Le Diagnosticien de Fuites de Tunnel : Découvrez quelle étape du tunnel vous coûte réellement le plus de revenus
Pourquoi ce prompt est important
Teams default to fixing whichever funnel stage has the scariest-looking percentage drop, which is usually the top of the funnel where casual browsers never intended to convert anyway. A mid-funnel stage with a smaller percentage drop but a much larger user base often hides the bigger dollar opportunity. Picking the wrong stage burns an entire sprint of engineering time for a fix that doesn't move revenue, and the team doesn't find out it was the wrong call until the next month's numbers come in unchanged.
À quoi nous l'utilisons
A product or growth manager at a B2B SaaS company notices overall trial-to-paid conversion dropped this month. They have stage-by-stage funnel counts from their analytics tool and one sprint of engineering capacity before the next planning cycle, and need to decide which single stage to fix rather than spreading effort across three plausible-looking problems.
Prompt
Role: You are a senior product analyst who specializes in conversion funnel diagnostics for [TYPE OF PRODUCT, e.g., B2B SaaS signup flow]. You have run hundreds of funnel analyses and know that teams usually fix the wrong stage because they chase the biggest percentage drop instead of the biggest absolute business impact. Context: - Funnel stage data: [PASTE EACH STAGE NAME, USER COUNT AT THAT STAGE, AND THE TIME PERIOD, e.g., Visited Pricing Page: 10,000 / Started Signup: 3,200 / Completed Signup: 1,850 / Activated (connected first data source): 740 / Converted to Paid after trial: 210 — all for the last 30 days] - Average revenue or lifetime value per converted user: [INSERT VALUE, e.g., $1,800 average first-year contract value] - Recent changes in the period that might explain shifts: [LIST ANY CHANGES, e.g., new onboarding wizard shipped 3 weeks ago, pricing page redesign 10 days ago, a marketing campaign started 2 weeks ago] - Known constraints on what can realistically be fixed this sprint: [LIST CONSTRAINTS, e.g., engineering has roughly 1 sprint of capacity, cannot touch the billing system, cannot change pricing] Task: 1. Calculate the stage-to-stage conversion rate and drop-off percentage for every transition in the funnel. 2. Identify two candidate stages: the stage with the single largest percentage-point drop, and separately the stage where a modest improvement (assume a conservative 10% relative lift) would produce the largest absolute increase in revenue, using the value-per-converted-user figure provided. 3. For the revenue-impact stage specifically, propose at least two distinct, competing hypotheses for why users are leaking out there. Do not default to the obvious explanation — consider at least one hypothesis tied to the recent changes listed and at least one unrelated to them. 4. For each hypothesis, specify one concrete, cheap way to confirm or rule it out this week, before any engineering work starts (for example: a short user survey, a funnel-stage recording review, a cohort comparison against a time period before a recent change, or a support-ticket keyword search). 5. Recommend exactly one stage to focus the sprint on, with a one-sentence justification that references the revenue math, not the percentage-point drop, unless they happen to be the same stage. Constraints: - Do not recommend fixing more than one stage in this sprint, even if several stages look broken — the point is forcing a single, defensible priority call. - Do not state a hypothesis as fact. Every hypothesis must come with its cheap validation method attached. - If the revenue-per-user figure or stage counts are missing or clearly inconsistent, ask for the missing number specifically rather than estimating it yourself. - Keep the tone analytical and skeptical of the first obvious-looking answer, not reassuring. Output Format: 1. Funnel Table — stage name, user count, conversion rate from the previous stage, and drop-off percentage, for every stage. 2. Two Candidate Stages — a short head-to-head comparison: the biggest percentage-point drop vs. the biggest absolute revenue-impact opportunity, stated as a dollar range. 3. Hypotheses for the Recommended Stage — 2-3 competing explanations, each paired with its one-week validation method. 4. Final Recommendation — one stage, one sentence, tied explicitly to the revenue math.
Résultat
**1. Funnel Table** | Stage | Users | Conversion from Previous | Drop-Off | |---|---|---|---| | Visited Pricing Page | 10,000 | — | — | | Started Signup | 3,200 | 32.0% | 68.0% | | Completed Signup | 1,850 | 57.8% | 42.2% | | Activated (connected first data source) | 740 | 40.0% | 60.0% | | Converted to Paid (post-trial) | 210 | 28.4% | 71.6% |
**2. Two Candidate Stages** The largest percentage-point drop is Visited Pricing Page → Started Signup (68.0% drop). That looks alarming, but it's largely top-of-funnel browsing behavior, not a broken product experience — a 10% relative lift there (320 more signups) flows through the existing downstream conversion rates to roughly 7 additional paid customers, worth about $12,600 at $1,800 average contract value.
