Powerhouse - Session 3 - Performance
Powerhouse - Session 3 - Performance
30%
…
Alphabet Amazon Apple Microsoft
M
A
A
apps1
R&D as % of total revenue 20222
Sources:
1. Meta Q4 2022 Earnings Call
2. Q4 2022 Earnings Call from Alphabet, Amazon, Apple, Meta, Microsoft
UNLOCK AI: REALIZING THE POTENTIAL OF AI
+20%
increase in conversions
YOY and decline in
cost per acquisition
Ranking Distribution Tooling
AI foundations
Business goal Audience liquidity
FUEL FREEDOM
Diverse creative Simple account structure
Data
AI performance index CAPI/Pixel/EMQ
Learning phase
Maximize all to improve business results Catalog health (match rate, item availability)
Creative
Creative differentiation
Creative fatigue
Media
Account simplification
Liquidity (placement, audience, Reels, Budgets)
Advantage suite (ACA/ACABA)
Advantage+ shopping campaigns
Measurement
Incrementality testing (Conv lift, GeoLift, MMM)
Multi-cell experiments that measure incremental
sales (Conv lift, GeoLift)
Data marktgoods
Sponsored
01
Shop now
What is the
technology to be effective
META PIXEL
WEBSITE
Conversions API?
INTERACTION
OTHER
INTERACTION
*Based on analysis of >1K global large advertisers with direct, **Based on analysis of >1K global large advertisers with ***The meta analysis is based on the Global Conversion API
Conversions API Gateway or partner integrations (partner direct, Conversions API Gateway or partner integrations based Lift studies (with 90+% confidence) ran between
integrations only included if >1000 integrations exist) between (partner integrations only included if >1000 integrations 2022-02-05 to 2022-05-03 with different levels of EMQ
2022-01-01 and 2022-03-31 with criteria of at least 50 server exist) between 2022-01-01 and 2022-03-31 with criteria scores. The difference in incrementality share was
and browser events, EMQ score > 5, server coverage of > 90% of at least 50 server and browser events, EMQ score > 5, observed between an EMQ score of 5.0 to 7.0 for
(but no higher than 300% to eliminate outliers) Source: Meta server coverage of > 90% (but no higher than 300% to advertisers who are passing subscription events with
case study eliminate outliers) Source: Meta case study Conversions APISource: Meta case study
DATA - LEARNING PHASE
system through a
consolidated and streamlined
EXPECTED CONVERSION
campaign setup. Learning phase Optimised
delivery
• During the learning phase, the
delivery system is exploring
the best way to deliver your
RATE
ad set
NUMBER OF CONVERSIONS
As of 6/18/2019
DATA - LEARNING PHASE
19% LEARNING
On average, advertisers saw a 19% lower CPA for ad sets that successfully exited the
learning phase (moved from learning to active), versus those that failed to exit the
learning phase (were learning limited).
*Analysis excludes smaller tail-end advertisers who are not eligible to see spend in learning details, and uses only purchase optimised activity from October 2022. Based on median within advertiser improvements.
DATA - CATALOG HEALTH
90%
or higher indicates a high match rate if you have a
Effective use Increased Improved ad single catalog connected to the event source*.
of signals audience size performance
• *As recommended by Meta
Business Engineering
Creative Golden matte
eyeshadow
02
Blue hand-dyed knit
sweater
Creative differentiation
Creative fatigue
CREATIVE
Product adoption
1 (new formats, placements, mediums)
Creative concepts
2 (visual messaging content, style)
Creative differentiation
has been shown to drive
32%
9%
incremental reach
increased efficiency
Right stat: We found that the more visually distinct the creatives,** the higher the performance lift of diverse creatives over identical creatives. We saw that the lift in performance (in
test vs control) was 32% when the creatives in the diverse cell were most visually distinct,** compared to 2% when the creatives in the diverse cell were most visually similar.**
These results were derived from a scaled backend test comparing the performance of two identical vs two non-identical (creatively diverse) creatives in 2703 ad sets across verticals
from 04-04-2021 to 04-26-2021. This analysis was limited to static ads only.
