Conversion funnels with Claude: connect stages and data

Analyze the journey with Market Funnel →

The dashboard shows a number that worries the manager of a small training organization: over the last thirty days, barely two out of ten visitors who open the registration form make it to the end. The number is clear, but the reason isn't — too many fields, a page that doesn't reassure enough, a misunderstanding about what registering actually involves. Without breaking the journey down step by step, she risks fixing the wrong thing based on a hunch rather than an observation.

A conversion funnel is never read as a single overall number: it breaks down into successive stages, each with its own pass-through rate, to pinpoint exactly where visitors drop off rather than guessing from an average that blends everything together.

This article details the method for analyzing a funnel stage by stage, the difference between what a page reading reveals and what only measured data can calculate, the rules already governing the audience-measurement tools used for this work, and the limits of an analysis that never fixes the page itself nor configures the tracking tool on its user's behalf.

Contents

1. A funnel is read stage by stage, never as one overall average
2. What a page reading reveals, what only data measures
3. Trackers and audience measurement: what the law already governs
4. Analyzing a conversion funnel in five steps
5. When funnel analysis is no longer enough
6. What we observe among self-employed professionals analyzing their first funnel
7. Training to make your funnel analysis more reliable
8. Frequently asked questions

A funnel is read stage by stage, never as one overall average

An overall conversion rate almost always hides the stage that is the real problem: two journeys with the same final rate can lose their visitors at completely different points, for reasons that have nothing in common. Breaking the funnel into distinct stages, each with its own pass-through rate, reveals where to focus attention instead of randomly fixing a page that may not have been the cause.

The average rate hides the stage that's actually blocking

A single conversion rate, calculated from the first visit through to the final action, adds up losses that don't share the same cause: some visitors leave on the very first page, others abandon a form that's nearly finished. Treating both situations with the same fix — rewriting the homepage headline, for example — only solves one of the two, if it was even the real cause.

Visitors, sessions and clicks: three counters, three realities

A visitor can open several sessions, a session can generate several clicks, and mixing these three counters into one formula produces a rate that doesn't correspond to anything precisely measurable. Before any calculation, you need to know which of the three counters is actually used at each stage of the funnel, and check that the period and population being compared stay consistent from one stage to the next.

What a page reading reveals, what only data measures

A conversion funnel combines two very different sources of information, which must never be confused on pain of presenting an impression as a measurement.

A messaging break is visible to the eye, a drop-off rate is not

Reading the journey's pages can reveal an obvious messaging break — a promise worded differently from one stage to the next, a form visibly longer than the previous stage led you to expect — without ever showing how many visitors that break actually drives away. That number requires real pass-through data, comparable from one stage to the next over a consistent period and population.

The invisible stages after login

Some funnel stages happen after login or in a private area — finishing an account setup, an internal booking — and fall entirely outside the reach of a public-page reading. These stages must be flagged as such in the analysis, with an explicit note that they still need checking through a separate means, rather than being silently ignored as if they didn't exist.

Trackers and audience measurement: what the law already governs

A conversion funnel almost always relies on an audience-measurement tool, and French law specifies precisely under what conditions that tool can operate without asking for the visitor's prior consent.

Audience-measurement cookies exempt from consent, under strict conditions

CNIL specifies the conditions under which an audience-measurement cookie can be exempt from prior consent: a purpose strictly limited to audience measurement or the proper functioning of the site, exclusively anonymous data, no combination with other processing and no transmission to a third party, a cookie lifespan capped at thirteen months, and data retention capped at twenty-five months. These conditions are cumulative: a tool that meets only part of them gets no partial exemption.

A tool that shares data with advertising loses the exemption

An audience-measurement tool that transmits data to a third party, or reuses it for advertising purposes beyond measuring the site itself, cannot claim the consent exemption, whatever its configuration otherwise. Before analyzing a funnel with a given tool, actually checking whether it meets these conditions — or whether a compliant consent banner has been correctly implemented and honored by the tool — avoids basing an analysis on data collected in a non-compliant way.

Analyzing a conversion funnel in five steps

An analysis run without a method often mixes impressions drawn from reading pages with numbers coming from a measurement tool, without ever specifying which supports which claim.

Step 1: Map the actual stages, from discovery to the intended action. Listing every stage a visitor actually goes through, including those that happen after login, gives a complete view before even starting to look for where the main loss sits.

Step 2: Separate what comes from reading pages from what comes from measured data. Explicitly labeling, for each observation, its source — visual reading or measured figure — avoids presenting an impression as an established fact once the analysis is shared.

