Impact analysis with Application Insights

Impact analyzes how load times and other properties influence conversion rates for various parts of your app. To put it more precisely, it discovers how any dimension of a page view, custom event, or request affects the usage of a different page view or custom event.

Impact tool

Still not sure what Impact does?

One way to think of Impact is as the ultimate tool for settling arguments with someone on your team about how slowness in some aspect of your site is affecting whether users stick around. While users may tolerate a certain amount of slowness, Impact gives you insight into how best to balance optimization and performance to maximize user conversion.

But analyzing performance is just a subset of Impact's capabilities. Since Impact supports custom events and dimensions, answering questions like how does user browser choice correlate with different rates of conversion are just a few clicks away.

Screenshot conversion by browsers


Your Application Insights resource must contain page views or custom events to use the Impact tool. Learn how to set up your app to collect page views automatically with the Application Insights JavaScript SDK. Also keep in mind that since you are analyzing correlation, sample size matters.

Is page load time impacting how many people convert on my page?

To begin answering questions with the Impact tool, choose an initial page view, custom event, or request.

Impact tool

  1. Select a page view from the For the page view dropdown.
  2. Leave the analyze how its dropdown on the default selection of Duration (In this context Duration is an alias for Page Load Time.)
  3. For the impacts the usage of dropdown, select a custom event. This event should correspond to a UI element on the page view you selected in step 1.

Screenshot of results

In this instance as Product Page load time increases the conversion rate to Purchase Product clicked goes down. Based on the distribution above, an optimal page load duration of 3.5 seconds could be targeted to achieve a potential 55% conversion rate. Further performance improvements to reduce load time below 3.5 seconds do not currently correlate with additional conversion benefits.

What if I’m tracking page views or load times in custom ways?

Impact supports both standard and custom properties and measurements. Use whatever you want. Instead of duration, use filters on the primary and secondary events to get more specific.

Do users from different countries or regions convert at different rates?

  1. Select a page view from the For the page view dropdown.
  2. Choose “Country or region” in analyze how its dropdown
  3. For the impacts the usage of dropdown, select a custom event that corresponds to a UI element on the page view you chose in step 1.

In this case, the results no longer fit into a continuous x-axis model as they did in the first example. Instead, a visualization similar to a segmented funnel is presented. Sort by Usage to view the variation of conversion to your custom event based on country.

How does the Impact tool calculate these conversion rates?

Under the hood, the Impact tool relies on the Pearson correlation coefficient. Results are computed between -1 and 1 with -1 representing zero correlation and 1 representing a positive correlation.

The basic breakdown of how Impact Analysis works is as follows:

Let A = the main page view/custom event/request you select in the first dropdown. (For the page view).

Let B = the secondary page view/custom event you select (impacts the usage of).

Impact looks at a sample of all the sessions from users in the selected time range. For each session, it looks for each occurrence of A.

Sessions are then broken into two different kinds of subsessions based on one of two conditions:

  • A converted subsession consists of a session ending with a B event and encompasses all A events that occur prior to B.
  • An unconverted subsession occurs when all A's occur without a terminal B.

How Impact is ultimately calculated varies based on whether we are analyzing by metric or by dimension. For metrics all A's in a subsession are averaged. Whereas for dimensions the value of each A contributes 1/N to the value assigned to B where N is the number of A's in the subsession.

Next steps