Secondary
Easy wins Increased efficiency Incremental reach business goals
● Automatic ● Dynamic Ads ● Video/in-stream ● Click to Messenger
placements ● Commerce product ● Branded content with ● Lead Ads
● Placement tagging handshake
customization ● Multiple asset types ● On-platform
● Dynamic creative ● Instant Experiences engagement audiences
optimization ● Messenger and lead
● Dynamic language gen
optimization ● Cross-border business
Product adoption
Same format, new and improved message
CREATIVE
Avoid
Easy wins Increased efficiency Incremental reach
creative fatigue
● Frequency - How
● Vertical/square ● Mixed asset types ● Dissimilar concepts often to deploy new
(over horizontal) ● Messaging matrix ● Diverse people creative
● Feature brand ● Custom overlays ● Optimizing across ● Where - Which
upfront ● Promotion testing the funnel campaign to deploy
● Showcase the ● Community ● Influencer strategies new creative
product or service connections ● Strategy - How to
early rotate new vs. existing
● Video + static mix assets
Product adoption
Same great message, new and improved format
CREATIVE
10
Stats overview:
8
Our analysis found that greater levels of fatigue
were predictive of stronger CPA improvement from 6
● Won’t build clear future learnings ● Splits audience to eliminate cross ● Holdout group is mandatory
(which ad is best and by how much?) contamination ● Would need enough test power to see
● Full credit given to only one ad which Cons significant results
can lead to cross contamination ● Generally requires more budget
● Tests are limited to campaigns with
the same attribution setting
Media
Account simplification
Liquidity (placement, reels, automation)
Advantage suite (ACA/ACABA)
Advantage+ shopping campaigns
MEDIA- ACCOUNT SIMPLIFICATION
Account structure
Adopt the simplest campaign structure to offer the flexibility needed for the system to learn efficiently
STRUCTURE 1 STRUCTURE 2
Total weekly
Max number budget
AD of =
ad sets Cost per acquisition
X 50
Not recommended Preferred
MEDIA - ACCOUNT SIMPLIFICATION
Account simplification
Utilize broad targeting which
simplifies an account structure
and drives a markt
5%
Sponsored
Better median
cost per purchase
Source: Based on a Meta meta-analysis of 10 split tests run from December 2021 – January 2022
MEDIA – LIQUIDITY (PLACEMENTS)
73% probability.
INSTAGRAM
REELS MESSENGER
STORIES
Source: Based on historical results from Brand and Conversion lift studies
conducted between October 2020 and September 2021.
MEDIA – LIQUIDITY (REELS)
Source: 1. Page visits results are based on 12 lift studies, add to cart results are based on 13 lift studies, and purchase results are
based on 11 lift studies. All studies were run from June 2022 to December 2022 by global advertisers from various verticals
including Ecommerce, CPG, Retail, and Professional Services. Business-as-usual campaigns means Facebook Feed, Instagram
Feed and Instagram Stories. 2. We conducted a 2-week long large-scale study measuring incremental effects with 14k global
advertisers across multiple verticals in Q3 2022. Results showed that adding ads that were opt-in to FB Reels placement
delivered 3% more incremental purchases per dollar spent compared to not opt-in to FB Reels placement. To minimize skew,
we cap conversion counts at 5 for each user for any given ad account in the test. The result is statistically significant at 90%
MEDIA – LIQUIDITY (REELS)
Sources: 1 Nielse custom MMM meta-analysis commissioned by Meta ‘Global CPG Execution Best Practce to Maximise Meta
Performance, Nielsen Metrics That Matter’ 119 CPG Brands, 2019-2021 2. On average across 12 Disruptors clients tests June
2022
MEDIA – LIQUIDITY (AUTOMATION)
Automation
Use automation to enhance various aspects of your campaign.
Advantage+ campaign Audience automation solutions Advantage+ creative and Advantage+ placements
budget helps distribute help reach more people likely to Advantage+ creative for enables you to take
your budget to your best convert. catalog help personalise advantage of the lowest-
performing ad sets. creatives at scale. cost auctions across all
Advantage detailed targeting
placements.