The largest absolute revenue opportunity is Completed Signup → Activated (60.0% drop). A 10% relative lift here means 111 additional activated users. Applying the existing Activated → Paid rate of 28.4%, that's roughly 31 additional paid customers — worth approximately $55,800. This is nearly 4.5x the revenue impact of fixing the pricing-page stage, despite having a smaller percentage-point drop. This is the stage to focus on.
**3. Hypotheses for Completed Signup → Activated** - Hypothesis A (tied to recent change): The onboarding wizard that shipped 3 weeks ago may be asking users to connect a data source before they've understood the product's value, front-loading a high-friction step too early. Validate by comparing activation rates for users who signed up in the 3 weeks before the wizard shipped against the 3 weeks after — if activation dropped sharply right at the ship date, this is likely the cause. - Hypothesis B (unrelated to recent changes): Users may not have the specific data source credentials on hand at signup time (e.g., they signed up without admin access to connect their CRM or warehouse). Validate with a 10-question exit survey triggered for users who abandon at this exact step, asking directly what stopped them. - Hypothesis C: The "first data source" step may require a data format or integration your current customer base doesn't actually have, filtering out a segment that was never a good fit. Validate by pulling the top 20 support tickets mentioning "connect" or "integration" from the last 30 days and checking whether a specific integration type comes up disproportionately.
**4. Final Recommendation** Focus this sprint on the Completed Signup → Activated stage — a 10% relative lift there is worth roughly $55,800 in projected first-year revenue, about 4.5 times the impact of fixing the stage with the largest raw percentage-point drop.
Les rapports de tunnel de conversion sont trompeusement faciles à mal interpréter. L'étape avec la plus grande chute en pourcentage attire l'attention sur un tableau de bord, mais chute en pourcentage et impact en dollars ne sont pas la même chose — et les confondre est l'une des façons les plus courantes dont une équipe produit brûle un sprint entier sur une correction qui ne fait pas bouger les revenus.
Ce prompt construit une comparaison structurée en deux étapes : il calcule chaque chute entre étapes de votre tunnel, puis classe séparément les étapes selon le revenu réel qu'une amélioration modeste débloquerait, en utilisant votre propre chiffre de valeur par utilisateur converti. Il impose de comparer ces deux classements côte à côte, car ils pointent fréquemment vers des étapes différentes — et le classement par revenu est presque toujours celui qui mérite d'être suivi.
Pourquoi cela fonctionne
La structure sépare délibérément le diagnostic de la priorisation. La plupart des prompts d'analyse de tunnel s'arrêtent à « voici où les utilisateurs abandonnent », ce qui est descriptif mais pas prêt à la décision. Celui-ci ajoute une contrainte selon laquelle une seule étape doit être choisie pour le sprint, forçant une véritable décision d'arbitrage plutôt qu'une liste de souhaits de trois choses à corriger.
Il exige aussi au moins deux hypothèses concurrentes pour l'étape choisie — une liée à un changement produit récent, une indépendante de celui-ci — car le plus grand mode d'échec dans le débogage de tunnel est de s'ancrer sur la première histoire plausible sans vérifier si une explication plus simple et non liée correspond tout aussi bien aux données. Chaque hypothèse doit être accompagnée d'une méthode de validation peu coûteuse d'une semaine, afin que l'équipe teste avant de construire plutôt que de construire sur la base d'une supposition.
Comment l'utiliser
Extrayez les comptages d'utilisateurs étape par étape de l'outil analytique que vous utilisez — Amplitude, Mixpanel, une requête SQL sur votre table d'événements, ou même un export de feuille de calcul fonctionne. Vous avez besoin de : les noms et comptages des étapes pour la période analysée, votre revenu moyen ou valeur vie client par client converti, une courte liste de ce qui a changé récemment, et une description d'une ligne de votre contrainte réelle de capacité d'ingénierie pour le sprint. Collez tout cela dans les champs entre crochets et exécutez-le.
Le résultat vous donne un tableau de tunnel, une comparaison directe des deux étapes candidates, deux à trois hypothèses testables pour l'étape recommandée, et une phrase finale liée au calcul en dollars. Cette phrase finale est ce que vous apportez à la planification du sprint — elle est défendable car c'est un chiffre, pas une intuition.