{"id":1374,"date":"2026-09-24T08:00:00","date_gmt":"2026-09-24T13:00:00","guid":{"rendered":"https:\/\/www.seomos.com\/?p=1374"},"modified":"2026-09-10T03:22:28","modified_gmt":"2026-09-10T08:22:28","slug":"embudo-de-conversion-marketing","status":"publish","type":"post","link":"https:\/\/www.seomos.com\/en\/blog\/embudo-de-conversion-marketing\/","title":{"rendered":"Conversion funnel: stages, metrics, and drop-off points"},"content":{"rendered":"
\n

A conversion funnel represents how many units in a cohort advance between observable steps toward a valuable action.<\/strong> You can model everything from a visit to a purchase, or from a read receipt to a qualified lead. To serve, you must define the unit, events, criteria, window, and outcome; then you combine rates with volume, time, quality, and experience.<\/p>\n<\/div>\n

The funnel shape is a simplification. People return, use multiple devices, compare channels, and change their intentions. The model shouldn't simulate a linear journey; it should help pinpoint a question and lead to a decision.<\/p>\n

What is a conversion funnel?<\/h2>\n

It's an analytical sequence that counts entries and exits between states. A store might use session with product, cart, checkout, payment, and purchase. A B2B company might use visit, content, valid form, and accepted meeting.<\/p>\n

The final conversion rate depends on the goal. Don't treat every interaction as a success. A play or scroll might explain behavior, but it doesn't equate to revenue.<\/p>\n

Funnel, customer journey and sales funnel<\/h2>\n

The funnel quantifies progress; the customer journey describes tasks, channels, emotions, and context; sales funnel<\/a> It usually encompasses qualification and closure. They can be related without sharing identical stages.<\/p>\n

In e-commerce, behavior is recorded in minutes; in consultative selling, a single stage requires a human decision. Don't mix sessions with opportunities on the same rate without a defined bridge.<\/p>\n

Choosing the unit of analysis<\/h2>\n

Session, user, account, cart, order, and opportunity are not equivalent. A person can have multiple sessions or orders. Define what counts at each step and how to deduplicate it.<\/p>\n

If the unit changes, show the explicit transition. For example, 10,000 sessions produce 600 forms and 420 unique contacts. Hiding the change makes consolidation appear as if it's a leak.<\/p>\n

How to define stages<\/h2>\n

It includes meaningful changes: relevant arrival, exploration, intent, action initiation, confirmation, and validated result. Each stage requires an event, condition, fields, exclusions, and owner. Avoid creating a stage for every micro-click.<\/p>\n

The sequence must be explainable to both product and business. If no one knows what decision an event represents, it probably doesn't deserve to be in the main funnel.<\/p>\n

\"Secuencia
The useful stages correspond to observable decisions or commitments.<\/figcaption><\/figure>\n

Instrumentation with events<\/h2>\n

Google Analytics lets you measure interactions through events and parameters, with automatic, recommended, and custom categories. Use the recommended event when appropriate and create your own only if the task requires it.<\/p>\n

Define name, trigger, parameters, unit, consent, and test. Validate in the browser and DebugView, then verify processing in reports. A real-time visible event does not prove that the value, currency, or identifier is correct.<\/p>\n

Key events and results<\/h2>\n

Key actions that are important to the business. Don't mark every step: you could optimize resources for easy signals. A purchase, successful shipment, or confirmed reservation is usually closer to a value than a page view.<\/p>\n

In B2B, connect CRM to identify opportunities and close sales. Maintain micro-actions for diagnostic purposes. Define how duplicates, cancellations, and returns are handled.<\/p>\n

How to calculate conversion rate<\/h2>\n

Stage rate = units completing the next step \u00f7 eligible units entering the step. Total conversion = results \u00f7 initial cohort. Express numerator, denominator, and period.<\/p>\n

