What this checkpoint covers
Brookhurst Concert Tickets is an invented seller of tickets for small and mid-sized venues. Its Google Analytics property (GA4Google's event-based analytics platform.) exports to BigQuery, two marketers run Tag Manager (GTMTag management system.), and on-sale days bring sharp traffic spikes. The questions draw on the BigQuery export basics guide, the diagnosis of an export that does not match the interface and the client-side versus server-side tagging comparison.
Task
Answer each question in a sentence or two with your reason, then compare with the explanations. A right action with the wrong reason is a miss. Reread the linked page for each miss.
Questions
- An analyst extracts the ticket quantity parameter from purchase events and gets nothing but nulls, although the interface shows the parameter populated. What is the likely mistake?
- In a market where many visitors decline consent, the interface shows noticeably more purchases than the export. A small venue's row appears in the export but not in the interface. Which of these is an error?
- Daily ticket sales from the export disagree with the interface on most days, yet the weekly totals agree. A separate query reports more sessions than the interface by a wide margin. What do you check in each?
- The box office wants to watch sales minute by minute on on-sale days, and finance wants closed weekly figures. How do daily and streaming export serve each, and where does the cost really come from?
- An agency proposes moving to server-side tagging so that conversions from visitors who declined advertising consent can still reach the ad platforms. What do you say?
- Brookhurst has no engineer, a tidy container, and ad platforms that under-report purchases. Does it need a server container now?
- A vendor pitches a customer data platform (CDPUnified first-party customer profiles.) as a replacement for a server container. Are they the same thing?
Answers explained
- The wrong value slot, or no unnesting. Parameters are stored as a repeated key and value record, and only one typed slot holds each value. A quantity sent as a number sits in the integer slot, so reading the string slot returns nulls. Extract it with a sub-select over the unnested parameters, filtered to the key, from the right slot. See the schema.
- Neither. The interface blends modelled purchases where consent was declined; the export holds collected events only. Thresholds hide small rows in the interface but not in the export. Both are expected; label figures as observed or modelled and use the export for small segments. See modelled data on one side only.
- Dates in the first, session logic in the second. Daily tables follow the property's time zone while timestamps are UTC, so dates cut from timestamps move evening sales across midnight. A session is the user identifier and the session identifier together, counted once over the range; counting the session identifier per day and adding days overcounts. See dates cut in another time zone.
- Streaming for monitoring, daily for reporting. Intraday tables are provisional and disappear when the daily table lands, so they suit on-sale monitoring, not closed figures. Most cost comes from bytes scanned by queries: a monitoring dashboard on raw events scans them at every refresh. Summarise first, select columns, bound every query by date and set a limit on bytes billed; check Google's pricing page for current charges. See daily versus streaming.
- No. Server-side tagging changes where code runs, not what the business may collect. It makes central consent enforcement easier; it never replaces a lawful basis. A server that forwards data for people who declined is a compliance failure. See the tagging comparison.
- Probably not yet. A server container needs someone to own hosting, monitoring and debugging; unowned, it puts all measurement at risk. Check consent behaviour and the purchase trigger first, and use the platforms' documented server-to-server routes from the ticketing system later if the gap stays material. A hybrid is the usual end state for a team that grows into it. See the decision table.
- No. A server container routes and shapes collected events on their way to analytics and ad platforms. A customer data platform resolves identity, holds profiles and pushes segments to channels. They can sit in the same stack, but one cannot stand in for the other. The guide to data flow and integrations places each in the flow.
Where to go next
If the reconciliation questions caught you out, redo the BigQuery reconciliation exercise for a soap maker. Then move to causal measurement with the geo holdout design exercise. To revisit event design and data quality, return to the intermediate analytics level.
