01
One landing, one ad campaign
New ideas get a single page and a Google Ads campaign first. Real clicks, real cost-per-lead data — before a line of product code. The older products on this board predate the method.
rathgeber.dev — build in public
Stéphane Rathgeber
I build small SaaS products and kill them fast when the market says no.
This page is the whole portfolio: live products, experiments running on real ad budgets, and honest kills — with the strategy behind each call.
14
products
4
in validation
3
live today
€0
current MRR — all of them
01 — The board
Every product, its real status.
Four stages, in the order a product goes through them. Unfold a card for the strategy — or the lesson.
-
Landing — 4 products
A landing page and ads are live. The product does not exist yet.
-
ReminderProof
Appointment reminders that reach the inbox, sent from your own domainStrategy
No ads on this one: the qualification is free and public. A DNS scan tells any dental practice whether its domain enforces DMARC — 87% of the 2,482 practices we measured do not, which means their reminders can be filtered before a patient ever sees them. That scan is both the targeting and the opening line of the cold email, and the prospect can verify it themselves in seconds. Priced at $179/month behind a payment wall, so the test measures intent to pay, not curiosity.
Live data · as of 2026-08-09
- Cold emails sent
- 60
- Bounced
- 3
- Leads
- 0
- Price tested
- $179/mo
Read so far — Day one, so nothing is decided. What the first 60 emails did settle is the plumbing. The opening batch bounced at 10% — three addresses that had a valid MX but no live mailbox behind it, because the list had only ever been checked at the domain level, never at the mailbox. An SMTP probe that stops short of sending re-qualified the whole list: 525 of 684 addresses confirmed, catch-all servers excluded because their acceptance proves nothing. The next 30 bounced at 0%. On 30 sends that does not prove a true zero — the confidence interval still reaches 11% — but the fault we identified did not recur, and it cost nothing: the $90 paid-validation service was never bought.
-
CertifiedWeekly
Certified payroll filed weekly for public-works contractors — unlimited jobs, one flat priceStrategy
No fake door on this one: the method is to recruit five design partners and build it with them. They send a real payroll week, we return a file they can actually upload, they say where it is wrong — first year free, founding price after that. The targeting comes from a public record rather than a scrape: Washington publishes both its public-works contractors and their upcoming jobs, so the 193 companies contacted all have a job starting before 31 December and apprentices on it — apprentices being the part of the filing that takes six lookups across three separate tools.
Live data · as of 2026-08-18
- Targeted
- 193
- Emails sent
- 0
- Design partners
- 0 / 5
- Offer
- 1st year free
Read so far — The whole pitch rested on one assumption: that filing certified payroll could not be automated by a file import, which would make a paid service pointless. The question went to Washington L&I on 8 August. The reply came back on the 17th, ten days later, confirming it and attaching the guide — which had been published on their own website since 2020. Worth noting is what the reply leaves out: the question listed CSV, XML, Excel and any other format, and only XML came back. No API either, so converting payroll into a valid file is entirely the contractor's problem. L&I has accepted XML uploads since 2020, schema published. That should have killed the idea. It did the opposite, because of how the state tells you to produce the file: "Program to Create/Edit a XML File: Notepad (PC), Textedit (Mac)." Case-sensitive trade codes, classifications that shift with the award date, six apprentice fields spread across three lookup tools. The vendors who automate it charge $175/month plus $995 to set up, or $145/month per job; the ones with a modern self-serve model — including one launched five months ago on this exact positioning — do not cover Washington at all. So the product moved from "we file for you" to "you never touch XML", and the method moved with it: not a fake door, but five contractors who get the first year free for building it with us. The lesson worth keeping is the week spent waiting on a government reply that was already published on its own website.
-
BeforeTheChair
Insurance benefits verified before the patient sits down, at a flat monthly priceStrategy
Same dental practices as ReminderProof, different pain: verifying insurance benefits eats front-desk hours and gets billed per verification by everyone else. Flat $199/month against an incumbent at $450/month for 50 verifications.
