Skip to main content



We've gotten an early taste of funnels during the initial tour. Here, we learn about funnels in much more depth. Funnels are a powerful tool for identifying meaningful sequences of steps that actors take on their journeys through the system. Interana can rapidly compute these funnels across all the actors in the dataset, and calculate useful metrics for each step and the funnel as a whole.

We have detailed documentation of funnel operations so you can read up on the subject later. For now, we're going to concentrate on getting some hands-on experience working with them.

Another look

Head back to the list of All funnels from the navigation panel:

Then select the Contentious Articles funnel:

You'll see the funnel showing a long sequence of steps around new articles that have been edited at least three times before hitting an abuse filter and then getting deleted. If you hover over the steps and time ranges between steps, it'll look something like this:

Head over to the funnel controls and click on Copy to check out how the funnel is defined:

What you'll see is a sequence of six steps, each of which is defined in terms event characteristics that represent the step being reached. Toward the bottom of the funnel definition are settings, including the time period within which all the steps should occur. It will look like this:

For each article observed during the selected time period, this funnel looks for a starting point where an article is first created (type is one of new), goes through three edits (type is one of edit), then somebody hits an abuse filter while attempting to change the article (log_action is one of hit), and ultimately the article gets deleted (log_action is one of delete). The events are limited to those in the Main space of English Wikipedia, and to qualify for the funnel all the events need to happen within 12 weeks of the first event. 


How can I use it? The example above looked for three edits, but Interana funnels can have an arbitrary number of steps. You could look for as many repeated events as you'd like. For example:

  • An eCommerce company can use the funnel to identify shoppers who added and then deleted an item to their cart twice before finally completing a purchase. Use that information to cohort those shoppers and drill into their motivation or to better understand their journey.
  • UX designers might consider the funnel of steps that ended with an undesirable event and understand how to encourage users to do the right thing.
  • A publisher might might consider the which access limit strategies finally encouraged an active consumer to convert to a paid subscriber. 


Funnels in queries

Funnels are one of the most powerful tools for behavioral analytics because they let you directly examine relevant sequences of events. Interana instruments our funnels with tags and metrics for each step. These are in turn used to flexibly build queries that result in powerful insights. Interana can analyze and compute funnel metrics across millions of actors in seconds, putting this powerful tool into the hands of anybody willing to try different queries. There's no cost to exploration and asking questions to help you discover the right direction of inquiry that lead to the perfect query.

Head back to the landing dashboard by clicking the Interana logo at the top of the navigation bar. Once there, scroll all the way to the bottom and explore from the last chart:

Let's take a look at how the query was built:

All we're doing is charting a Time View plot for the average of five automatic metrics that Interana calculates for funnels, but only for those funnels which have gone through all six steps (terminal_state is one of 6). The automatic metric is the time between particular steps.

How can I use it?  Imagine that these are steps in a relevant conversion flow. For example:

  • The steps customers follow on an eCommerce site as they browse, register, add payment information, and check out. Monitoring the time they spend between steps as you improve their shopping experience is a powerful indication of whether the changes are helping.
  • The installation steps for software, where various errors or aborts might delay the time it takes for a new customer to become a productive user.
  • The successful update flow for an IoT device that has a new version of firmware available and requires the owner to initiate the process after receiving an alert about the new update.


You're probably curious about what some of those contentious articles might be. Let's figure that out.

Change to Table View, delete the Average measures and replace them with a Count Events, Count Unique of user, and a Sum for Contentious Articles_transition_time_to_current_state. Set to group by article and leave the filter for article.max.Contentious Articles.terminal_state is one of 6 the same.

It should looks something like this:

You might need to go to the chart controls and adjust them to run the query unsampled and sort by the count of unique users (as a measure of participation on the article):

Check out the results. Are there certain articles that seem like recent and contentious topics that might have been part of various edit wars and vandalism? Are there certain topics that are likely not allowed on Wikipedia? One indicator might be a large number of participants (Count unique user) combined with a relatively short Sum of time in the funnels. That's likely an indication that Wikipedia administrators stepped in quickly and shut down the article. For example, the article on Kekistan garnered nearly 300 changes from 90 users in less than 2 hours of combined time. It is now protected on Wikipedia and can't be recreated without administrative approval:

If you follow the link to check out the deletion log, you can see that it was deleted based on being a hoax:

Using funnels to calculate conversion rate

One simple use for funnel metrics is to calculate conversion rates. By combining funnels with ratio metrics, we can take a look at the fraction of actors that undergo some conversion event. Let's go through the steps to create a new funnel and ratio metric to try this for yourself. We're going to look at articles that are deleted and then restored. It's a very simple flow that's perfect for hands-on learning.

Head on back to the Funnel icon on the navigation panel and select the Create New Funnel option:


This will open up an empty New Funnel box.

Name the new funnel something like Delete-Restore for each article. Define the first event a having log_action is one of delete, and the second event as log_action is one of restore.

The box should look like this:

Click the big blue Save button when done.

This will create and run the funnel. Unlike our previous example, this funnel looks for events that happen within a week of each other, and isn't limited to a single wiki or name space.

