Data and Web Analytics: Measurement Plans, GA4 and Attribution
Analytics questions test whether you can turn a business question into a measurement plan, read GA4-style data correctly, and interpret attribution without overclaiming. The exam rewards candidates who know the limits of their data as well as its numbers.
Measurement planning
Work backwards: business objective → KPI → supporting metrics → events and dimensions → tagging. GA4 is event-based: every interaction is an event with parameters, and conversions are designated key events. Engagement rate and engaged sessions replace some older session-centric reporting. Document a measurement plan naming each event, its trigger and its owner, or implementations drift within months.
Dimensions and metrics
Dimensions describe - source/medium, campaign, device, page, country. Metrics quantify - users, sessions, event count, engagement rate, revenue. Mixing them up is the most common reporting error. UTM parameters (source, medium, campaign, term, content) standardise campaign tracking; inconsistent casing or missing mediums create split, duplicated channels.
Data quality and attribution
Exclude internal traffic, apply consent mode and expect modelled gaps, and note sampling or thresholding in reports. Attribution models answer different questions: last click credits the closer, first click credits discovery, linear spreads evenly, position-based gives most credit to first and last contact, time decay favours recent touchpoints, and data-driven allocates by observed contribution. Last-click systematically under-credits upper-funnel work.
Worked example
A £100 purchase follows: paid search → organic blog → email → direct. Last click gives email 100%; first click gives paid search 100%; linear gives 25% to each of the four. Position-based gives 40% each to paid search and email, and 20% split between the blog and direct. Reasoning: none is objectively correct. If the team is deciding whether to fund organic content, the linear or position-based view reveals a contribution that last click hides; if the team is optimising the closing email, last click is defensible. State the decision the model must inform before choosing it.
Traps
- Reporting sessions when the question asks for users.
- Duplicated or mis-cased UTM values splitting one campaign into several.
- Claiming causation from a dashboard correlation without a controlled test.
- Comparing year-on-year figures without acknowledging consent-mode data gaps.
- Choosing an attribution model after seeing which one flatters the channel.