Diving into a case study to cover the promising aspects of Consent Mode as a technology and the big challenges it promises to solve while addressing some of the immediate issues and required workarounds to get full advantage of it.
3. The major issues that Consent Mode aims to address, along with
the significant benefits it promises to deliver.
4. The major issues that Consent Mode aims to address, along with
the significant benefits it promises to deliver.
5. Big problems to solve.
Regulatory Changes
Regulations such as GDPR and
ePrivacy Directive are impacting
how data can be collected and
used
3rd Party Cookies
Use cases for measuring and
reaching users with pixels are
limited and will continue to
decrease over time
Browser Updates
Heightened controls are
impacting traditional data
collection (e.g. third party
cookies and device identifiers)
7. >70%
On average, Consent Mode recovers
more than 70% of ad-click-to-
conversion journeys lost due to user
cookie consent choices.
Big promise.
8. How it works?
A customer visits your
website and makes
their consent selection
for the use of cookies
on your consent
banner.
Consent Mode adjusts
how your Google
tracking behaves,
based on the
customer’s consent
choices.
If consent is granted,
conversions and behavior
data are reported
normally. If not, Consent
Mode recovers the data
using modelling.
9. The prerequisites.
Consent mode for web pages must be implemented so
that tags are loaded before the consent dialog appears,
and Google tags load in all cases (gcs=100), not only if the
user consents (gcs=111)
The property collects at least 1,000 events per day with
analytics_storage='denied' for at least 7 days.
The property has at least 1,000 daily users sending
events with analytics_storage='granted' for at least 7 of
the previous 28 days.
10. My team implemented Google Consent Mode via Google Tag Manager. This approach allowed to adjust how
Google's tags behave based on the consent status of end-users, thereby maximizing data collection while
respecting user consent choices.
Consent
Mode
Observed data
(with Cookies)
Cookieless pings
for Modelling
By adjusting the settings of analytics_storage and ad_storage, we could control the behavior of Google
Analytics cookies and ad conversion tracking cookies, respectively.
Cookiebot
Case Study
11. Data exported from HubSpot, Google Analytics 4 and Cookiebot
Date range Q2 2023 - 01/04/23 to 30/06/2023
Conversion = website form completion
Without
Consent Mode
With Consent Mode,
Firing cookieless pings for
ad_storage or analtics_storage
= denied users
Modelled
Conversions
Observed
Conversions
Observed
Conversions
80% of lost conversions recovered
Actual
Conversions
Data exported from
the CRM
5060 conversions
4517 conversions
2337 conversions
Average opt-in
rate
76%
Data exported from HubSpot, Google Analytics 4 and Cookiebot
Date range Q2 2023 - 01/04/23 to 30/06/2023
Conversion = website form completion
Case Study - The Results?
+ Same level of data recovery
found for Users and Sessions.
12. Data generated through modelling in GA4 is only accessible if the
“Blended” Reporting Identity in the Property is selected.
13. Data generated through modelling in GA4 is only accessible if the
“Blended” Reporting Identity in the Property is selected.
14. Reporting Identity
User ID (unique, business provided)
Google Signals ID (Google profile)
Modelled (estimated)
Device ID (anon. cookie)
18. Reporting Identity Options
(When Google Signals are enabled) Data thresholds are
applied to prevent anyone viewing a report or exploration
from inferring the identity of individual users based on
demographics, interests, or other signals present in the
data.
Must always be enabled as otherwise links with Advertising
platforms do not work.
Data thresholds are system defined. You can’t adjust them
19. Reporting Identity Options
More data when aggregates are viewed e.g. Users,
Sessions, Conversions etc
Less data when granular information is needed - e.g.
pages and page-level metrics
21. How does modelling work?
Black
Box
Input Output
Observed data
Unobserved data
Modelled behavior and conversion
data
Modelled
Observed
22. A peek inside the black box
This is all based on an overview of how modelling using Consent
Mode works for Google Ads, based on what Google provided as an
explainer.
23. First, we separate ad interactions into 2 groups: one where
we can observe the link between an ad interaction and
conversion, and one where we cannot observe the link.
1
A peek inside the black box
24. First, we separate ad interactions into 2 groups: one where
we can observe the link between an ad interaction and
conversion, and one where we cannot observe the link.
1 Then, we divide the observed group into subgroups that
share non-sensitive characteristics like:
device type, browser, country, conversion type, etc.
2
Within each of these subgroups, we calculate conversion
rates.
3
A peek inside the black box
25. Next, we take the ad interactions and conversions that
are missing a link, and assign them to one of the
existing subgroups based on shared characteristics.
