Supported events

The funnel uses supported events such as widget_open, quiz_started, question_answered, results_opened, recommendation_shown, product_clicked, and add_to_cart. An event is an observation of the embedded experience. It is not a completed payment. Browser blocking, consent choices, interrupted requests, or unsupported integrations can make event coverage incomplete.

How an attempt is formed

Events are grouped using the run and session identities and ordered by valid timestamps. Records without a usable identity cannot be safely assigned to a journey and are counted as excluded. The reviewed aggregation splits attempts after thirty minutes of inactivity and starts a new journey when a new quiz start occurs after an already started attempt. This is an analytics session rule, not a way to recognise the same person across devices.

Explicit and inferred starts

A recorded quiz_started event is the strongest direct start signal. Later events such as an answered question or opened result can imply that a journey started even when the initial event is missing. Explicit starts and inferred starts should remain distinguishable. A widget being opened is not automatically equivalent to a shopper answering the quiz.

Revising answers does not invent stages

If the shopper goes back and changes the answer to the same question, the aggregation updates that question's recorded choice within the attempt rather than adding an extra question stage. Question and answer labels identify choices in the reviewed aggregation, so wording changes can split historical labels. Keep a stable flow when comparing paths, or explicitly account for changed questions.

Choice-level statistics

A choice can show how many attempts selected it, continued to another step or a result, completed, remained active, or exited at that choice. Completed attempts can contribute to several earlier choices, so summing completion counts across all choices double-counts journeys. Path counts describe the retained sequence of choices; they are not a set of independent shoppers for each row.

Completion, click, and exit

Opening results, showing recommendations, clicking a product, or a supported add-to-cart event establishes completion of the recommendation journey in the reviewed funnel. Product clicks are also tracked separately. An unfinished attempt becomes exited after the inactivity threshold; before that it is active. This does not observe a browser close in real time or prove why the shopper stopped.

Example
Synthetic journey cohort
100 started attempts
 60 reached a recommendation result
 25 clicked at least one product
 30 unfinished attempts inactive for >= 30 minutes
 10 unfinished attempts still active

Completion: 60 / 100 = 60%
Product click: 25 / 100 = 25%
Exited + active + completed = 100
No order conversion can be inferred from these events alone.

How to investigate a drop

Compare choices within the same question and time range. A popular choice followed by many inactive attempts can indicate confusing wording, too few matching products, slow rendering, or genuine lack of interest. Inspect the matching rules and live flow before attributing a cause. Make one documented change and collect a compatible follow-up sample; the funnel is observational rather than a randomised experiment.

How Product Advisor questions and recommendations workInstall, update, or remove a Product AdvisorCockpit periods, Search Console, and connector freshness