient forums, not a single interviewed user.
Why now
The Lancet publishes the rename of PCOS to PMOS, after fourteen years and 56 organisations, specifically to reject the cysts and fertility framing that never fit most patients.
NICE runs its consultation on adopting the name (GID-NG10436). Publication is expected on 9 December 2026. The clinical vocabulary is being rewritten right now.
Search behaviour, clinical language and guideline text all shift together. An incumbent with millions of users has the most to lose by rewording. We have nothing to unlearn.
Source: The Lancet, 12 May 2026; NICE guideline in development GID-NG10436.
Research brief
A Monash-led global consensus renamed PCOS to PMOS in The Lancet, May 2026. The clinical world moved; no consumer app reflects it yet.
13.6 to 19.6 percent of Indian women, by the ICMR national survey (n = 8,993), published in JAMA Network Open, 2024.
About 7 months pass between first symptom and diagnosis (Rao 2022, PMID 36497927). Nothing carries forward between visits.
Cycle apps log symptoms but never explain how they relate to the cycle, an analysis of 70,685 cycle-app reviews found (JAMIA Open, 2021).
Trustworthy, evidence-based information ranks first among what women want from a symptom app (Seminars in Reproductive Medicine, 2018, n = 264).
PMOS presentation and response to treatment differ person to person, so per-person statistics are the point, not a nice-to-have.
The comorbidity load behind the safety features: anxiety runs about six times the general population, depression about four times, and sleep issues are elevated, including sleep apnoea at about 37 percent versus 6 percent (Frontiers in Endocrinology, 2025).
Source: The Lancet, May 2026 (PCOS to PMOS rename); ICMR national survey, JAMA Network Open 2024 (n = 8,993); Rao et al. 2022, PMID 36497927; JAMIA Open 2021 (70,685 app reviews); Seminars in Reproductive Medicine 2018 (n = 264); Frontiers in Endocrinology 2025. Definition: the 12 tracked dimensions are a curated set mapping to PMOS symptom and comorbidity domains, capped by design, not drawn from any single "12 is enough" paper, to keep the daily check-in short and the multiple-testing correction powered.
Solution
1 · Check in
Check in
2 · Her own range
Your cycle
days since your last logged period
Your own cycles have run 27 to 52 days apart across 9 logged cycles. Today sits inside that range.
3 · The finding
What we found
On nights you sleep under six hours, your skin tends to flare about three days later.
Seen across 9 cycles. The confidence interval does not cross zero.
4 · For her doctor
Summary for your doctor
Eighteen months of cycle history, symptom trends and logged medication, on one page.
Labelled a wellness trend, never a clinical measurement.
Definition: screens are rendered from the shipping copy modules in the codebase. The finding shown is an illustrative example of the output format, not a claim about a real user.
How it works
Not a claim glued on top of the output. It refuses to guess: if the evidence is not there yet, it says so instead of inventing a pattern.
Definition: this is the product-level view. The statistical gates behind "clears the evidence bar", minimum observations and a false-discovery correction, are documented in the technical review deck, not repeated here.
What makes this different
A deterministic statistician finds real, lagged patterns in her own history. A generative layer only phrases and interprets; it can never compute a finding.
Nobody else in the category splits it this wayOne tap produces a document built from real engine output, sized for a seven minute consultation.
BuiltPHQ-4 and GAD-2 on a cadence, immediate escalation to Tele-MANAS, and not one generated word anywhere on that path.
BuiltEvery health claim carries a grade and a citation. The build fails if one does not.
BuiltAsk maps her question to two of her own tracked variables and answers from the same deterministic engine as Insight, never a generated number.
BuiltSix cycle states, a range instead of a due date, and no word for "late" anywhere in the product.
BuiltPhotograph a prescription or lab report. AI vision extracts the fields; she reviews and confirms every one before anything is saved.
Built, liveA population risk that is substantially elevated here and almost never surfaced by a cycle app.
BuiltCompanion circle
She initiates, always. She can remove someone silently, with no notification to the person removed. The message text takes zero inputs, so there is no way for a health value to leak into it, which is precisely what makes it safe to use in a household where the condition is not discussed.
Contacts, per-contact consent, a confirmed send for every message, and a Worker route that records the outreach request server side.
Through her phone's own share sheet, she picks the app and taps send herself. That is the real delivery mechanism right now, and it is exactly the plumbing the roadmap builds on.
"Good decision on not sending mood-triggered alerts to family members." "I wouldn't share my details with anyone to be honest." - Dr. Sneha Kannan, BDS, on reviewing the design.
Definition: the notify route records the outreach request; it does not yet deliver it, because no SMS or push provider is wired in. Real app-to-app push is future work, gated on a pairing code, because there are no accounts.
Document scanning
A photo of a prescription or lab report is sent, encrypted in transit, to a Worker-hosted AI model that extracts provider, medications and lab results.
Nothing is written until she reviews it. Each extracted field appears on its own line with its own on or off switch, ticked by default, every one of them untickable.