4
A peek inside the black box
26. Next, we take the ad interactions and conversions that
are missing a link, and assign them to one of the
existing subgroups based on shared characteristics.
4 Using the known conversion rates from the
observed population and machine learning, we can
model
which unlinked ad interactions belong to which
unlinked conversions.
5
A peek inside the black box
27. Once the ad interactions and conversions have the appropriate links
between them, we aggregate them and surface them in the reporting.
We only include modeled data in reporting when we have
high confidence that conversions actually occurred as a result of ad
interactions.
6
A peek inside the black box
28. Once the ad interactions and conversions have the appropriate links
between them, we aggregate them and surface them in the reporting.
We only include modeled data in reporting when we have
high confidence that conversions actually occurred as a result of ad
interactions.
6
A peek inside the black box
Problem: data is aggregated and
you cannot tell what % of what
you see is modelled and what is
observed.
29. Conversion = website form submission
Conversions
Blended
Device-
based
You cannot get this kind of a
report within the GA4
interface.
A peek inside the black box
30. Reporting Identity Options
More data when aggregates are viewed e.g. Users,
Sessions, Conversions etc
Less data when granular information is needed - e.g.
pages and page-level metrics
31. How do you actually leverage Consent Mode, without sacrificing
granularity and limiting other aspects of GA4?
// Discuss
32. Some of the options we consider
Reporting
Identity
Blended
Device-based
Analyse
Report
1
Dummy way
33. Some of the options we consider
1 - Blended
2 - Device - based
Analyse
Report
Duplicate tracking
2
A bit less dummy way
34. Some of the options we consider
Report
Analyse
3
A reasonably not dummy way
35. Some of the options we consider
4
A bit out-of-scope way
36. Some of the options we consider
5 Have you considered something different?
37. Promising tech to solve complex problems
Black box and WYSIWYG
Thresholding issues require workarounds to get all benefits.
Recap
38. Yasen Lilov
Data & Analytics Service Lead,
VertoDigital
Thank you!
/in/yasenlilov/
yasen.lilov@vertodigital.com
Scan to connect!
Editor's Notes
The Good:
> How it works (simplified) - tech, setup, limit requiements and what this results in.
> > Case Study Plug-in - SnapLogic - Results in real world + mention other clients where we see similar volumes and results
The Good:
> How it works (simplified) - tech, setup, limit requiements and what this results in.
> > Case Study Plug-in - SnapLogic - Results in real world + mention other clients where we see similar volumes and results
The Good:
> How it works (simplified) - tech, setup, limit requiements and what this results in.
> > Case Study Plug-in - SnapLogic - Results in real world + mention other clients where we see similar volumes and results
The Good:
> How it works (simplified) - tech, setup, limit requiements and what this results in.
> > Case Study Plug-in - SnapLogic - Results in real world + mention other clients where we see similar volumes and results
Before we get to the “UGLY” there are two more things to depict here.
The Ugly:
How to deal with the situation.
0. BigQuery where you model the data and create your own reporting (there’s info on consented vs non-consented)1. BigQuery for reporting (exact numbers with no thresholding) & Interface & Sperate LS for "Data with estimates" (pros/cons)
2. Constantly On/Off when working with the data (pros/cons)
3. ? -a few ideas and open the floor for a discussion
The Ugly:
How to deal with the situation.
0. BigQuery where you model the data and create your own reporting (there’s info on consented vs non-consented)1. BigQuery for reporting (exact numbers with no thresholding) & Interface & Sperate LS for "Data with estimates" (pros/cons)
2. Constantly On/Off when working with the data (pros/cons)
3. ? -a few ideas and open the floor for a discussion
The Ugly:
How to deal with the situation.
0. BigQuery where you model the data and create your own reporting (there’s info on consented vs non-consented)1. BigQuery for reporting (exact numbers with no thresholding) & Interface & Sperate LS for "Data with estimates" (pros/cons)
2. Constantly On/Off when working with the data (pros/cons)
3. ? -a few ideas and open the floor for a discussion
The Ugly:
How to deal with the situation.
0. BigQuery where you model the data and create your own reporting (there’s info on consented vs non-consented)1. BigQuery for reporting (exact numbers with no thresholding) & Interface & Sperate LS for "Data with estimates" (pros/cons)
2. Constantly On/Off when working with the data (pros/cons)
3. ? -a few ideas and open the floor for a discussion
The Ugly:
How to deal with the situation.
0. BigQuery where you model the data and create your own reporting (there’s info on consented vs non-consented)1. BigQuery for reporting (exact numbers with no thresholding) & Interface & Sperate LS for "Data with estimates" (pros/cons)
2. Constantly On/Off when working with the data (pros/cons)
3. ? -a few ideas and open the floor for a discussion