About Tour de Growth
I'm Antoine Berthaud, a Senior Growth PM with ten years in product and growth (AB Tasty, SNCF Connect & Tech…). I built Tour de Growth because the fastest way to explain what growth work actually is, is to hand someone a mirror: fifteen honest questions and a score they can argue with. It is also a live case study — the tool has to grow the way it tells others to.
Why these fifteen questions
Three questions per stage of the AARRR framework — Acquisition, Activation, Retention, Referral, Revenue. They deliberately ask what you do and measure, not what you believe: is a channel identified and tracked, is an aha moment defined, is retention measured, is referral built into the product, has pricing been tested. A founder can answer in three minutes, and two people on the same team should land on the same score.
- Do you have a primary acquisition channel that's identified and measured?
- Have you tested more than one acquisition channel?
- Do you know your customer acquisition cost, even roughly?
- Have you defined a specific "aha" moment for new users?
- Do you know what percentage of users reach that moment?
- Has your onboarding been tested or iterated on at least once?
- Do you track a retention rate (D7/D30 or similar)?
- Do you have a re-engagement mechanism (email, notification...)?
- Do you know your main cause of churn?
- Does your product have a sharing or referral mechanism?
- If it exists, is it actually used by customers?
- Do you measure a viral coefficient or equivalent?
- Has your pricing model been tested, not just chosen?
- Do you know your LTV, even roughly?
- Do you have an expansion playbook (upsell/cross-sell)?
How the score is calculated, exactly
- Each question has three answers, worth 20, 7 or 0 points. No middle tier: either it is in place and measured, in place but unmeasured, or not there.
- A stage's raw total (0 to 60) is divided by 3 and rounded to the nearest whole number: that is the stage score, out of 20.
- The five rounded stage scores are added up to give the total out of 100. Rounding happens before the addition, so the numbers on the page always add up exactly.
- The weakest stage is the one with the lowest score; ties go to whichever comes first in AARRR order. It gives your result its headline and the red line on your share image.
- No AI touches the number. The score is a fixed rule, the same for everyone, and you can redo it by hand: your own result shows you the fifteen answers and the arithmetic, stage by stage.
Three answers worth 20, 7 and 7 points make 34 out of 60. 34 divided by 3 is 11.33, rounded to 11: the stage scores 11/20. Five stages at 18, 12, 8, 16 and 20 add up to 74/100, and Retention, at 8, is the weakest stage.
Where AI is used, and where it isn't
The quick result contains none: the headline and the per-stage sentences are pre-written texts, picked by score band (0 to 9, 10 to 15, 16 to 20). Only the Deep dive — ten more questions and an optional free-text field — calls a language model (Google's Gemini) to write recommendations from your twenty-five answers. Both tones and both languages are generated at once, the rule "aim at the strategy, never the person" is hard-coded into the instructions, and the score itself does not move by a single point.
Built in the open
The code is public under the AGPL, with every architecture decision and every review documented in the repository: this project is a demonstration of method as much as a tool. The tool's own growth numbers — analyses, shares, viral coefficient — are measured with the same rules it applies to everyone else. See the code on GitHub
Get in touch
A question about the method, a score you disagree with, or a growth problem you'd like an outside eye on: message me on LinkedIn, or read my background.