From data to decisions

Marketing generates more data than ever. Google Analytics, CRM, email marketing, social media, advertising, SEO. The problem is not a lack of data. It is the excess. So much data that nobody has time to analyze it, and decisions are still made by intuition.

AI changes this equation in 3 ways:

• Automated analysis: AI can process thousands of data points and summarize the key findings in natural language. "Your organic traffic dropped 15% this month, mainly due to position losses on 3 high-volume keywords. The probable cause: a competitor published more updated content on those topics."
• Anomaly detection: AI identifies unusual changes in your metrics before you notice them: a spike in bounce rate, a drop in conversion, a sudden increase in cancellations.
• Prediction: based on historical data, AI can predict future trends: next quarter's revenue, expected traffic, campaign conversion.

The goal of analytics is not to have beautiful dashboards. It is to make better decisions faster.

Marketing dashboards with AI

A marketing dashboard should answer 3 questions: what is happening (current metrics), why it is happening (cause analysis), and what we should do (recommendations). Traditional dashboards only answer the first. Dashboards with AI answer all three.

Content performance dashboard

• Metrics: traffic per article, time on page, bounce rate, conversions (leads generated from each article), social media shares.
• AI analysis: "Articles about [topic X] generate 3x more leads than those about [topic Y]. Articles with comparison tables have 40% less bounce rate. Recommendation: prioritize [topic X] content in comparison format."

Funnel dashboard

• Metrics: visitors, leads, MQLs, SQLs, opportunities, customers. Conversion between each stage.
• AI analysis: "The bottleneck is in MQL-to-SQL conversion (12%, when the industry benchmark is 20%). MQLs who download [resource X] convert to SQL at 25%. Those who only visit the blog convert at 8%. Recommendation: create more [resource X] type content and use it as the main CTA."

Campaign dashboard

• Metrics: cost per lead, cost per acquisition, ROI per campaign, performance by channel.
• AI analysis: "The LinkedIn Ads campaign has a CPL of 45 EUR and a ROAS of 3.2x. The Google Ads campaign has a CPL of 28 EUR but a ROAS of 1.8x. Despite the higher CPL, LinkedIn generates higher quality leads (35% conversion to opportunity vs 12%). Recommendation: increase LinkedIn Ads budget by 20% and reduce Google Ads by 10%."

Creating dashboards with AI

If you do not have a BI tool (Tableau, Looker, Power BI), you can create ad hoc dashboards with AI:

"Here is my marketing data from the last 3 months [paste CSV with metrics]. Generate an executive report with: 1) Resumen of main KPIs (traffic, leads, conversion, revenue) with MoM trend. 2) Top 5 insights (what is working and what is not). 3) 3 detected anomalies (unusual changes in metrics). 4) 3 action recommendations prioritized by impact. Format: structured text with concrete data, not vague opinions."

Cohort analysis

Cohort analysis groups users by a common characteristic (registration date, acquisition channel, first action) and follows their behavior over time. It is the most powerful tool for understanding retention and lifecycle.

Types of cohorts

• Acquisition date cohort: groups customers by the month they registered. Question: do January customers retain the same as March customers? If retention improves, your onboarding improvements are working.
• Channel cohort: groups by acquisition source (organic, paid, referral, direct). Question: do Google Ads customers retain the same as referral customers? If not, the channel influences customer quality.
• First action cohort: groups by what they did upon arrival (downloaded ebook, requested demo, signed up for trial). Question: do those who requested a demo convert and retain more?

Cohort analysis with AI

"Here is data from 1,000 customers with these columns: registration date, acquisition channel, first action, revenue per month (M1 to M12), current status (active/churned). Create a cohort analysis by acquisition channel. For each channel: number of customers, average M1 revenue, retention at M3/M6/M12, average LTV, CAC (if you have it), LTV/CAC ratio. Identify the channel with the best LTV/CAC and the channel with the worst retention. Recommendations."

Cohort analysis reveals truths that aggregate metrics hide. An "average" CAC of 50 EUR can hide that Google Ads CAC is 120 EUR (with an LTV of 80 EUR, losing money) while referral CAC is 15 EUR (with an LTV of 400 EUR).

Attribution models

Attribution answers the question: which channel or marketing action generated this sale? It is the hardest question in marketing because customers interact with multiple channels before buying.

Classic attribution models

• Last click: all credit to the last touchpoint before conversion. Simple but unfair: ignores everything that happened before.
• First click: all credit to the first touchpoint (how they discovered your brand). Ignores the rest of the journey.
• Linear: credit shared equally among all touchpoints. Democratic but does not reflect reality.
• Time decay: more credit to the most recent touchpoints. Assumes what happened last matters most.
• Position-based (U-shaped): 40% to the first contact, 40% to the last, 20% shared among intermediate ones. The most used in B2B.

Data-driven attribution

Data-driven attribution models use machine learning to assign credit based on the real impact of each touchpoint on conversion. Google Analytics 4 includes this model if you have enough data (at least 300 conversions in 30 days).

For companies with less volume, AI can help with ad hoc analysis: "Here is the customer journey of 200 converted customers [paste touchpoint sequence per customer: channel, action, date]. Analyze: which channels appear most frequently in journeys that convert? Which channel combinations have the highest conversion rate? Which channel is most effective as a first contact vs as a last contact? Suggest a customized attribution model for our business."

The truth about attribution

No attribution model is perfect. Models measure clicks and visits, but they do not measure the LinkedIn post your prospect read 3 weeks ago that made them think of you when the need arose. Nor the colleague's recommendation at lunch. Nor the podcast where they heard you speak.

Attribution is a useful approximation, not an absolute truth. Use it to prioritize budget across channels, not for binary decisions like "this channel doesn't work, let's cut it".