Live data · as of 2026-08-09
- Verified prospects
- 465
- Cold emails sent
- 0
- Leads
- 0
- Price tested
- $199/mo
Read so far — The least advanced of the three, and it has a scheduling problem rather than a product one. It shares its prospect list with ReminderProof — and a practice already contacted about email deliverability is no longer a clean test for insurance verification. 60 of the 525 verified addresses are spent, 465 remain. No cold email has been written for this angle yet, so nothing has been sent. It waits until the first test has said something either way.
-
MonLogicielPraticien
All-in-one practice software for wellness practitioners (agenda, patients, billing)Strategy
Different validation channel: free practice audits run through a Facebook community of practitioners, instead of paid ads. GDPR-first, hosted in France.
-
-
Beta — 6 products
The product exists and people can use it. Early, rough, still moving.
-
refply
Flat-fee affiliate platform for SaaSStrategy
The only one of the three fake doors that stopped being a door. An application now runs on app.refply.com — Stripe Connect and Lemon Squeezy connected, attribution read off payment-provider webhooks rather than a third-party pixel, so a conversion survives Safari and ad blockers. The private beta is invitation-only and asks for a card that it is structurally unable to charge: the billing client allowlists three Stripe endpoints and throws before any other HTTP call. The point is to find out whether someone will hand over a card at all, which no landing page can answer.
Live data · as of 2026-09-14
- Ad spend
- 793 €
- Clicks
- 277
- Leads
- 16
- Cost / lead
- 50 €
Read so far — Frozen. €793 bought 277 clicks and sixteen signups, most of them an artefact of Google's default geo setting rather than interest, and the thirty-one addresses captured on the landing have produced exactly one account in the application: the founder's own. All sixteen contactable leads were written to in August; not one replied. The two-contact rule applies, so nobody will be written to again. The site stays online and costs nothing, the campaign stays paused, and the affiliate rails that were built here will be useful the day a product with paying customers needs referrals. No hours are being spent on it.
-
redemo
Product demo videos re-rendered from your live product, not re-shotStrategy
Total dogfooding: the demo video on redemo's own landing page is produced by the very pipeline the product sells. Delivery is still concierge — each video is rendered by hand for the first accounts — so automatic regeneration is what the product is being built towards, not what it does today.
Live data · as of 2026-09-14
- Ad spend
- 1 718 €
- Clicks
- 654
- Leads
- 15
- Cost / lead
- 115 €
Read so far — €1,718 and 654 clicks bought fifteen strangers a sign-in link, and not one of them started a project. That is the whole result. The tunnel itself works end to end — brief, script, automatic capture of the live app, render, payment, published — and it was walked through in a single day in August. What did not work was the measurement: the conversion action fired on a thank-you page that no button on the landing had pointed to for months, while the app where signing up actually happens carried no tag at all. Every euro spent before 9 September was therefore unmeasurable by construction, whatever visitors did. Tag, conversion action and cross-domain linker were repaired that evening and verified live; the campaign was narrowed to six states the same day and has barely served since. Two things follow, and only one of them costs money. The fifteen who signed in and left are being asked what they were looking for. And a cold-email test is running on founders who launched their product in the last few weeks, whose demo video is already behind their release — the one claim we can make that they can check themselves.
-
Gaina
AI phone receptionist that answers, qualifies and books appointmentsStrategy
€49/month, plugged into the existing phone line without changing the number. Sold one vertical at a time rather than broadly: cold outreach to driving schools produced the first real signup, and during the beta the product is given away in exchange for feedback — early users are worth more as evidence than as revenue.