Saved to a separate encrypted document store, only after her review. Never merged silently into her daily check-in log.
Off by default and gated twice: a feature flag, and a separate consent she grants on this screen specifically, before a single byte leaves the phone.
Definition: this is live in the current build, not a mockup. It calls a real Worker-hosted model and a real extraction prompt.
Privacy and trust
Her check-ins, cycle history, symptoms and screening scores. There is no account, no login, and by default, nothing is transmitted.
A document she scans. A message she confirms to her circle. A question she asks the engine. Each one is encrypted in transit and requires an explicit action from her first.
Field-level encryption in her local store, keyed by a master key wrapped by the Android Keystore. Every feature that reaches beyond the phone fails closed to on-device behaviour if she has not consented.
Definition: an earlier claim of "nothing transmitted, ever" described a version of the product that no longer exists. What is unchanged is that nothing leaves without her taking an action that implies it.
Competition
Leads on every capability axis, and sits at the floor on shipped availability and regulatory standing. Those two are the honest gap, and they are the subject of the ask.
It is shipped, on both stores, with real users today. It also matches our privacy posture closely. It does no interpretation at all.
Registered EU MDR Class I. We hold no regulatory classification whatsoever. Josie is still a waitlist, and is cloud hosted rather than on device.
Distribution at a scale we cannot approach. Flo reached a $1B valuation in July 2024 with 77M monthly actives. Neither is built for irregular cycles.
Definition: each axis is scored 0 to 5 from published feature documentation and our own build state, by us. This is a self assessment and should be read as one. Source: pomaia.app; josie.care; Crunchbase News, July 2024.
Traction
63 unique visitors a day to mycyclesafe.ai across 52 countries, and a complete working build of the product running on device and against a deployed Worker.
There is no cohort retention curve, no net revenue retention and no payback period, because all three require paying users over time and we have none. Those arrive after the first pilot cohort. We are not going to draw them from assumptions.
Definition: "unique visitors" counts distinct daily IPs to the marketing site, not product users. The app is not yet listed on any store.
Market · who is actually reachable
Ten million diagnosed women is the number we design against. We are not citing a femtech market size, because published 2026 figures range from $32B to $73B depending on which report and which definition you take, and that spread is too wide to put on a slide.
Source: prevalence, ICMR national survey (n = 8,993, 8 states); undiagnosed share, WHO, January 2026; population, Worldometer 2026. Definition: the 18 to 35 share of India's 715.5M women is an age pyramid estimate, not a cited figure.
Market · users × price × frequency
| Reach into the diagnosed population | 0.5% | 2% | 5% |
|---|---|---|---|
| Installed users | 50,000 | 200,000 | 500,000 |
| Paying, at 3.9% healthcare freemium conversion | 1,950 | 7,800 | 19,500 |
| Price per month, midpoint of ₹200 to ₹500 | ₹350 | ₹350 | ₹350 |
| Months billed per year | 12 | 12 | 12 |
| Annual recurring revenue | ₹0.8 Cr | ₹3.3 Cr | ₹8.2 Cr |
Doubling the price moves this less than doubling reach does, and the free tier is where the safety features live, which we will not put behind a paywall. Distribution, not pricing, is therefore the first problem to solve.
Source: conversion benchmark, First Page Sage 2026 healthcare and medtech freemium study (3.9%). Definition: ARR here is gross subscription revenue before store fees and payment costs. The reach percentages are scenarios, not forecasts, because we hold no cohort data to forecast from.
Business model · unit economics
Statistical inference cost, because the engine that decides what is true runs on hardware she already owns. The AI layer's cloud cost is real but tiny, fractions of a cent per call, and is not modelled here as a per-transaction line yet.
Logging, the cycle engine, the doctor summary and every safety path. A woman does not lose a crisis screen because she cannot pay.
No user has been asked to pay yet. The price point is anchored to Indian consumer subscription norms, not validated by a single transaction.
Definition: store commission is 15% under small business programmes and 30% on the standard tier; 30% is shown as the conservative case. Methodology: structural claim about the architecture, not an observed cost per user.
Go to market
mycyclesafe.ai already receives 63 unique visitors a day across 52 countries, 2.8% of them from India, built and operated by this team. It is small, and it is genuine search traffic from women looking for exactly this.
There is no customer acquisition cost, no payback period and no channel attribution, because nobody has been acquired yet. Any number here would be invented. The first real test is whether that existing organic traffic converts to installs at all, which is measurable within two weeks of a store listing and costs nothing to run.
The second test, if the first one works, is whether a gynaecologist will hand the doctor summary to a patient without being asked. That is the channel that would actually compound, because it arrives carrying the one endorsement that matters.
Definition: visitor figures are first party analytics from the existing marketing site, not from the product.
Ask
We are not asking for a valuation on a product with no users. The three items in the first column are free, and they are the only things currently keeping this from reaching the person it was built for.