// Case study / Funnel instrumentation
A 50%+ candidate drop-off, decomposed step by step.
Front-end behavioural instrumentation turned a vague candidate-sourcing drop-off into a step-level funnel question, replacing manual Excel reporting with a GA4-to-BigQuery Looker Studio dashboard.
What happened
The sourcing team moved from narrative to step-level evidence.
A global AI services firm's sourcing team knew there was a 50%+ drop-off between candidate selection and start of work, but could not tell where the loss happened. Full GA4 instrumentation of the Candidate Management Platform and a Looker Studio dashboard on the GA4-to-BigQuery pipeline decomposed the vague problem into a step-level view.
The example outcome was the conversation shifting from "we're losing candidates somewhere" to "step 4 has a 43% abandonment rate; let's look at what's happening there."
Context
The team had a drop-off problem, not a location.
Metrics lived in Excel spreadsheets hand-populated from the operational system. No funnel decomposition was possible, and any conversation about fixing the drop-off was stuck at the qualitative level. The engagement had previously evaluated Mixpanel as the analytics platform, but the client chose to remain on GA4 for cost and infrastructure-compatibility reasons — a decision that contributed to low-friction adoption of the resulting dashboards.
Task
Instrument the candidate journey without over-prescribing the finding.
The goal was to make the journey measurable across application start, exam steps, compliance checkpoints, and selection status. The stakeholder framing mattered: this was presented as new data available for sourcing decisions, not as a premature conclusion to react to.
I also built and presented the analytics roadmap for the engagement, establishing measurement priorities across the Candidate Sourcing Platform, Refer a Friend, and My Performance Dashboard product areas.
Application startexplicit denominator
Exam stepswhere candidates slow or stop
Compliance checkpointsfriction vs intent
Selection statusper-candidate, per-programme
Captured the entry point into the candidate-management flow so the denominator was explicit.
Instrumented exam-stage behaviour to locate where candidates slowed, stopped, or abandoned.
Tracked required checkpoints so operational friction could be separated from candidate intent.
Connected funnel progress to selection state, enabling per-candidate and per-programme segmentation.
Action
The instrumentation replaced a manual reporting workaround.
I led front-end behavioural instrumentation of the Candidate Management Platform with two web engineers. GA4 event capture covered application start, exam steps, compliance checkpoints, and selection status, with user-identity parameters enabling per-candidate and per-programme segmentation.
I then built a Looker Studio dashboard on the GA4-to-BigQuery pipeline that replaced the manual Excel workflow with a queryable view of candidate flow.
Outcome
A qualitative concern became a measurable funnel.
- 50%+ candidate drop-off decomposed into step-level abandonment analysis.
- Step-four abandonment example surfaced as a concrete investigation point.
- Manual Excel workflow replaced with a Looker Studio dashboard on the GA4-to-BigQuery pipeline.
- Stakeholder sync in May 2025 framed the work as new data available for sourcing decisions.
Design lesson
Measurement can make an organisational problem actionable without pretending to solve it alone.
The instrumentation did not explain why candidates left at step four. It made the abandonment visible enough that the sourcing team could investigate the right moment instead of debating the whole journey at once.