Once the funnel completes, look at the statistics for each step and the time between steps:

What we see is that (at least for the last week) there were very few pages restored after they've been deleted, and those were generally restored quickly. Now we have what we need to construct our ratio metric.

Head over to the Metrics icon on the navigation panel and once there click the New Metric button for Custom Metrics and Ratios:

Name the new metric something like Restored Article Ratio, and define it as a custom ratio with Count Unique of article. For the numerator, filter for the funnel terminal state being 2 (article was restored). For the denominator, filter for terminal state being either 1 or 2 (article was deleted but may or may not have been restored).

The box should look something like this:

Click the Save button and then click the compass icon to explore from your new metric:

There you go: a conversion metric showing the ratio of funnel completions over time. We can see that in general between 1-4% of deleted articles get restored. You might want to use the chart controls to adjust the resolution of the chart and rerun the query unsampled. It might also be useful to change to the Number view, and display the results as a percentage:

Can you figure out how to use the chart controls to get these results? If conversion rates are especially relevant for you, check out our how-to article on using funnels to track conversion rates


How can I use it? Conversion flows are important in a wide range of applications. From apps and websites, to ecommerce, to devices, or even physical businesses. For example:

  • Conversion can be a positive event like a purchase from an eCommerce site.
  • Conversion can be a negative event like a device failure after several early symptoms.
  • Conversion can be quick, like activating a new phone after initial setup.
  • Conversion can be very slow, like using a freemium service for years before upgrading.
  • Conversion can even be low-tech like a customer using a receipt coupon.


Using funnels to tag events occurring after a trigger event

One other very cool power of funnels is to tag all events that occur with a certain time range of an important triggering event. To illustrate this, consider how we earlier used a cohort of newly registered users to look at their actions within 28 days of registration. We focused on registered users, but some Wikipedia users submit changes without registering or logging in. We can use a funnel to do something similar to the cohort.

Head on over to the funnel icon on the navigation panel and start creating a new funnel. Fill out the definition as follows:

Be sure to define the funnel for each user (instead of article). The first step is defined as one that's carried out by an anonymous user — one where the IP address is recorded. The second step is the cool trick: we've defined a derived column that always holds the value 1 — named always1 — and are checking that it's equal to 0. That will never happen, which means the funnel will tag all events associated with that particular user until the funnel expires. That's the last part: we set the funnel to expire 28 days after the first event for the anonymous user. All events that happen within the funnel are thus ones that the user does within their first 28 days.

Click the big blue Save button and let's see what happens...

Interesting! We know that Step 2 will never get reached. By hovering over the bar for Step 1, we can see that over 54% of all unique users were not logged in.

Click that bar and let's explore further.

What we see in the Explorer is a Time View chart showing the number of unique users who left the funnel in either Steps 1 or 2. Since nobody reached Step 2, these are the users who first showed up within 28 days of the time on the chart:

Change the filter to something a bit more direct by matching events where the current_state is one of 1:

Now let's use this to take a look at the sorts of things users who aren't logged in do within 28 days of first showing up.

Set the query builder to Pie view, limit it to English Wikipedia's Main name space, and group by type and log_type.

The result looks something like this:

So it looks like the vast majority of actions are edits to existing articles (84%), with a significant number of events (16%) hitting an abuse filter. But given the difference in those two numbers, it does look like Wikipedia administrators are justified in their trust of anonymous edits. There's abuse but there is still a fair amount of work contributing to Wikipedia that's likely not abusive. Do you find it interesting that there aren't any new articles getting created? That's because by the rules of Wikipedia, users need to be logged in to create new articles. 

Let's compare this with logged in users.

Switch the filters over to Compare mode, and name the first clause something like Anonymous. For the second clause, name it something like Logged In and delete the funnel filter. Then add a new filter for user is in cohort Users Registered Last 28 Days. Click GO.

The chart that comes up should look something like this:

Interesting stuff! It looks like logged in users are actually hitting the abuse filter more frequently than anonymous users. That's unexpected behavior and something that might be worth exploring in more depth later on...


How can I use it? A funnel with a final step that never occurs can be used to tag and explore all the events that happen after some triggering event. That could be any event for a class of actors like our anonymous users. Or it could be used at the end of a longer funnel that explores what happens after a certain kind of journey.
For example:

  • What do shoppers do after they add something to their cart, regardless of whether they check out or not?
  • What does a certain device do after first detecting some sort of problem?
  • How close to home and how quickly does a new rider hail their first ride after signing up for ride sharing?

These are all situations where this kind of funnel can help you analyze behavior. These funnels are more powerful than cohorts because they tag the actual events, not just the actors, and are not tied to a particular time range in the data.

Wrapping Up

That's it! You've now gained hands-on experience working with complex funnels, using them to construct conversion metrics, and even constructing funnels that tag all events after a certain trigger. Funnels are a very powerful tool for exploring behavior, and this is just a small taste of what they can do. Check out our discussions and how-to articles for more about using funnels in Interana. 

Next steps 

This is the last lesson in the Live Demo Guide. Next up, we're going to wrap up the whole experience and share some places you can go for more information. 

Keep in mind that you're using a shared demo system meant for learning by everybody. The dashboards and objects you create will stick around for a while, but we will periodically clean up the system and remove stale accounts.

  • Was this article helpful?