Advantage lookalike
Advantage custom audience
Jasper’s Market
Sponsored
Ceramic hand-
painted mugs Polka dot mug
$45
Source: 1. 10 lift studies in NA, LATAM and EMEA, from May-July 2021, to test whether Advantage detailed targeting was more cost-effective on conversions than not utilizing Advantage detailed targeting. 2. 14 cross-vertical AB tests in NA, EMEA, and
APAC, from April-May 2020, to test whether Advantage Lookalike drives more cost-effective purchase conversions than BAU Lookalike Targeting. 3. Meta global budget-controlled experiment of 589 campaigns randomized at the user level and conducted
in April 2022. For both Product Catalog Sales and App Installs, a statistical simulation framework gave a 99.9% statistical confidence that Advantage Custom Audience outperformed Custom Audience for driving cost-effective conversions. Given the scale
of this study, outlier data points will occur and were retained and accounted for through rule-based methods and by using the median to describe results.
MEDIA - ADVANTAGE SUITE (ACA & ACABA)
Broad audiences
Your expanded audiences
Advantage+
catalog ads help you
connect with people
likely to purchase. Retargeting audiences
Your existing audience
MEDIA - ADVANTAGE SUITE (ACA & ACABA)
17% 32%
improvement in cost increase in return
Creative Destination
per action on ad spend
04
Knitted yellow
sweater
Source : Beyond the Cookies: Mapping the Future of Marketing Measurement, IAB Australia
MEASUREMENT
First-party
experiments On average, last click
attribution undervalues
Facebook by
56%
More effective
Geo-based
budget allocation
testing
across channels
*Granularity and accuracy of digital attribution and Ads Manager reporting will continue to decrease over time with further aggregation and delaying of reporting due to increased privacy
MEASUREMENT
Marketing Mix Modeling
helps fill the gaps left by
attribution (Opt-outs, SANs,
non-digital cross-publishers,
non-media marketing levers,
project ROAS, LTV)
Some advanced
marketers use all MMM ATTRIBUTION
Validate non-incremental
Each solution can Calibrate models to ensure
model like rule-based
attribution. Calibrate
they align with ground-truth.
supplement another incremental attribution such
EXPERIMENTS as Statistical MTA.
Validation of
observations using "gold
standard" methodology.
MEASUREMENT-IMPORTANCE OF EXPERIMENTATION
Why is experimentation
important?
30%
higher ad performance that year.
Source: Runge, Julian and Geinitz, Steven and Ejdemyr, Simon, Experimentation and Performance in
Advertising: An Observational Survey of Firm Practices on Facebook (May 23, 2020). Expert Systems
with Applications, Forthcoming, Available at SSRN: https://ssrn.com/abstract=3608998
MEASUREMENT - INCREMENTALITY
Last-click attribution:
Only the last click-through touchpoint will
get credit for the conversion
(Last-click)
No ad impression
1. 2. 3. 4. 5.
Identify the audience Meta randomizes your Ads are delivered to You pass your conversion Meta compares the
you want to target and audience, then separates the test group data to Meta via the Meta conversions in the test
the business objective people into test and pixel(s), a secure data and control groups to
you want to measure control groups upload or app events calculate lift, then posts
results in Ads Manager
MEASUREMENT - INCREMENTALITY
lift study
For both options: How does this compare to my current attribution reporting of Meta’s impact?
MEASUREMENT - INCREMENTALITY
Why GeoLift?
It’s grounded in incrementality, arguably the best
indicator of return on investment.
Gardening Glo…
It’s open source. You can use it freely, build on top of it
$10.00
and replicate results.
MEASUREMENT - INCREMENTALITY
GeoLift enables omnichannel GeoLift allows you to GeoLift allows you to calibrate
strategies and measurement identify which channel and optimize campaigns based
with apples-to-apples provides the best on the true return on
comparisons throughout incremental value and act on investment of every effort.
your entire business. that information.
MEASUREMENT - INCREMENTALITY
1. 2. 3. 4. 5.
Identify your business Choose test-and-control Run your campaign—ads Collect conversion data from Calculate lift by comparing
objective and campaign geographic areas per the will be delivered to the the geographic areas conversions in the test areas
to measure pre-campaign matching test areas and will not be to conversions in the control
process based on your delivered to the areas
objective control areas
Q&A
Next Week
Signals & AI – 31/05