Step 3: Check that the period and population being compared stay consistent across stages. Comparing a rate calculated over one week against one calculated over a month, or a population of new visitors against one including returning visitors, produces differences that reflect nothing about the funnel itself.

Step 4: Identify the stage with the most significant loss before proposing a fix. Prioritizing a single stage at a time, the one where the loss weighs most on the final goal, avoids spreading the fixing effort across several points at once without knowing which one actually mattered.

Step 5: Formulate a testable hypothesis rather than a certainty about the cause of the loss. A messaging break spotted by reading remains a hypothesis until a test or new measurement confirms it actually explains most of the loss observed at that stage.

When funnel analysis is no longer enough

Analyzing a funnel locates the stage losing the most visitors; it does not fix that page's structure or its copy. If the identified stage is a full page with a form, visual hierarchy and trust signals, the fix belongs to a dedicated page audit rather than to the funnel analysis itself. If the problem concerns the wording of a promise or a specific piece of copy, our evidence-based rewriting method takes over once the stage is identified. Finally, without revenue actually attributed to each stage by a connected measurement tool, no analysis based on reading pages should present a revenue gain as measured: that number comes only from the tracking tool, never from an estimate deduced from the funnel.

What we observe among self-employed professionals analyzing their first funnel

In the conversations we have with self-employed professionals about their conversion funnel, the difficulty is almost never reading a dashboard: most can spot a percentage that's dropping. What holds people up is connecting that percentage to a precise stage of the journey rather than the site as a whole, and resisting the temptation to fix the first thing that looks imperfect rather than the stage where the loss actually weighs most.

Training to make your funnel analysis more reliable

Knowing how to break a funnel into stages once doesn't guarantee repeating that exercise with the same rigor every month, under the pressure of other priorities. According to the OPIIEC study of June 2025, 64% of French companies already use AI solutions and 88% expect to adopt them within three years; productivity gains observed after AI training in SMEs sit, as an order of magnitude, between 15 and 25%, a benchmark to read as a trend rather than a guaranteed result.

Frequently asked questions

Can a conversion rate be calculated without a measurement tool connected to the site?

No: an actual conversion rate requires counting visitors and completed actions over a defined period, something no visual reading of pages allows, however careful. Without a connected measurement tool, the analysis can identify probable friction from what's visible, but it must present that friction as hypotheses to verify, never as a numeric rate. Installing a compliant measurement tool remains the only reliable way to obtain that number.

Can the analysis say why visitors abandon a specific stage?

It can propose reasonable hypotheses from what's observable — an overly long form, a message inconsistent with the previous stage — but it cannot state the exact cause without additional data, such as a session recording or direct visitor feedback. Presenting a hypothesis as the confirmed cause of abandonment, without verifying it, risks fixing something that wasn't actually the cause. Confirming a chosen hypothesis takes a direct test — a session recording, a user test, or an isolated change tracked over time, as step 5 of the method sets out — not another read of the same pages.

Is consent required for the analytics tool used on the funnel?

It depends on the tool's exact configuration: an audience-measurement tool that meets the cumulative conditions set by CNIL — limited purpose, anonymous data, no transmission to third parties, capped retention periods — can be exempt from prior consent. A tool that transmits data for advertising purposes or combines it with other processing gets no exemption and must collect valid consent before gathering any data about the visitor. That qualification is checked configuration by configuration at the time of installation, not settled once and for all from the tool's commercial name alone.

What is the difference between this analysis and a landing page audit?

Funnel analysis covers the chain of several distinct stages, often across several pages, to locate where the journey's main loss sits. A landing page audit then goes deeper into the detail of a single page identified as problematic, examining its structure section by section. The first locates the problem at the scale of the journey; the second addresses it at the scale of a single page, once it has been flagged as the priority.

Analyzing a conversion funnel means breaking a journey into distinct stages, rigorously separating what a page reading reveals from what only measured data confirms, and prioritizing a single stage to fix rather than spreading the effort thin. That is exactly what Market Funnel analysis automates, leaving its user the decision on which hypothesis to test and which fix to pursue.

For a quick first look before breaking down an entire funnel, our fast marketing diagnosis helps confirm where to start; once the most problematic stage is identified, our landing page audit examines it in detail.

Training 100% fundable via OPCO/FIFPL. Qualiopi-certified programme. To learn how to analyze and de-risk your conversion funnels with AI, as part of our AI for marketers training, contact Educasium and specify your status (employee, self-employed, business owner) and your goal.