Don't use the total number of events if a person can repeat them. Choose users, sessions, or transactions depending on the question. Show volume alongside the percentage: 50% from two cases is not solid evidence.<\/p>\n

Cohorts and maturity windows<\/h2>\n

Group visitors by entry within a specific period and allow sufficient time for completion. Comparing yesterday's visitors to a 30-day cohort penalizes recent visitors. Use curves at 1, 7, 30, or more days depending on the cycle.<\/p>\n

Separate open or incomplete items. Subscriptions, acquisition, and retention require different windows. Document time zone and measurement changes.<\/p>\n

Time between steps<\/h2>\n

The rate can be maintained even as the journey slows down. It measures medians and percentiles between events. An average is distorted by extreme cases. It differentiates active waiting time if the data exists.<\/p>\n

The delay could be due to a website issue, approval process, support, or customer issue. Assign the step to the team capable of intervening.<\/p>\n

Segment without inventing stories<\/h2>\n

Compare device, channel, region, audience, product, and cohort. Start with segments that are both realistic and large enough. Don't explore hundreds until you find a random difference.<\/p>\n

Keep context in mind. A lower mobile rate may reflect mobile research and subsequent purchase on desktop, or it could be a real barrier. Use data and qualitative evidence.<\/p>\n

What is a vanishing point?<\/h2>\n

This slowdown or stagnation between stages warrants investigation. Some of the decline is normal: not everyone needs to buy. The problem arises when there is avoidable friction, poor procurement, insufficient capacity, or faulty measurement.<\/p>\n

Don't call every outflow a leak. Define the expected range and segment value. A filter that excludes unattended profiles can improve the system even if it reduces volume.<\/p>\n

\"Embudo
A leak is diagnosed by separating behavior, quality, operation, and data.<\/figcaption><\/figure>\n

Method for diagnosing a fall<\/h2>\n
    \n
  1. It confirms that it exists in a mature cohort.<\/li>\n
  2. Validates events and definition changes.<\/li>\n
  3. Locate step, segment, and date.<\/li>\n
  4. Check speed, errors, and compatibility.<\/li>\n
  5. Observe sessions with adequate privacy.<\/li>\n
  6. Interview users or team.<\/li>\n
  7. Formulate alternative causes.<\/li>\n
  8. Prioritize a limited trial.<\/li>\n<\/ol>\n

    Acquisition and quality of input<\/h2>\n

    A channel can increase visits but decrease conversions because it broadens the audience. Evaluate intent, message, and destination. Don't blame the checkout process before reviewing who enters the funnel.<\/p>\n

    Relate UTM<\/a> With campaigns based on conventions, without overwriting the original source. Compare subsequent quality and margin, not just the immediate rate.<\/p>\n

    Friction of experience<\/h2>\n

    Errors, slowness, fields, navigation, trust, and accessibility block actions. Mobile test, keyboard, messages, failed payments, and slow connections. A heat map<\/a> It provides signs, not causes.<\/p>\n

    It combines analytics, usability testing, support, and technical review. Recordings require consent and data protection.<\/p>\n

    Capacity and operation<\/h2>\n

    A digital conversion can fail later: missed calls, no available appointments, out-of-stock inventory, or late follow-up. Incorporate availability and SLAs into the funnel.<\/p>\n

    Optimizing the form to generate more leads can backfire if sales can't handle them. The overall result matters.<\/p>\n

    Design an experiment<\/h2>\n

    Write down the evidence, hypothesis, change, population, primary metric, guardrails, duration, and criterion. Change one plausible cause and maintain other conditions where possible. Calculate size and avoid stopping at the first positive result.<\/p>\n

    Guardrails protect quality, revenue, accessibility, and returns. A shorter form isn't beneficial if it generates more invalid leads.<\/p>\n

    \"Mesa
    A responsible test seeks learning and protects results that should not be sacrificed.<\/figcaption><\/figure>\n

