{"id":1278,"date":"2026-09-16T08:00:00","date_gmt":"2026-09-16T13:00:00","guid":{"rendered":"https:\/\/www.seomos.com\/?p=1278"},"modified":"2026-09-17T16:08:07","modified_gmt":"2026-09-17T21:08:07","slug":"que-es-la-analitica-web","status":"publish","type":"post","link":"https:\/\/www.seomos.com\/en\/blog\/que-es-la-analitica-web\/","title":{"rendered":"What to Measure in Web Analytics to Make Better Decisions"},"content":{"rendered":"
\n

Web analytics is the discipline of collecting, validating, analyzing, and communicating data about interaction with websites and applications to make decisions.<\/strong> It's not about installing Google Analytics or accumulating dashboards. A useful system starts with business questions, defines events and metrics, respects consent, verifies implementation, and connects digital behavior with outcomes such as sales, qualified leads, usage, or retention.<\/p>\n<\/div>\n

A tool can record millions of events and still leave the team without answers. If "conversion" is lumped together to include a click, a form submission, and a purchase, the volume may seem high, but the decision-making process suffers. Quality begins with the definitions.<\/p>\n

What is web analytics?<\/h2>\n

Analysis involves transforming observations into a limited explanation and action. The discipline covers acquisition, behavior, content, experience, conversion, and value. It also acknowledges what cannot be observed due to privacy, devices, attribution, or errors.<\/p>\n

A mature measurement combines quantitative data \u2014events, sessions, revenue\u2014 with qualitative research \u2014interviews, tests, feedback\u2014 and operational data from CRM, ecommerce or product.<\/p>\n

Measurement, reporting and analysis<\/h2>\n

Measurement<\/strong> defines and captures signals. Reporting<\/strong> organizes recurring indicators. Analysis<\/strong> It investigates a question, tests hypotheses, and recommends a decision. A weekly dashboard does not replace causal analysis.<\/p>\n

Analytics is not individual surveillance. Its purpose must be legitimate, proportionate, and explained. Aggregating data does not automatically eliminate privacy risks.<\/p>\n

How to create a measurement plan<\/h2>\n
    \n
  1. Aim:<\/strong> express what result the business needs.<\/li>\n
  2. Decision:<\/strong> Define what the team would change with the information.<\/li>\n
  3. Ask:<\/strong> formulate what should be investigated.<\/li>\n
  4. Indicator:<\/strong> Select the measurement and its formula.<\/li>\n
  5. Event:<\/strong> Describe the observable action and exact moment.<\/li>\n
  6. Parameters:<\/strong> It adds necessary context, without unnecessary personal data.<\/li>\n
  7. Fountain:<\/strong> Identify site, CRM, gateway, or survey.<\/li>\n
  8. QA:<\/strong> establishes proof, tolerance, and responsibility.<\/li>\n
  9. Cadence:<\/strong> decide when to review and act.<\/li>\n<\/ol>\n

    Example: The goal is to increase qualified requests. The decision is to prioritize pages. The event generate_lead<\/code> This occurs after a successful server response, not upon clicking. The CRM then determines whether the lead is qualified.<\/p>\n

    \"Marco
    A traceable chain avoids measuring actions that no one will use.<\/figcaption><\/figure>\n

    Events, parameters, dimensions, and metrics<\/h2>\n

    In GA4, an event measures an interaction or occurrence. There are automatically collected events, enhanced measurement events, recommended events, and custom events. Recommended events allow you to leverage reports when implemented with your structure.<\/p>\n

    Parameters add context: method, search location, currency, or items. They can then feed dimensions and metrics. Don't create a separate event for each button if a controlled parameter can distinguish it.<\/p>\n

    An event contract must include name, condition, parameters, types, example, owner, consent, and exclusions. Use stable names: renaming without migration splits series.<\/p>\n

    Metrics and KPIs<\/h2>\n

    A metric quantifies; a KPI is critical to an objective and has a target, timeframe, responsible party, and decision rule. Users, sessions, and clicks can be diagnostic, but they are rarely business KPIs on their own.<\/p>\n

    \n\n\n\n\n\n\n\n\n\n
    From system to value<\/caption>\n
    Level<\/th>\nExamples<\/th>\nDecision<\/th>\n<\/tr>\n<\/thead>\n
    Availability<\/th>\nErrors, latency, active tags<\/td>\nCorrect capture<\/td>\n<\/tr>\n
    Acquisition<\/th>\nSource, campaign, cost<\/td>\nAllocate investment<\/td>\n<\/tr>\n
    Behavior<\/th>\nView, search, advance<\/td>\nImprove route<\/td>\n<\/tr>\n
    Result<\/th>\nLead, purchase, use<\/td>\nPrioritize offer<\/td>\n<\/tr>\n
    Worth<\/th>\nMargin, quality, retention<\/td>\nEscalate or stop<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

    Acquisition and attribution<\/h2>\n

    UTM parameters, referrer identifiers, advertising identifiers, and channel rules help classify arrival. Maintain a consistent convention of UTM parameters<\/a>, It preserves values and avoids labeling internal links.<\/p>\n

    Attribution assigns credit under a model; it does not demonstrate causality. Windows, consent, devices, and offline channels produce gaps. Compare reports with the same configuration and highlight model changes.<\/p>\n

    Ecommerce and lead generation<\/h2>\n

    GA4 recommends an ecommerce sequence such as list view, product, cart, checkout, payment, purchase, and refund. Currency is required when sending value; items require a consistent structure. Purchases should use a transactional identifier to control duplicates.<\/p>\n

    In leads, it measures successful submissions, connected calls, or confirmed appointments. Then it links CRM statuses: contact, qualification, opportunity, and close. A form does not equal a customer.<\/p>\n