Live data · as of 2026-09-14
- Businesses live
- 7
- Calls answered
- 213
- Appointments booked
- 13
- Payment rail
- none
Read so far — Seven businesses have a working line, the agent has answered 213 calls, booked 13 appointments and taken 13 messages it sent on by SMS. Four driving schools came on in the last fortnight, all through people who know the founder rather than any campaign. The August verdict on this line — six landing pages and not one form — is now out of date: the site points at the sign-up page since 9 September. What is missing has moved, and it is more awkward. There is no way to pay. No Stripe, no enforced end of trial, an invoice written by hand. Every yes from the network becomes either an unpaid account or a manual invoice, which is why this product has real usage and zero revenue. The rail belongs to the Gaina team; until it exists, the accounts stay free on purpose and the point is to learn what these businesses actually need — the 63 questions callers asked and the agent could not answer are a better roadmap than any survey.
-
IdeaScorer
Scores your SaaS idea 0–100 from real signals (Reddit, HN, GitHub, Trends)Strategy
Eat-your-own-dogfood tool: the same market-signal scoring used to pick which SaaS ideas on this page get a validation budget. Two agents run it unattended every day, one sourcing pains, one ranking them the next morning.
-
ludotracker
Board-game tracker growing into a players' networkStrategy
Started as a board-game collection and plays tracker. The real bet: grow it into a social network of players, then monetise through affiliate links to the games they track, want, and recommend to each other.
-
Portulan
A travel folder that still works on a foreign platform with no signal, white-labelled for agenciesStrategy
Sold to the layer above the traveller: a travel agency deploys it under its own brand and hands it to its clients. The product does not plan a trip — it ingests the documents of a trip already booked, PDF tickets, hotel confirmations, photos, links, extracts the steps, has a human validate them, and shows the journey day by day with real weather and a climate-driven packing list. Everything stays readable offline. The market is not empty: mhikes, owned by Decathlon, equips around seventy trekking operators, and Voyageurs du Monde runs seven apps of its own. Two gaps remain — guided group tours, explicitly excluded from the existing apps, and document ingestion, which no competitor does. The differentiator is not offline access, which everyone promises, but having nothing to re-enter.
Live data · as of 2026-09-07
- Live since
- 2026-09-07
- Agencies contacted
- 0
- Trips created
- 0
- Competing white-label vendors found
- 7
Read so far — Deployed, healthy, and empty. The application runs in production on its own domain with a valid certificate and no errors, but the database holds no trip and no document, and not one agency has been approached. The market research came back sobering rather than encouraging: thirty-three French players were mapped, the trekking segment is locked by a Decathlon subsidiary, and the largest tour operators already run their own apps. What keeps the idea alive is arithmetic rather than novelty — the British competitor publishes its rates, and a single mid-sized operator would bring in roughly the whole monthly revenue target, where the file-transfer product would need fifty-three subscriptions for the same sum. That is a reason to keep it on the shelf, not a reason to make it the priority.
-
-
Prod — 3 products
Shipped, in production, used for real.
-
ZeroTrust Share
A shared vault opened by two separate passphrases — one for each sideStrategy
Transfer is ephemeral: a link, a few days, gone. A wealth adviser, an accountant or a lawyer follows the same client file for years, which is the wrong shape for a transfer tool and the right one for a vault. Share holds documents, encrypted in the browser before upload, and opens with two distinct passphrases — the professional has one, the client the other. We host bytes we cannot read, and the encryption layer is public under MIT so the claim can be checked rather than believed. It has no checkout of its own on purpose: entitlements are read from Transfer, so the €19 Pro subscription already wired to Stripe opens the vault quotas. Nothing to build in order to charge for it.
Live data · as of 2026-09-14
- Vaults in use
- 3
- Members
- 5
- Files stored
- 84
- Ad spend
- 0 €
Read so far — The quietest product in the portfolio is the only one being used by people who owe us nothing. Three vaults, five members, eighty-four files — and among the account holders, the advisory firm that has been recommending us. No advertising has ever pointed at it and no one has ever been asked to try it: it spread by one person showing it to another. That is the opposite of every other line on this board, where traffic was bought and nobody stayed. It still has no paying customer, and the honest reason is that nobody has been asked to pay yet.
