Funnels analysis: Conversion drivers

Knowing which events lead to conversions and which events don’t is a crucial part of any analytics program. With Amplitude, you also have the ability to conduct deeper analyses and learn why users convert or churn after a specific event, with conversion drivers.

A conversion drivers analysis begins with a simple, two-step funnel. Step one should be the starting event—for example, Search Song or Video in a music app—while step two should be the conversion event you’re interested in, like Purchase Song or Video. Amplitude will then automatically sort through, aggregate, and analyze all events that occur for each user between these two steps, in order to identify which user actions are most strongly correlated with that outcome. 

Use this feature to understand which behaviors are driving key outcomes in your customer journey. To help you do that, Amplitude provides several relevant metrics in each conversion driver analysis:

• A correlation score
• Behavior frequency
• Percentage of users engaging in that behavior
• Overall time to convert when a user engages in that behavior

These metrics should help clarify the frequency of different user actions, and whether they help or impede conversion.

Before you begin:

• This feature is only available to Enterprise, Growth, and Scholarship plans.
• If you haven’t already, familiarize yourself with the funnels analysis chart.
• Always keep in mind that correlation does not equal causation.
• This feature will only work for funnel charts that are set up for the conversion metric, and the order of the events is set to “this order.”

See this article for a more advanced look at funnels analysis. 

Analyze events performed between funnel steps

To start a conversion drivers analysis, follow these steps:

1. From within a funnel analysis chart, find the step you’re interested in examining as a conversion event. It can be any step after the initial event in the funnel. Doing this will open the Microscope.
   
   
2. From the Microscope, click Investigate Conversion Drivers. This will open the conversion drivers panel.

The conversion drivers panel has two sections. The step controller gives you the option to choose the beginning and ending steps for your conversion driver analysis. In the screenshot below, the analysis is set to look at what happens between the Play Song or Video and Purchase Song or Video events. You can change these by clicking on them and selecting the events you’re most interested in.

This section also displays the conversion and drop-off numbers for the steps you’ve selected, in terms of both unique users and a percentage of users.

Below the step controller is the events table. This lists all the events users have performed between the two selected steps. At the top of the table, you can choose to look at the event list either for users who converted, or for users who dropped off.

The events table shows four relevant metrics for each event listed:

• Correlation Score: Correlation means there's a relationship between two things. In this context, the correlation column quantifies the relationship between the event in question and conversion (or drop-off), depending on which tab you selected (Converted vs. Dropped Off). The higher the score, the stronger the relationship.
  For more information, read the next section on "Understanding Correlation."
• Frequency: The average number of times users fired a given event between the two selected funnel steps.
• % Who Did Event: The percentage (and absolute numbers) of users in the selected cohort who fired a given event.
• Time Between Steps: How long it took users who fired a given event to convert between the two selected funnel steps. This helps understand if firing a given event acts to accelerate conversion, or to impede it.

How Amplitude identifies events to include in a conversion drivers analysis

For users who convert, Amplitude looks at the events performed between the timestamps of the two selected funnel steps. For users who churn, Amplitude looks at the timestamps of the first selected funnel step, and their entry into the funnel plus the conversion window.

Imagine a funnel defined as A --> B --> C, and you wanted to investigate drivers of conversion at step C. The time periods analyzed each set of users is shown below, where t() represents the timestamp of the event performed:

Understand the correlation score

Correlation is a measure (ranging from -1 to 1) of how two variables relate to each other. In a conversion drivers analysis, the variables for each user are:

• whether the user performed the selected event; and 
• whether the user was in the cohort (converted or dropped off) you selected. 

NOTE: You may have heard of different variations and definitions of correlation, including Matthews correlation, Pearson correlation, phi coefficient, and R-value. In this case, all these definitions are equivalent because a conversion drivers analysis looks at pairs of binary random variables.

In the screenshot above, we see a strong positive correlation for the Add Content to Basket event for users who have converted. Positive correlations indicate that these events tend to occur together. Therefore, Add Content to Basket could be a conversion driver. 

For a detailed confusion matrix that shows the count of users in your base cohort who constitute your True Positives (TP), False Positives (FP), False Negatives (FN), and True Negatives (TN), click Correlation data.

Remember, correlation is not causation, so hypotheses generated by a conversion drivers analysis still must be tested and verified in the real world. Here are some more technical definitions of correlation:

1. Correlation of X and Y is the covariance of X and Y divided by the geometric mean of their variances;
2. If X is modeled as an affine function of Y and Y is modeled as an affine function of X, each with minimal root mean squared error, then the correlation of X and Y is the geometric mean of the predictive coefficients of these two functions.

Event properties and conversion drivers

When you look at combinations of attributes on an event, you get a more accurate picture of what a user is actually doing in your product, which leads to a more layered and nuanced analysis.

To use this feature, open a funnel chart and follow these steps:

1. On the chart, find the event you’re interested in analyzing. Open the Microscope by clicking on either the top section (for churn) or the bottom section (for conversions). Then click on Investigate Conversion Drivers. This will open the conversion drivers tab.
   
   
   
   The conversion drivers tab lists every event included in your project, along with each event’s correlation with either conversion or churn. In this example, the Add Content to Cart event is very highly correlated (+0.97) with conversion on the Purchase Song or Video event.
   
   NOTE: You can switch between viewing correlations with conversions and correlations with churn by clicking Converted or Dropped Off, directly above the list of events.
   
   
2. Locate the event you’re interested in. Below the event name, click Expand by Property.
3. Click the Select property … button and click the property you’re interested in analyzing.
   
   
   
   In this example, we’re interested in finding out which genres are most frequently added to customers’ carts and then purchased—remember, we are looking at users who converted on Purchase Song or Video. Here, we see pop is the most popular genre, with a correlation of +0.41.

NOTE: You can add up to three different properties. You can also create another, separate property view by clicking +. Note that each property view is completely independent of any other property views you may have already created.

Share the report

When you find a valuable insight using conversion drivers, you can easily share it with a teammate:

1. Click Share.
2. Click Copy Chart Link to copy the chart’s unique URL to your clipboard, from which you can send your analysis to others.