FUNNELCOUNT

This section contains reference documentation for the FUNNELCOUNT function.

Funnel analytics aggregation function.

Returns array of distinct correlated counts for each funnel step.

Signature

FUNNEL_COUNT (

STEPS ( predicate1, predicate2 ... ),

CORRELATE_BY ( correlation_column ),

SETTINGS ( setting1, setting2 ... ) )

ParameterArgumentsDescription

STEPS

predicates 1...n

(required) These are individual predicates representing funnel steps which are applied on rows selected by the where clause. Distinct values from the correlation_column that satisfy these predicates are counted per step. For example, all filtered rows that match url = '/checkout' are unionized into a set. Sets are intersected with the sets resulted from the preceding steps, each step retaining only individuals present in previous steps. Finally, unique counts are returned for each step in the funnel.

CORRELATE_BY

correlation_column

(required) Column to leverage for funnel correlation, distinct values from this column are counted per step during aggregation. Only dictionary-encoded columns are supported.

SETTINGS

settings 1...n

(optional) Settings to select and configure a funnel counting strategy:

bitmap (default): This strategy is accurate for INT column, but approximate for other cases where hash codes are used in distinct counting and there may be hash collisions. For accurate distinct counting on all column types, use 'set' instead. See also DISTINCTCOUNTBITMAP.

set: This strategy uses fastutil hash sets. Use with care, unbounded memory cost. See also DISTINCTCOUNT.

theta_sketch: This strategy leverages Theta Sketch framework to provide an approximate funnel count with a small memory footprint. See also DISTINCTCOUNTTHETASKETCH.

nominalEntries: theta-sketch strategy parameter (defaults to 4096). Can only be used in conjunction with theta_sketch setting.

partitioned: This strategy counts funnel steps per segment, then sums up step counts across segments. Correlation column should be configured as partition column for this strategy. See also SEGMENTPARTITIONEDDISTINCTCOUNT.

sorted: This strategy counts funnel steps per segment with zero memory footprint. Correlation column should be configured as sort column for this strategy. Can only be used in conjunction with partitioned setting.

Usage Examples

Many datasets are time series in nature, tracking events of an entity over time. An example of such a dataset could be a user analytics activity log from a commerce web application.

Example

user_idevent_timeurl

U1

2021-10-01 09:01:00.000

/product/listing

U2

2021-10-01 09:17:00.000

/product/search

U1

2021-10-01 09:33:00.000

/product/details

U1

2021-10-01 09:47:00.000

/cart/add

U3

2021-10-01 10:02:00.000

/product/listing

U3

2021-10-01 10:05:00.000

/product/search

U2

2021-10-01 10:06:00.000

/product/search

U2

2021-10-01 10:15:00.000

/checkout/start

U2

2021-10-01 10:16:00.000

/cart/add

U3

2021-10-01 11:17:00.000

/product/details

U2

2021-10-01 11:18:00.000

/checkout/confirmation

U3

2021-10-01 11:21:00.000

/cart/add

U1

2021-10-01 11:33:00.000

/cart/add

U1

2021-10-01 11:46:00.000

/checkout/start

U1

2021-10-01 11:54:00.000

/checkout/confirmation

Funnel

We want to analyse the following checkout funnel:

  • /cart/add

  • /checkout/start

  • /checkout/confirmation

Counts

We want to answer the following questions about the above funnel:

  • How many users entered the top of the funnel?

  • How many of these users proceeded to the second step?

  • How many users reached the bottom of the funnel after completing all steps?

Query

select 
  FUNNEL_COUNT(
    STEPS(
      url = '/cart/add', 
      url = '/checkout/start', 
      url = '/checkout/confirmation'),
    CORRELATE_BY(user_id)
  ) AS counts
from user_log 
counts

3, 2, 2

Notes

Notice that although U1 user added to cart twice, it still counted as one conversion in the first step, as we report on unique counts rather than total events. Also notice that although U2 events were logged out of order, we still counted the user as converted.

Equivalence

The above query is equivalent to the below presto SQL query:

select 
   ARRAY[
     count_if(steps[1]),
     count_if(steps[1] and steps[2]),
     count_if(steps[1] and steps[2] and steps[3])
   ] as counts
 from (
   select 
     ARRAY[
       bool_or(url = '/cart/add'),
       bool_or(url = '/checkout/start'),
       bool_or(url = '/checkout/confirmation')
     ] as steps
   from user_log
   group by user_id
 )

Settings

For a large dataset we could use for example a theta_sketch strategy, or furthermore, partition the data by user_id and leverage a partitioned strategy. It is also important to filter in the where clause so to aggregate only necessary rows.

select 
  FUNNEL_COUNT(
    STEPS(
      url = '/cart/add', 
      url = '/checkout/start', 
      url = '/checkout/confirmation'),
    CORRELATE_BY(user_id),
    SETTINGS('theta_sketch', 'nominalEntries=4096')
  ) AS counts
from user_log 
where url in ('/cart/add', '/checkout/start', '/checkout/confirmation')
counts

3, 2, 2

Another Example

We now want to learn how many users checkout after a text search; as opposed to other entry points such as browsing a product category listing. We want to then analyse the following funnel:

  • /product/search

  • /cart/add

  • /checkout/start

  • /checkout/confirmation

Query

select 
  FUNNEL_COUNT(
    STEPS(
      url = '/product/search',
      url = '/cart/add', 
      url = '/checkout/start', 
      url = '/checkout/confirmation'),
    CORRELATE_BY(user_id)
  ) AS counts
from user_log 
counts

2, 2, 1, 1

Notes

Notice that U1 is not counted in this funnel, as the user did not perform any product search. Both U2 and U3 entered the top of the funnel and performed the second step, but only U2 converted to the bottom of the funnel.

Last updated