Understand your influence analysis

This article explains how to interpret the results of an Influence analysis in Optimizely Analytics. It covers the key statistics calculated by the template and describes how to read and understand the individual modules within the exploration template. For information on how to set up an Influence analysis, see Influence analysis overview. The Influence template has the following modules:

The Actors module lets you define the primary entity for your analysis and specify which subset of actors to include. For example, set the Actor to Accounts and include only the Enterprise Customers cohort to analyze how group-level activity drives their renewal rate.

The Actors module has the following fields: 

The Test module is where you define the events, metrics, or cohorts you want to analyze as potential drivers of your chosen target. To get the most value, align your Test type with your specific analytical goal. Use the Type field to specify whether your test is an Event, Metric, or Cohort. For information, see Test types. You can select the following test types:

When you select Event as the test type, you analyze how specific events influence a target (event, metric, or cohort) by comparing the likelihood or impact for users who did or did not perform those events. For example, you can analyze if an Add to Cart event influences a Make Purchase target differently depending on whether it occurred on a Mobile or Desktop device.

When you select Event as the test type, the following fields display: 

When you select Cohort as the test type, you analyze how belonging to a specific user cohort influences your chosen target. The Cohorts field lets you select one or more cohorts as your test variable and compare them against your target to determine their impact. For example, a streaming service can set Premium Subscribers as the test cohort to analyze if they are significantly more likely to Start a New Series compared to the average user base.

When you select Metric as the test type, you analyze how the value of a specific metric influences your chosen target. This can involve correlation analysis (for Metric → Metric) or impact analysis (for Metric → Event/Cohort). For example, you can analyze if a higher Average session duration (test metric) strongly correlates with an increase in Monthly recurring revenue (target metric) for your subscription platform.

The Test Metrics field lets you add one or more metrics as your test variable by clicking +. 

The Target module is where you define the specific outcome or behavior you want to understand or influence. For example, define Made Purchase (an event), Average Session Length (a metric), or High-Value Customers (a cohort) as your target depending on your analytical goal.

Use the Type field to specify whether your target is an Event, Metric, or Cohort. For information, see Target types. You can select the following target types:

Select Event as your target to define a specific user action or conversion point as your primary outcome, and analyze how various test factors (events, metrics, or cohorts) influence the likelihood of it occurring.

The following fields display:

When Event is your target, you can combine it with the following test types:

For example, if your target is Make purchase and your test is Add to cart, the template calculates the Likelihood Impact by dividing the purchase rate of users who added to cart by the purchase rate of all users; a result of 2x indicates that users who added to cart are twice as likely to complete a purchase compared to the average user.

Select Cohort as your target to define a specific user group as your primary outcome of interest. This is ideal when your goal is to grow a user segment, improve retention, or understand the factors that define important user groups.

The Cohorts field selects a single user cohort as your target. The analysis identifies how test factors associate with belonging to that cohort.

When Cohort is your target, you can combine it with the following test types:

For example, if your target cohort is Loyalty program subscribers and your test is First time visitors, the template calculates the Likelihood Impact by dividing the percentage of first-time visitors who are in that cohort by the overall percentage of users in that cohort; a result of 1.5x indicates that first-time visitors are 50% more likely to be loyalty members than the average user.

Select Metric as target when your primary objective is to optimize a quantifiable business outcome. This mode helps you understand the drivers of performance indicators. When you select Metric as the target type, you define a specific quantifiable metric as your primary outcome of interest. 

The Metric field lets you select a single metric as your target. You can add only one metric. 

When Metric is your target, you can combine it with the following test types:

For example, if you select Total purchase revenue as your target metric and Average order value as your test metric, the template calculates the Correlation (Pearson r) between them; a result such as r = 0.85 indicates a strong positive relationship, meaning that as users' average order values increase, their total purchase revenue also tends to rise significantly.

After setting up your influence template by defining your target, test, and actor modules, the Filters module lets you narrow down the data presented in your visualizations. This makes it easier to answer exploratory questions and drill down into specific segments of your analysis.

For instance, if your main analysis is for All Users, you can use a filter to quickly see results for Mobile Users or users from a specific region without having to reconfigure your entire Actor module. This provides flexibility for ad-hoc exploration and deeper dives into your insights.

Analytics also lets you choose JSON columns in this module. When you click a JSON column, it expands to display the available keys for that column. You can choose a key and click Apply. The selected end key becomes the display name for that column.

The impact template's visualization module lets you run and view the analysis in a chart. It also lets you edit the visualization, examine the underlying SQL query, and add the chart to a dashboard.

There are five important sections in this module.

These options let you switch the values on the Y-axis from absolute values to percentages and vice versa in one click. This feature is particularly useful when dealing with multiple segments, as using absolute values lets you see a clearer breakdown of how many individuals are in each segment.

To choose the number of top segments you want to display in your impact analysis, click the Show top drop-down list and enter the required number. You can also use the + and – symbols to enter your input. The total count of selected events defines the top attribute values. You can also set the number of top 'n' segments when you create beelines.

You can set the time range using a drop-down list or choose from the quick options and iterate through different choices of time range and time grain without leaving the chart. It is also possible to set a lag by clicking Offset and setting the Ending.

In the visualization module, you can configure sampling modes for your exploration by clicking the lightning icon at the top of the exploration. The drop-down list displays two sampling modes: Enabled Faster Response or Enabled Higher Precision.

When you define your target, test, actors, and time range, the Influence template automatically calculates key statistics to quantify the relationships in your analysis. This module provides a detailed breakdown of these statistical outputs, helping you understand the "why" and "how much" behind the influences you identify.

