Digital strategy
B2B conversion funnel: measure from visit to contract
Short answer
A conversion funnel is a chain of denominators, not a drawing.
A useful B2B funnel identifies who could complete each step, what counts as completion, when it occurred and which system holds the evidence. It separates discoverability, web behaviour, valid enquiries and commercial outcomes. Without those definitions, a percentage can be precise and still describe the wrong population.
This edition does not claim an end-to-end Edikka conversion rate. Traffic is currently too limited and only valid contact submissions are recorded server-side. The absence of a headline percentage is a result of the method, not a missing result.
Editorial boundary
The funnel measures the chain; CRO decides what to change.
| Question | Owned here | Owned by CRO |
|---|---|---|
| Where does a transition lose eligible cases? | Population, event, denominator, time window and status. | Uses the observation as an input. |
| Why does the loss happen? | The cause remains a hypothesis. | Diagnosis and evidence selection. |
| Should a change be kept? | Supplies the measured transition. | Decision rule and guardrails. |
Units
Sessions, people, enquiries and opportunities are not interchangeable.
The denominator must be the population that was eligible to complete the next step. A page view cannot silently become a person; a form submission cannot silently become a qualified opportunity. Repeated visits, duplicate requests, spam rejection and reopened opportunities all change the unit.
Web layer
Exposure and interaction
Search impression, landing session, CTA exposure, form start, validation error and successful server response.
Commercial layer
Enquiry and business status
Valid request, qualified lead, working lead, won, lost or disqualified. These states use the business record, not a browser cookie.
Documented join
Relate the layers without adding their percentages
A controlled identifier may connect a request to its commercial history. It does not turn web sessions and opportunities into the same statistical population.
Measurement contract
Every transition needs seven fields before it gets a rate.
| Field | Question to close | Failure prevented |
|---|---|---|
| Population | Who was eligible? | Inflated or incomparable denominator. |
| Completion event | What exact observable state closes the step? | Clicks mistaken for completed outcomes. |
| Identity rule | Session, device, person, request or opportunity? | Silent deduplication changes. |
| Window | Which start, end and cohort dates? | Incomplete cohorts compared with mature ones. |
| Source | Which system is authoritative? | Two tools producing two truths. |
| Exclusions | Spam, tests, duplicates, internal traffic? | Operational noise treated as demand. |
| Limit | What does this observation not prove? | Correlation promoted to causation. |
Calculations
Four formulas are enough when their populations are explicit.
Transition rate
Completed next step ÷ eligible previous step
Both counts use the same cohort and observation window.
Loss rate
Eligible previous step minus completions ÷ eligible previous step
A loss location does not establish the reason for the loss.
Qualified enquiry rate
Qualified requests ÷ valid requests
The qualification rule must be written before the period is analysed.
Median delay
Median elapsed time between two dated states
Open or immature cases remain visible; they are not silently discarded.
Edikka · public state
One server event is active; the rest remains explicitly unmeasured.
| Stage | Status | Current evidence | Permitted conclusion |
|---|---|---|---|
| Organic discovery | External system | Search Console impressions and clicks. | Discoverability, not complete site traffic. |
| Page and CTA exposure | Not instrumented | No exhaustive first-party event. | No exposure or click-through rate. |
| Form start and error | Not instrumented | Accessible error states exist; no aggregate counter. | Implementation can be tested, frequency cannot. |
| Valid contact request | Instrumented | generate_lead after server validation and successful storage. | Count of recorded submissions, not unique people. |
| Qualified, won or lost | Not instrumented | No versioned public status journal. | No lead-to-contract rate. |
Data minimisation
The measurement record stores the event name, UTC time, language and a controlled source context. It stores no name, email, message, IP address or user agent. A logging incident must never prevent delivery of a valid contact request; it is reported server-side and narrows the measurement coverage.
Low traffic
With a small sample, publish counts and uncertainty—not decorative percentages.
A young B2B site may receive too few valid requests to estimate a stable rate. That does not make measurement useless. It changes the questions: are events valid, are sources traceable, how many cases entered each state, which cases are still open, and how long has the observation window been running?
Edikka will not publish a rate from an immature cohort merely because the division is possible. The default status is insufficient data until the denominator, observation window and completed cases support a business decision.
Interpretation
A funnel locates a loss; it does not diagnose its cause.
A lower transition may coincide with an unclear proposition, an inaccessible form, low-intent traffic, a pricing mismatch, a technical failure or a deliberate qualification filter. The recorded fall cannot distinguish those explanations alone.
The correct handoff to CRO contains the affected population, the transition definition, the observation window, raw counts, anomalies and competing explanations. CRO then chooses whether the next proof should be a technical retest, usability session, content review, sales analysis or controlled experiment.
Governance
A small measurement system still needs an owner and a change log.
- Record the definition before interpreting the result.
- Keep raw counts beside every percentage.
- Version changes to event names, exclusions and qualification rules.
- Use UTC for event storage and state the reporting timezone.
- Keep failed, open and out-of-scope cases visible.
- Recalculate historical figures only when the rule and impact are documented.
Architecture
Continue with the page that owns the next question.
Primary references
The sources define collection and privacy constraints—not Edikka results.
- Google Analytics · Recommended events — names the lead-generation and commercial events used as an interoperability reference; this page does not claim that Edikka runs GA4.
- Google Search Console · Performance report — defines search impressions and clicks, not complete site traffic.
- CNIL · Audience measurement solutions — frames the conditions and limits of consent-exempt audience measurement in France.
- W3C · Privacy Principles — supports purpose limitation, data minimisation and transparent processing.
Method limit: this is an Edikka measurement contract, not a Google, CNIL or W3C standard. Version 1.0 was reviewed on 4 September 2026. Next review: 4 December 2026.
Measuring a funnel means preserving the meaning of every transition—not filling a dashboard.
Useful measurement connects exposure, action and business outcome without silently mixing sessions, people, enquiries and opportunities.
One count, one population
Every figure names its unit, eligible denominator, observation window and source before a rate is calculated.
Measure what informs a decision
A small, stable measurement system is more useful than an exhaustive dashboard built on shifting definitions.
Keep the unknown visible
Low volume, missing coverage and attribution limits are published as states instead of being hidden behind a percentage.
The objective is not to display activity. It is to preserve the meaning of every count from first exposure to business outcome.
Measure without mixing populations
Eight answers about denominators, web and sales layers, low traffic and Edikka’s actual coverage.