-
zerotrusttransfer
Digital vault & zero-knowledge file transfer, hosted in FranceStrategy
Two linked modules: ZeroTrust ID for encrypted identity documents, and ZeroTrust Share for two-key encrypted vaults. Both are shipped and online; paid acquisition was stopped on 30 July 2026 and the product now waits for a channel that is not Google Ads.
Live data · as of 2026-09-14
- Accounts
- 27
- Paying subscribers
- 0
- Addresses captured
- 110
- Ad spend, two named campaigns
- 437 €
Read so far — Two measured slices, 1,976 clicks, €437, twenty-seven accounts and still not one paying subscriber. The tracking excuse is gone: the tag, the conversion action and — since 10 September — the cross-domain linker between the site and the app are all in place and verified live. What the slices actually exposed is worse and more useful. A free account gave strictly less than no account at all: the total-storage quota was only enforced for signed-in users, so an anonymous visitor escaped it entirely, and one of them stored 6.4 GB in a single session. The 2 GB per-file wall that the paid plans were sold against was never hit by anyone — largest file measured, 1,131 MB out of 1,364. There was no reason to sign up, so nobody did. Fixed on 9 September: without an account, 100 MB per file and one day of retention; with a free account, 2 GB and three days. The ladder now climbs. Ads are off, no budget is planned, and the one segment with a reason to pay €19 — professionals sending confidential documents — is being approached through people rather than through search.
-
stareply
AI-written replies to Google reviewsStrategy
Aimed at the French market. Try it without an account for three replies, then €9.99/month for one Google Business Profile with unlimited replies. A US version was considered and dropped: 230 searches a month, a price floor set by competitors at $5–19, and Google folding the feature into its own console.
Live data · as of 2026-09-14
- Price
- 9,99 €/mois
- Users
- 2
- Replies generated
- 1
- Ad spend
- 0 €
Read so far — The plumbing works and nobody is using it. Live Stripe key, a coherent €9.99 price, and two users who between them have generated one AI reply — the last on 18 July. The three contradictory prices reported in August are down to one: the brochure site was corrected on 27 August to match what the app actually charges. So this is the exact mirror of Zero Trust Transfer. One product has traffic and nothing to sell it; this one can take money and has nobody to take it from. What it does have, and no other line here does, is a target you can enumerate: French businesses with Google reviews left unanswered, where the opening line states a fact the reader can check on their own listing. That test has not been run yet.
-
-
Killed — 1 product
Stopped on its own thresholds. The lesson stays public.
-
sourcemylead
Lead-source attribution that writes into your CRMStrategy
Fake-door test: one landing page plus Google Ads, and a GO/KILL decision against thresholds written down before launch. Hard budget cap of €1,000, ads included.
Live data · as of 2026-08-09
- Ad spend
- 169 €
- Clicks
- 13
- Leads
- 1
- Cost / lead
- 169 €
Read so far — Stopped at €169 of a €1,000 budget, campaign paused since 30 July and silent since. The €13 cost-per-click was not a settings problem: once the bids were corrected and the ad-tech noise filtered out, the campaign stopped serving altogether — that price was the market, not a mistake. Keyword research then showed the wall. Every cluster holds two of the three qualities this needed, never the third. "Call tracking" has the volume — 22,000 searches a month on the CallRail brand alone — and real buyers, but demands phone tracking this product does not do. "UTM tracking" has the volume and matches the product exactly, and zero advertisers, because Google hands that tool out for free. "Lead source tracking" has the buyers and the product fit, and twenty searches a month. The landing converted at roughly 8%: the channel was wrong, not the page.
-
02 — Method
How an idea lives or dies here
02
An honest fake door
The signup form is a fake door, and it says so on the page: “private beta, onboarding within 48 hours.” Nobody gets access to a product that does not exist yet.
03
Contractual GO / KILL thresholds
Thresholds are written down before launch and read after 6–8 weeks. No reinterpretation, no extension because it “almost” worked.
04
€1,000 per idea, communicating vessels
Each test is capped at €1,000, ads included. Whatever a killed idea doesn't burn flows to the best-performing one still on the board.