The specific statistical interpretations available depend on the combination of your chosen test and target types. At a high level, these statistics help you understand:

This statistic measures how much more (or less) likely users who performed a test action are to also perform the target action, compared to those who did not. It helps you identify which user behaviors or features are most influential in driving a target outcome.

It applies to the following test → target types:

Calculation – (Conversion rate of test group) / (Baseline conversion rate)

Example

If users who trigger the Product Video Viewed event are 3x more likely to also trigger the Add to Cart event, it shows videos strongly drive purchase intent.

Imagine you have 100 total users included in the analysis. Out of these 100 users, 20 (20%) Add to Cart. This is your baseline conversion rate. Now, let's say 50 users watched the Product Video Viewed (your test group). Out of those 50 video watchers, 30 (60%) went on to Add to Cart.

The likelihood impact is calculated as: (60% conversion rate for video watchers) / (20% baseline conversion rate) = 3x. This means video watchers are three times more likely to add to cart.  

A result of 3x means video watchers are three times more likely to add to cart, highlighting the video's strong influence.

The following Events → Event analysis tests multiple events against the Make Purchase target for Users within a 7-day window. The scatter plot shows that Browse, View Content Details, and Search are the strongest drivers of purchase, sitting in the Higher Impact, Higher Reach quadrant — meaning they are both widely performed and highly influential. Pause Content, Resume Content, and Play Content have an even higher likelihood impact but are performed by fewer users, while Marketing Touchpoint shows comparatively limited reach and influence.

![A screenshot of a computer

AI-generated content may be incorrect.](https://support.optimizely.com/hc/article_attachments/44397696296333)

This statistic tells you what percentage of your total audience actually performed a specific test action or belonged to a certain group (cohort). It is the proportion of your entire user base that engaged with the test. This helps you understand the reach or adoption of a particular feature or campaign, giving context to other impact metrics.

It applies to the following test → target types:

Calculation – (Count of users who did test event) / (Count of all users) 

Example

If only 10% of your total user base triggered the Promotional Banner Clicked event, it tells you the banner's reach is limited.

Imagine you have 100 total users. Only 10 of those users clicked on Promotional Banner Clicked. 

The % of actors who did the test is calculated as: (10 users who clicked) / (100 total users) = 10%. This shows that only a small fraction of your audience interacted with the banner.

The following screenshot shows a Cohorts → Cohort analysis that tests three cohorts — Made Purchase within 7 Days of Signup, Regular Watchers, and Viewed Recommended Content — against the target cohort Loyalty Program Subscribers for All Users.

The scatter plot shows that Viewed Recommended Content is the strongest predictor of loyalty program membership, sitting in the Higher Impact, Higher Reach quadrant — meaning it is both widely adopted and highly influential. Made Purchase within 7 Days of Signup shows moderate reach and impact, while Regular Watchers is the weakest predictor of the three. Overall, engaging with recommended content appears to be the most significant behavioral driver of loyalty program membership, though these relationships should be validated through further experimentation.

This statistic tells you what percentage of your total audience belongs to a specific test cohort. It is the proportion of your entire user base that falls within the defined group you are testing. This helps you understand the size and reach of a particular cohort, giving context to other impact metrics and helping you assess how representative or niche the group is within your overall audience.

It applies to the following test → target types:

Calculation – (Count of users who belong to test cohort) / (Count of all users)

Example

If only 15% of your total user base belongs to the Loyalty Program Members cohort, it tells you this is a relatively small but potentially high-value segment of your audience.

Imagine you have 100 total users. Only 15 of those users belong to the Loyalty Program Members cohort.

The % of actors who are in the test cohort is calculated as: (15 users in cohort) / (100 total users) = 15%. 

This shows that while the cohort represents a small fraction of your overall audience, any strong influence this group has on your target outcome is particularly meaningful given its limited size.

This statistic measures how much a specific test action changes a key number (a metric) for users who performed that action, compared to the average for all users. It is a ratio comparing the average metric value for the test group versus the overall average.

It applies to the following test → target types:

Calculation – (Average target or test metric value for test group / target cohort) / (Average target/test metric value for all users)

Example

If users who trigger the Chatbot Interaction event have an Average Order Value that is 1.2x higher, it suggests that the chatbot boosts spending.

Consider 100 total users. The Average Order Value across all 100 users is $50. For example, 30 users had a Chatbot Interaction. For these 30 users, their Average Order Value is $60.

The Metric Impact is calculated as: ($60 Average Order Value for chatbot users) / ($50 Average Order Value for all users) = 1.2x.  

This means that chatbot users spend 20% more on average.

This statistic measures whether two metrics tend to move together, producing a value between -1 and +1. It identifies potential relationships within your user experience, helping you spot trends and areas for further investigation, but does not imply that one metric causes the other.

It applies to the Metric(s) → Metric test and target type.

Calculation – Pearson correlation coefficient (r).

Example

A strong positive correlation (for example, r = 0.9) between two metrics means the two metrics tend to increase together. For example, users who spend 60 seconds on a product page might have a 5% purchase rate, while users who spend 120 seconds might have a 10% purchase rate.

However, correlation does not imply causation; other factors may explain the relationship. For example, highly motivated buyers may naturally spend more time on a page before converting. Use this as a signal to form a hypothesis and validate it through controlled experimentation before drawing conclusions.

The following example is a Metrics → Metric analysis testing three metrics — Average Titles Viewed per User, Average Ad Revenue per User, and Total Purchase Revenue — against the target metric Total Time Viewed (TTV) for All Users over a 30-day period.

The bar chart displays the Pearson Correlation (r) value for each test metric against the target, revealing the following:

All three metrics show strong positive correlations with Total Time Viewed, indicating that engagement and revenue metrics tend to move together. However, it is important to note that these correlations do not imply causation — they signal relationships worth investigating further through controlled experimentation.