Skip to contents

Synthetic touchpoint event log

Source: R/data.R
mm_events.Rd

A raw marketing event log of the kind exported from a web analytics or customer data platform, containing on purpose every case that breaks a naive journey-building pipeline.

Usage

mm_events

Format

A data frame with 20,799 rows and 5 columns:

customer_id

Customer identifier.

channel

Channel label. Includes direct-traffic labels ("direct", "(none)", "") and genuine NA values.

timestamp

Event time, POSIXct in UTC.

conversion

1 on a converting event, 0 otherwise.

value

Conversion value, NA except on converting events.

Source

Simulated. See data-raw/make_mm_events.R in the package sources.

Details

The awkward cases are deliberate, because they are what a journey builder has to handle:

  • Repeat converters. Some customers convert two or three times, separated by dormant periods of 20 to 140 days. Treated as one journey per customer they become a single implausibly long path; split on conversion they are several short ones.

  • Consecutive duplicates. Half of all multi-touch bursts contain the same channel twice in a row – some placed deliberately, the rest arising by chance from a six-channel sampler.

  • Direct, none, blank and missing channels. A quarter of bursts contain one; 261 events have a genuinely missing channel.

  • Non-converters. 62% of customers never convert. They are the null paths that Markov removal effects are computed against.

  • Timestamp ties. 624 events share a customer and a timestamp exactly, so journey order depends on a deterministic tie-break.

  • Channels with positional roles. Display and social tend to open journeys, paid search and email to close them, so first-touch and last-touch attribution genuinely disagree about them – which is what makes attribution_spread() and channel_positions() worth looking at.

  • Unsorted rows. The table is shuffled, as a real export would be.

See also

build_paths(), mm_weekly for the media-transform counterpart.

Examples

data(mm_events)
str(mm_events)
#> 'data.frame':	20799 obs. of  5 variables:
#>  $ customer_id: chr  "cust_03227" "cust_03940" "cust_04838" "cust_00749" ...
#>  $ channel    : chr  "email" "" "organic_search" "affiliate" ...
#>  $ timestamp  : POSIXct, format: "2025-02-10 18:08:57" "2025-05-20 04:44:17" ...
#>  $ conversion : int  0 0 0 0 0 0 0 0 0 0 ...
#>  $ value      : num  NA NA NA NA NA NA NA NA NA NA ...

# The awkward cases really are in there
sum(is.na(mm_events$channel))
#> [1] 261
table(mm_events$channel, useNA = "ifany")
#> 
#>                        (none)      affiliate         direct        display 
#>            465            427           2016            444           3799 
#>          email organic_search    paid_search         social           <NA> 
#>           2537           3333           4017           3500            261 

# Rows are shuffled, as a real export would be
is.unsorted(mm_events$timestamp)
#> [1] TRUE

# Customers converting more than once
conv <- table(mm_events$customer_id[mm_events$conversion == 1])
table(conv)
#> conv
#>    1    2    3 
#> 1509  349  178