A Copilot Studio agent guides employees through a five-step benefits enrollment process. The product team has configured custom events in Application Insights for each step: 'EnrollmentStarted', 'PlanSelected', 'DependentsAdded', 'ReviewCompleted', and 'EnrollmentSubmitted'. After one month of data, the product team wants to identify where most users abandon the process. Using Application Insights, which analysis method and KQL query approach would directly reveal the step-by-step drop-off rates between each enrollment stage?
Select an answer to reveal the explanation.
Short Explanation
Funnel analysis needs unique sessions at each gate. A KQL query that counts distinct sessions per event and calculates survival between consecutive steps gives you the drop-off at every stage. The Funnels UI can show it visually, but the KQL method proves you know what's being computed.
Full Explanation
Funnel analysis in Application Insights identifies where users abandon a multi-step process by tracking the percentage of a starting cohort that completes each subsequent step. For the AB-620 exam, you need to understand both the KQL approach and the Application Insights Funnels feature.
The correct approach is the KQL query against the customEvents table that counts unique sessions per event name, calculates the conversion percentage between consecutive steps to build a funnel table, and optionally visualizes the result using the Application Insights Funnels feature for a guided UI alternative. This is correct because it describes both the analytical mechanism and the optional UI alternative. The KQL approach queries the customEvents table, groups by event name, counts distinct session_Id values for each event, and calculates the conversion rate between each consecutive step as (count_step_N / count_step_1) * 100. This produces a funnel table showing exactly which step has the greatest percentage drop-off. Example KQL:
`` customEvents | where name in ('EnrollmentStarted', 'PlanSelected', 'DependentsAdded', 'ReviewCompleted', 'EnrollmentSubmitted') | summarize Users = dcount(session_Id) by name | order by Users desc ``
The Funnels feature in Application Insights provides a GUI that performs this analysis without requiring KQL, but the correct approach acknowledges both approaches.
The Metrics Explorer approach is wrong because it shows event counts as time-series data. This is useful for trend analysis, such as enrollment volume over time, but it does not calculate conversion rates between steps. Comparing absolute counts across events does not account for users who started at different times or who may complete steps in a single session versus across multiple sessions. Absolute counts do not equal funnel conversion rates.
The User Flows approach is wrong because the built-in User Flows visualization is designed for web and app page navigation analysis. It shows the paths users take through page views and events in a Sankey diagram. While User Flows can visualize event sequences, it is not designed for sequential funnel analysis with percentage drop-off calculation between ordered steps. The Funnels feature is the purpose-built tool for that job.
The Funnels-only approach is wrong because it describes the Application Insights Funnels feature correctly but presents it as the complete answer without acknowledging the underlying KQL mechanism. While Funnels is a valid UI approach, the KQL query approach with optional Funnels visualization is more complete because it demonstrates understanding of both the KQL computation and the Funnels feature as a visual alternative. On an exam, demonstrating the deeper analytical understanding is always preferred.
Exam tip: For Application Insights funnel analysis on AB-620, know these key KQL functions: dcount() for counting distinct sessions/users, summarize for aggregation, and order by for ordering results. Know that the Funnels feature requires events to be configured as ordered steps, and that it tracks users who complete all steps in sequence. The key metric is conversion rate between steps, not absolute counts, and the goal is identifying the step with the highest drop-off percentage.