    \"Mapa
    The journey needs a common definition between web and results systems.<\/figcaption><\/figure>\n

    How to validate an implementation<\/h2>\n
      \n
    1. Build test cases that are accepted, rejected, and have errors.<\/li>\n
    2. Use DebugView or Realtime to confirm name and parameters.<\/li>\n
    3. Verify that the event occurs once.<\/li>\n
    4. Check currency, value, transaction ID, and items.<\/li>\n
    5. Test domains, gateways, mobile devices, and consent.<\/li>\n
    6. Compare counts with backend or CRM.<\/li>\n
    7. Wait for processing before concluding that a report is missing.<\/li>\n
    8. Document evidence and deployed version.<\/li>\n<\/ol>\n

      Tag Assistant confirms triggers, not the accuracy of the transaction. A visible event may contain zero value, incorrect currency, or duplicates.<\/p>\n

      Privacy and data quality<\/h2>\n

      Define what information you collect, why, for how long, and who has access. Do not send personally identifiable information to Analytics. Configure consent before tagging and respect opt-outs.<\/p>\n

      Google notes that Analytics may use identifiers and that behavior changes when storage is denied. Some results may be modeled; distinguish between observed and modeled data where relevant.<\/p>\n

      Quality includes completeness, validity, consistency, uniqueness, and timeliness. Create alerts for drops, spikes, empty parameters, and domain changes.<\/p>\n

      Process for converting data into a decision<\/h2>\n
        \n
      1. Write the question and the time period before opening the report.<\/li>\n
      2. Segment by page, source, country, device, or cohort.<\/li>\n
      3. Check instrumentation and business changes.<\/li>\n
      4. Describe the pattern without attributing a cause.<\/li>\n
      5. Formulate alternative explanations.<\/li>\n
      6. Look for qualitative or operational evidence.<\/li>\n
      7. Propose a reversible action and a success criterion.<\/li>\n
      8. Review the results and record what you've learned.<\/li>\n<\/ol>\n

        Don't attribute a conversion improvement to the new button if traffic, price, and inventory also changed. Design experiments or acknowledge the uncertainty.<\/p>\n

        Segmentation: the average rarely represents everyone<\/h2>\n

        An overall rate may increase because the traffic mix changed, even if no segment improved. First, separate the dimensions that affect the experience: landing page, source, campaign, country, device, new or returning customer, and site version.<\/p>\n

        Segmenting doesn't mean cross-referencing everything. Each division reduces the sample size and increases noise. Start with a hypothesis, maintain a comparison group, and avoid publishing conclusions about small cohorts. Record filters and time zone so someone else can reproduce the calculation.<\/p>\n

        Cohorts group people based on a common starting point, such as their first purchase or account opening. They are useful for retention and value tracking. Do not mix cohorts acquired through different promotions without disclosing this. In B2B, the long cycle requires linking arrival date, opportunity, and closing date without exposing personal data on the web tool.<\/p>\n

        What should a useful dashboard contain?<\/h2>\n

        An executive dashboard needs few KPIs, a target, comparisons, explanations, and an owner. An operational dashboard can display events, errors, and segments. Separating them reduces the temptation to treat every number as a target.<\/p>\n

        It includes definitions alongside the report: what a session counts, what event is key, what currency is used, and when it's updated. It marks instrumentation outages, campaigns, migrations, and incidents. Without annotations, a technical discontinuity appears as a change in behavior.<\/p>\n

        Use alerts with context. A drop in sales could be due to demand, site, payment gateway, or label. The alert initiates a diagnostic process; it doesn't assign a cause. Design a troubleshooting path from backend and payments to browser and platform.<\/p>\n

        Experiments and causality<\/h2>\n

        A before\/after comparison is subject to seasonality, campaigns, inventory, and learning. When risk and traffic allow, randomly assign variants, define the primary metric, duration, and rules before looking at the results.<\/p>\n

        Don't stop testing after the first positive result. Review exposure, allocation errors, novelty, multiple metrics, and impact by segment. A statistical improvement that reduces margin or accessibility is not a business success.<\/p>\n

        When experimentation is not possible, use convergent evidence: time series, comparable markets, stepwise changes, and interviews. State the degree of confidence. Responsible analytics distinguishes between \u201coccurred after\u201d and \u201coccurred because of.\u201d.<\/p>\n

        Governance and data dictionary<\/h2>\n

        Create a dictionary with event, description, parameters, system, owner, sensitivity, and date. Remove unused metrics and prevent each team from inventing names. Control who publishes tags and require review.<\/p>\n

        Define development, staging, and production environments with separate branding. Internal traffic and testing can contaminate live data; identify these sources without creating irreversible filters without backups.<\/p>\n

        Perform a review after each relevant release and another periodic review. Check for orphaned tags, events without parameters, obsolete goals, and access for users who are no longer involved. Maintenance is part of the system.<\/p>\n

        Hypothetical example<\/h2>\n

        Hypothetical example.<\/strong> A website reports 400 leads per month. QA discovers that clicking "Submit" counts even if the server rejects the form, and that double-clicking generates two events. Actual submissions are 250.<\/p>\n

        The event is moved to the successful response stage, a request ID is added, and the CRM is connected. Of 250 leads, 90 are qualified and 18 are opportunities. Marketing stops optimizing by total volume and prioritizes pages with the highest opportunity rate.<\/p>\n

        The report drops from 400 to 250 after the correction. This is not a commercial loss; it's a measurement improvement that should be noted to avoid comparing series as equivalent.<\/p>\n

        \"Equipo
        Shared definitions reduce conflicts between platform, marketing, and sales.<\/figcaption><\/figure>\n

        Common mistakes<\/h2>\n