US university surfaces 85% registration drop-off

The client is a US university's professional-education division. Prospective learners read about courses on a marketing site hosted on one domain, then register through a Salesforce portal on another. The team ran a Looker Studio registration funnel report built on GA4 data in BigQuery. It showed near-zero at every step. GA4 itself was recording dozens of 'Register now' clicks. Leadership could not tell where applicants were dropping out, or whether marketing spend was reaching registrations at all.

• The two domains were not linked. 3,170 portal users had never been seen on the marketing site, against 191 who had. Most registrants looked like new visitors with no source.
• The report was filtering on a course code that did not exist anywhere in the data, so every scorecard returned zero.
• Test, sandbox, and UAT portal hostnames were mixed into the live numbers.
• Most portal sessions carried no course code at all.
• A report that shows zero gets ignored. The 85% drop-off it was hiding kept running unchallenged.
• Leadership could not tell whether to fix the ads, the portal, or the content, because the data pointed at nothing.

• Replace the zeros with a step-by-step funnel showing real drop-off, by course.
• Rebuild the pipeline on the raw GA4 export in BigQuery, not the summary feeding the broken report.
• Clean test, sandbox, and UAT noise out of the live figures.
• Make the funnel queryable from one table, so new questions do not become new projects.
• Scope the next fix (cross-domain linking) so source attribution follows registrants.
• Give leadership a defensible answer to "where do people actually drop off?"

• Pulled the funnel from the raw GA4 export in BigQuery, bypassing the summary the broken report was using.
• Built one flat table with a row per session: course code, source, campaign, and a 0/1 flag for each of 8 funnel steps (login, sign-up, course selection, background, experience, information, consent, confirmation).
• Mapped the portal's real event names to the 8 funnel steps.
• Removed test, sandbox, and UAT hostnames from the live numbers.
• Replaced the broken course-code filter with codes that actually exist in the data.
• Pointed the Looker Studio report at the new table, by course.
• Scoped the cross-domain linking fix so every registration keeps its source.
• Delivered across three weeks: diagnosis (filter, domains, environments), flat table built and checked against GA4, report rebuilt by course.

• 0 to 351 confirmation sessions now visible to the report, where the old filter returned zero.
• 85% drop-off between login and course selection, previously hidden, now the team's top priority.
• 5 of 6 confirmed registrations come from returning learners, a retention opportunity that was not visible before.
• 3,314 users reached login, 497 reached sign-up, 250 reached course selection, 208 reached background. One clear leak, one page after login.
• One queryable BigQuery table carries the funnel by course, source, and month. New questions are a query, not a new project.
• Cross-domain source attribution scoped as the next fix. Currently 191 of 3,361 portal users seen on both domains.