    Prioritize CRO opportunities<\/h2>\n

    Evaluate evidence, scope, potential impact, effort, risk, and reversibility. A proven barrier at checkout often outweighs a cosmetic cover change. Punctuation organizes conversation, but it doesn't accurately predict outcomes.<\/p>\n

    Combine obvious fixes with experiments. A broken button can be repaired; it doesn't need A\/B testing. A new proposal, however, may require validation.<\/p>\n

    Privacy and ethics<\/h2>\n

    Collect the necessary information, obtain consent, and protect identifiers. Mask forms and sensitive data in behavioral tools. Review vendors and retention practices.<\/p>\n

    Don't use deceptive patterns to inflate conversions. An involuntary subscription or hidden cancellation inflates the metric and destroys trust.<\/p>\n

    Minimal Dashboard<\/h2>\n

    Display volume, conversion, time, and quality by stage, with unit and cohort. Add technical errors, priority segments, changes, and final result. Don't turn the dashboard into an inventory of all events.<\/p>\n

    Reconcile periodically with transactional data. Discrepancies are documented; they are not corrected by manually adjusting numbers.<\/p>\n

    Funnel and attribution do not answer the same question<\/h2>\n

    The funnel explains what proportion progressed through the stages; attribution attempts to distribute credit among marketing contacts. A drop-in between cart and purchase can be diagnosed without deciding which channel deserves the revenue. Similarly, an attribution model can assign value to a campaign without explaining why many people abandoned the checkout process. Keep both analyses separate and with compatible definitions.<\/p>\n

    For acquisition, retain the captured source, campaign, landing page, and date, but don't let a subsequent visit erase the first observation. In longer cycles, it can be helpful to compare first contact, last contact, and a multichannel view. None of these perspectives alone represents the entire causality; they are guidelines for interpreting the available history.<\/p>\n

    When a sale occurs off-site, create a verifiable link between the digital identifier and the business record, respecting consent and data minimization. If this link only exists for some cases, report coverage. Presenting a sales funnel as complete when it loses half the opportunities upon entering the CRM leads to misleading decisions.<\/p>\n

    Example of reading an ecommerce funnel<\/h2>\n

    Assume a weekly cohort of 20,000 eligible sessions: 8,000 view a product, 1,600 add to cart, 900 initiate checkout, and 540 purchase. The total conversion per session is 2.7%; product to cart, 20%; cart to checkout, 56.25%; and checkout to purchase, 60%. The percentages describe the journey, but they don't yet indicate what needs to change.<\/p>\n

    The team found that checkout conversion rates dropped significantly on mobile devices with a specific payment method. Before redesigning the entire page, they replicated the flow, distinguished between bank rejections and technical errors, and compared versions. If the problem stemmed from a faulty response from the vendor, fixing the integration proved more effective than simply changing the button's color or text.<\/p>\n

    After correcting the issue, compare equivalent cohorts and monitor key indicators: authorization rate, duplicate orders, returns, margin, and support requests. If checkout-to-purchase increases but duplicate charges also rise, the result is unacceptable. The learning must be documented with the date, target population, version, and responsible party.<\/p>\n

    Data governance and changes<\/h2>\n

    Maintain a dictionary with event name, description, trigger, parameters, owner, creation date, and dependencies. Version changes that alter a stage. If the form, consent policy, or user ID is modified, note it to avoid misinterpreting a measurement jump as a business improvement.<\/p>\n

    Configure alerts for absences, duplicates, and unlikely variations, but accompany them with thresholds and business context. A campaign or a seasonal drop can legitimately change the volume. The alert triggers a review; it doesn't replace diagnosis.<\/p>\n

    It regularly brings together analytics, marketing, product, technology, and sales teams to review definitions and decisions. The value of the funnel lies not in producing a fancy graphic, but in maintaining a shared language about where to intervene, which outcome to protect, and how to know if the change worked.<\/p>\n

    Common mistakes<\/h2>\n