Funnels
Last updated
Last updated
You can build funnels in the Hypertune UI to:
See drop-off rates between different event types
Compare conversion rates across different arms of a split
A funnel is composed of a sequence of steps.
Each step can be an:
Event, e.g. SignUpEvent
Split exposure, e.g. New Editor A/B Test or Landing Page Machine Learning Loop
For example, you can build a funnel with the following steps:
Event: PageViewEvent
Event: SignUpEvent
Event: UpgradeEvent
This lets you see drop-off rates between each step of your onboarding funnel.
When setting up each funnel step, you'll need to define a unit ID. For split exposures, this is automatically set to the unit ID used for randomization. However, for events, you'll need to configure it manually. Ensure that unit IDs are consistent across steps to build the funnel correctly.
Typically, the unit ID corresponds to a user ID or an anonymous ID stored in a cookie. It could also be a team ID or organization ID, as long as it's present in all event types in the funnel.
The first step of your funnel captures all relevant events within a specified timeframe. You'll see the unique count, which reflects the number of unique units (users, teams, etc.) based on your configured unit ID.
Subsequent steps will only include data for units that appeared in the previous step and only for events occurring after the earliest corresponding event in the previous step. This ensures a clear, sequential view of user journeys through the funnel.
For example, if the first step includes events of type A and the second step includes events of type B, and an event A was logged for a user at 10:00 AM and 11:00 AM, while event B was logged at 9:30 AM, 10:30 AM, and 11:30 AM, only the 10:30 AM and 11:30 AM occurrences of event B will be included in the second step.
You can add filters to each step of your funnel to only select the data you're interested in.
For example, if your app runs in multiple environments it can be useful to only include data from your production environment to avoid polluting your funnel results with test data. While disabling data logging in test environments is an option, it could hinder your ability to verify functionality before a production release.
If your unit IDs are environment-specific, filtering the first step should suffice.
Each step can be broken down by payload fields attached to the event or split exposure.
For example, if you included a utm_source field in the PageViewEvent type, you can add this as a breakdown to the first step in the example above, to get a segment for each value of utm_source.
Breakdowns are carried through to all the following steps in the funnel so you can see drop-off rates for each segment.
Split steps can also be broken down by their dimensions, to get a segment for each arm of the dimension.
For example, you can build a funnel with the following steps:
Split Exposure: New Landing Page A/B Test
Event: SignUpEvent
Event: UpgradeEvent
You can break down the first step by the dimension of the A/B test to compare the conversion rate, in terms of sign ups and upgrades, for the Test arm and the Control arm.
Multiple breakdowns can be added to each step or across different steps in the funnel, and you'll get a segment for each unique combination of payload field values and dimension arms.
High variability in certain fields can make it difficult to analyze breakdowns. In such cases, it can be useful to group field values. For example, the number of basket items can be grouped into ranges (1, 2, 3-5, 6-10, 11+), or URLs can be categorized by app sections like homepage, blog, pricing, etc.
While you can manually add these fields to your events, this may not be practical if you recognize the need midway through an experiment. Additionally, it increases the data sent to Hypertune. To resolve this, Hypertune offers derived fields, which you can define at any funnel step. First, select the type, then configure logic to calculate the field value for each event. These fields can then be used in breakdowns or aggregations.
Beyond simple event counts, you may need additional metrics to fully understand your experiment's results. For example, knowing the average purchase price might be crucial, especially if a lower price isn’t desirable despite an increase in purchases.
You can add aggregations to any funnel step that has numeric fields, including derived ones. This allows you to calculate the minimum, maximum, average, and sum for the selected field.
Once your funnel is set up, you can analyze data across different time periods and determine the winning A/B test variant. Hypertune provides Bayesian statistics to help you assess the probability that a specific variant is the best option, along with confidence intervals to confirm statistical significance.