Forecasting with AI

Forecasting predicts the future based on historical data. In marketing, forecasting helps plan budget, set realistic goals and anticipate problems.

What you can forecast

• Traffic: expected organic traffic next quarter based on current trend, seasonality and keyword rankings.
• Leads: expected number of leads based on projected traffic and historical conversion rate.
• Revenue: expected revenue based on current pipeline, close velocity and historical conversion.
• Churn: number of customers who will likely cancel based on engagement patterns.
• CAC: expected acquisition cost if you maintain current budget and efficiency.

Forecasting with AI

For basic forecasting, export historical data and pass it to AI:

"Here is monthly data from the last 24 months: traffic, leads, MQLs, SQLs, customers, revenue, marketing spend [paste CSV]. Generate a forecast for the next 6 months with 3 scenarios: conservative (current trend without changes), base (with the planned improvements I describe below), optimistic (+20% over base). For each scenario: traffic, leads, customers and revenue per month. Include the confidence range. Visualize with a table."

For advanced forecasting, tools like HubSpot (AI Forecast), Salesforce (Einstein Forecast) or dedicated tools like Clari predict revenue with machine learning models trained on your pipeline data.

Seasonality and events

Forecasting must consider external factors that affect your metrics:

• Seasonality: if your business has peaks (Black Friday, end of fiscal quarter, back from holidays), the forecast must reflect them.
• Planned events: a product launch, a major campaign or an industry event will generate traffic and lead spikes that do not naturally repeat.
• External factors: regulatory changes (a new regulation can boost compliance demand), competitor moves, market trends.

Automated executive reporting

Executive reporting is the bridge between marketing data and C-level decisions. A CEO does not want to see 47 charts. They want to know: are we growing? Where should we invest more? What risks are there?

Executive report structure

1. Executive summary (5 lines): general status, trend, the most important data point of the period.
2. Main KPIs (table): revenue, leads, new customers, CAC, LTV, pipeline. Actual vs target vs prior period.
3. Insights (3-5 points): what worked, what did not, why. With concrete data, not opinions.
4. Risks (1-3 points): negative trends, dependencies, potential problems.
5. Recommendations (3 points): concrete actions prioritized by impact. "Increase [channel X] budget by 15% based on 3.2x ROAS."
6. Forecast: prediction for the next period with 3 scenarios.

Generating reporting with AI

"Generate a marketing executive report for the CEO. Month's data [paste KPIs]. Format: 5-line executive summary, KPI table (actual vs target vs prior month), 3 main insights with data, 2 risks, 3 recommendations. Tone: direct, no filler, with concrete numbers. The CEO has 2 minutes to read it. Every sentence must add value."

Automating reporting

The ideal flow: data is automatically exported from your tools (GA4, CRM, email platform), AI analyzes it and generates the report, and you review and send it. Tools:

• Databox / Klipfolio: automated dashboards that pull data from multiple sources.
• Supermetrics: exports data from 100+ sources to Google Sheets, Excel or Looker Studio.
• Zapier + ChatGPT: automates data extraction and AI report generation.
• HubSpot / Salesforce: native reports with included AI analysis.

Analytics tools with AI

• Google Analytics 4: free, mandatory. AI for anomaly detection, automatic insights and data-driven attribution.
• PostHog: open source product analytics. Funnels, cohorts, feature flags, session recordings. Free up to 1M events/month.
• Amplitude: advanced product analytics. Cohorts, paths, predictions. For companies with digital products.
• Mixpanel: similar to Amplitude, with AI for insights and anomalies. Generous free plan.
• Looker Studio (Google): free dashboards connected with Google Ads, GA4, Sheets. Limited but sufficient to start.
• Tableau / Power BI: enterprise BI tools with integrated AI (Tableau Pulse, Power BI Copilot). For companies with advanced needs.
• Claude / ChatGPT: for ad hoc analysis. Export data as CSV, pass it to AI and get insights in minutes.

Metrics the CEO cares about

Marketers talk about impressions, engagement rate and CTR. CEOs talk about revenue, margin and growth. The CMO who speaks the CEO's language wins budget. The one who speaks marketing's language loses relevance.

Metrics the CEO wants to see

• CAC (Customer Acquisition Cost): how much it costs to acquire a customer. Total marketing and sales spend / new customers.
• LTV (Lifetime Value): how much a customer earns over their relationship with you. Average revenue per customer x average relationship duration.
• LTV/CAC ratio: if it is greater than 3, your unit economics are healthy. If it is less than 1, you lose money with every customer.
• Payback period: how many months it takes for a customer to pay back the acquisition cost.
• Pipeline-to-revenue ratio: how much pipeline you need to generate 1 EUR of revenue. If you need 4 EUR of pipeline per 1 EUR of revenue (4:1 ratio), you need 4M of pipeline to make 1M of revenue.
• Marketing-sourced revenue: percentage of total revenue that originated from marketing activities (not direct sales or existing inbound).

Errores comunes

1. Vanity metrics. Reporting impressions and followers to the CEO is irrelevant if you do not connect them with revenue. Always link marketing metrics with business impact.
2. Dashboards without action. A dashboard that shows data but does not suggest actions is decoration. Every data point should lead to a decision.
3. Attribution as absolute truth. No attribution model is perfect. Use it as a guide, not as law. Always complement with qualitative data.
4. Forecasting without humility. Models predict trends, not unforeseen events. Always present a range (conservative/base/optimistic) instead of a single number.
5. Analysis paralysis. Making a decision with 80% of the data is better than waiting for 100% and acting too late. AI accelerates analysis, but the decision remains human.

Ejercicio practico

Puntos clave