The Analysis Process: Four Analysis Types, Data Collection, a Worked Chart Diagnosis, Ad Cannibalisation, and the Attribution Problem

Descriptive = what happened (dashboards, trends, seasonality). Diagnostic = why (patterns, causal relationships, root-cause analysis). Predictive = what might happen (project trend lines; plan pricing and inventory). Prescriptive = what you should do (scenarios that recommend discount and stock levels). Reliable prescriptive tools are rare, so analysts who diagnose, project and then recommend add the most value. Collection methods: transactional tracking; surveys (avoid leading questions and socially acceptable answers); online behaviour tracking (heat maps, page sequences, mind compliance); interviews and focus groups for cultural insight.

Revenue up 1.4x September to October (descriptive); traffic up 1.26x with better conversion (diagnostic); but last year the same traffic jump gave 1.7x, so this year's conversion capture was weaker - discounts, delivery timelines or campaigns to investigate. Marketplace case: mandate is ROAS (return on ad spend) around 8; ads lifted ad revenue and total revenue, yet overall ROAS fell from 45 to 15.1 because organic revenue shrank - the ads cannibalised sales that would have happened anyway. Always see the complete picture and pick the right baseline (last year, the previous sale event).

Credit is hard to assign when several factors move together: website redesign vs the October sale and Diwali; email open rate credited to subject lines when send time, weekday and cohort engagement matter; calendars and gym memberships both spike in January. Mitigations: carefully controlled A/B tests, list all factors, accept that full attribution is impossible. Hand-off: the 'mega' Excel case - play category head, 11 questions, answer in Excel.