Funnels: Evaluate data from a series of related events

With NRQL, you can use funnels to evaluate sets of related actions. The actions generally share a relationship via an identifier such as a user ID or session ID.

Why it matters

NRQL funnel functions can answer questions like, "Of the people that completed step A, how many also completed step B, and of the people that completed steps A and B, how many also completed step C?"

For example, at New Relic, we could use funnel to track the number of users who completed these steps:

1. Visited the New Relic marketing page.
2. Signed up for an account.
3. Created a dashboard.

Funnel query structure

Here's the structure of a simple NRQL funnel query:

FROM DATA_TYPE 
  SELECT funnel(AN_ATTRIBUTE, ACTION_A, ACTION_B) 
  SINCE TIMEFRAME

See also a more complex funnel query example.

Funnel queries require the funnel function, an attribute to funnel, and at least two steps:

1. The first step is the anchor step, which will always represent 100% of the results.
2. The second and later steps describe the number of users who have also completed additional actions. This number typically will be less than 100%. However, it could be 100% if every user who completes action (A) also completes the additional actions being queried.

Funnel query technical details

One way to use funnel queries is to calculate the rate at which multi-step, ordered sequences were completed over a given timeframe, like in the New Relic signup example. But you can also use funnels to calculate overlap between several distinct actions that don't have a sequential relationship to one another. In other words, the order of steps doesn't impact the calculations performed.

The way funnel queries compute results varies depending on the size of the data set:

• If the funnel is dealing with fewer than 256 funnel attribute values, it will calculate the value exactly.
• If it's dealing with 256 or more funnel attribute values, it applies an algorithm called MinHash to calculate approximate results for optimized performance.

This means that for large data sets, there may be occasional "false positives" at the level of individual data points. These do not interfere greatly with the accuracy of the numerical estimates provided in query results.

Here's a detailed breakdown of technical details and constraints for funnel queries.

Funnel query example

This example queries the browser monitoring PageView event and its attributes. It queries unique browser sessions that have progressed from browsing a product, to adding to cart, to checkout. Labels are included for each step, indicated by the keyword AS.

FROM PageView SELECT funnel(session, 
    WHERE pageUrl LIKE '%browse%' as 'Browse', 
    WHERE pageUrl LIKE '%cart%' as 'Cart', 
    WHERE pageUrl LIKE '%checkout%' as 'Checkout') 
    SINCE 1 week AGO

This query returns the following:

Include additional actions inside the parentheses of the funnel function in a comma separated list:

SELECT funnel(session, 
    WHERE name = 'Controller/about/main' AS 'Action A', 
    WHERE name = 'Controller/about/careers' AS 'Action B', 
    WHERE name = 'Controller/about/insights' AS 'Action C', 
    WHERE name = 'Controller/about/apply' AS 'Action D') 
    FROM PageView 
    SINCE 1 week ago