There are two different ways that you can limit or organize the data you are looking at in your funnel: filtering in your funnel creation, or grouping the results of the funnel.
When you are building your filter, you can choose to filter a step by a particular property by clicking Add Filter next to the event name. This limits the data that you will see in the funnel to events with that property.
Multiple property filters can be chosen for each step in your funnel. Free Mixpanel plans can only apply up to two filters, while Paid Mixpanel plans can apply as many filters as they like.
You can add as many filters as you like to an event by clicking the small + button at the bottom of the list.
You can select whether you would like the user to match any of these filters, or all of the filters.
Grouping by property or cohorts, on the other hand, is something that you can choose to do in the results of the funnel.
Group by Property
Select the PROPERTY button in the "Overview" section below the funnel, and select the property you want to group by to see how your users operate in the funnel depending on city, country, browser type, etc.
This feature is useful for determining if a group factor such as browser type is having an impact on your conversion rates. For example, if there is a bug in your software on Firefox, you might see a decreased conversion rate in the Firefox group.
You can also choose to filter that grouping to a specific property by using the > arrow button.
Group by Cohorts
You can also choose to group your results by cohorts to compare how different cohorts of users move through your funnel. This can be used to determine whether specific factors of individual cohorts has an effect on the how effective your flows are.
To compare cohorts, select the COHORT button in the “Overview” section, and select any number of cohorts from the list.
In this example, you are comparing Android users and iPhone users in order to determine if mobile OS has an impact on conversion. The results in the chart below will appear as follows:
As you can see in the chart, Android users converted significantly more than iPhone users. This may be due to a bug in the iPhone app, which is making conversion more difficult for those users.
It is only possible to either group by properties or group by cohorts, but not both.
Grouping by property or cohort also allows you to see the statistical significance of each group segment in order to determine whether that segment was statistically significant in conversion.
Learn more about statistical significance in funnels here.
User Count Calculation
When grouping or filtering the results of your funnel, user count will be determined by the number of unique users for each property combination.
For example, if you are an e-commerce site grouping by the property “item” to determine what users are searching for and purchasing, users will appear in the table once for each property they use. In a funnel where Event A is “Search”, Event B is “Add to Cart”, and Event C is “Purchase”, a user may complete the funnel twice, once with the property “hat”, and once with the property “shirt”. This user would convert through the funnel, and be counted in the Group By table for both “hat” and “shirt” when grouping by property “item”. They would be counted in the “overall” row as completing the funnel once.
This behavior matches how users are counted when filtering by property: if a user goes through the entire funnel X times with Y distinct event property values, the user will be counted Y times.
Previously Group By and Filter By only counted unique users once, regardless of their property. This meant that only the first property a customer chose would be shown in the Group By table. This calculation has been changed to improve the accuracy and quality of your data. This change was made on 9/25/2018, and as a result, you will see your results differ from the results calculated using the previous algorithms.
Because of this change all automatic segmentation alerts that were generated with the previous calculation will automatically be removed. Moving forward the alerts will use the new calculation.