Attribution or Misattribution?
MAY 2018
2
Attribution or Misattribution?
How to make digital measurement less awful
INTRODUCTION TO EBIQUITY
3
Independent consultancy helping clients make better
Media and Marketing investment decisions
IN THE ROOM TODAY
4
Ebiquity Effectiveness: Our expertise lies in
measuring and improving ROI. We are commercial first!
Not an attribution
specialist
Nic Pietersma
Business Director
5
1. VALIDATE AND VERIFY
VALIDATE AND VERIFY
6
We need one version of the truth
The most harmful words…
GEOTESTING
7
Every attribution manager should consider geotesting
to validate the results coming out of their system
Ratings/Adspend
RegionalQuotes(Indexed)
Ratings/Adspend
Media Upweight Regions (Test Index)
Media Downweight Regions (Control Index)
GEOTESTING
8
This is what a successful geotest looks like; the size of the gap
can be quantified to calculate an ROI
Testing PeriodIndexing Period
Re-Indexing
Period
Quantify Incremental BenefitEstablish Regional “Match” Establish “Tail”
Ratings/Adspend
RegionalQuotes(Indexed)
Ratings/Adspend
Media Upweight Regions (Test Index)
Media Downweight Regions (Control Index)
GEOTESTING
9
If you have anchored your ROI expectations exclusively on digital
attribution, you may be in for a surprise…
Testing PeriodIndexing Period
Re-Indexing
Period
Quantify Incremental BenefitEstablish Regional “Match” Establish “Tail”
VALIDATION RESULTS
10
In our experience, some media lines (e.g. digital display)
consistently look worse from a geotesting perspective
First Touch Last Touch Machine Learning Geo Test
Conversions By Measurement Method
Machine Learning Geo Test
Zooming In…
PLACEBO TESTING / WHITE SPACE TESTING / AD SWAP TESTING
11
Placebo testing is a way of validating the tracking
framework by tagging “as if” advertising is your product
We have seen placebo tests in the
past where the client creative
works no better than the placebo!
Verification Tools
• Geotesting
• Econometrics
• Matchback Analysis
• Placebo Testing
• Etc.
Remedies
• Change attribution window
• Only feed viewable impressions into ML
• Use a multiplier ( x 0.2) to calibrate
• Only use attribution for “within channel”
budget decisions
We do not need to blindly trust attribution technology, there
are many tools that can help us verify the numbers
VALIDATION OPTIONS
12
13
2. OPTIMISE THE RIGHT THING
OPTIMISATION KPI
14
Sales Revenue by Channel
In Store E-commerce
Realtime machine learning optimisation sounds great…
but is it right for a business that mostly sells offline?
This business might be better off buying
campaigns on the basis of cost effective
reach rather than optimising to conversion
OPTIMISATION BUDGET
15
Media Budget
Online in Attribution Online Not In Attribtion Offline
Attribution may also ‘see’ less of your marketing budget than
you think (esp. if you factor in platforms that do not allow tagging)
An attribution framework only helps
measure a small share of this client’s
marketing budget
16
3. DO THE MATHS
DO THE MATHS
17
£12.52
£1.14
Digital Radio Regular Radio
Cost Per Thousand
Attribution ROI needs to be sense checked against media costs;
how plausible is your ROI estimate in light of big cost premiums?
£29
£6
VOD TV
x 10x 4.9
DO THE MATHS
18
53%
73%
88%
94% 97% 99% 100% 100% 100% 100%
14%
25%
41%
53%
67%
80%
89%
95% 98% 100%
<= 1 <= 2 <= 5 <= 10 <= 20 <= 40 <= 80 <= 160 <= 320 <= 3285
%ofDevices/Impressions
Cumulative Devices Cumulative Impressions
VOD: Your realtime optimised campaign might not look great
from a reach and frequency perspective!
THE PROMISE: “The complexity of our optimisation methodology… has thousands of possible
attributes with literally billions of potential combinations from multiple data partners”
DO THE MATHS
19
If 90% of your volume has a single touchpoint, you may want to
reconsider what role attribution plays in your decision making!
We often find that
media lines such as PPC
and Affiliates have only
one touchpoint (90%+)
20
4. DATA HYGIENE
DATA HYGIENE
21
Attribution & media platforms often don’t talk to the IT
environment that runs the rest of the rest of the business
COMMERCIAL & OPERATIONAL WORLD
Transaction ID
Transaction Value
Basket Details / SKUs
Conversion Rate
Customer ID
Customer Age
Customer Gender
Customer Purchase History
Delivery Address
DIGITAL MEDIA WORLD
Affiliate Reference ID
First Touch Source
Last Touch Source
Campaign ID
Checkout Page ID
Campaigns that are not tagged in the first place cannot feed
into any useful analytics, let alone cutting-edge attribution
DATA HYGIENE
22
Someone failed…
TIME SERIES TAGGING DATA
DATA HYGIENE
23
Is the product classification & hierarchy used in the
media tagging sensible?
DISPLAY ADVERT FOR
CORDLESS DRILL
SALE OF THIS FINE
GARDEN SHED
24
5. SOCIALISE THE KNOWLEDGE
SOCIALISE THE KNOWLEDGE
25
The whole marketing team should understand how Attribution
framework is set up and how ROI is calculated
In practice, we often find
only 1 or 2 people know
how their attribution
framework works!
26
LAST WORD
Our Total View Attribution is based on the idea that businesses
have different needs and should use the right tool for the job
Digital
Online
Shareofsales
Share of media
Econometrics
+
Digital
Attribution
Digital
attribution
Digital
attribution
(indirectly)
Econometrics
27
EBIQUITY
Head Office
Citypoint, One Ropemaker Street
LONDON,
EC2Y 9AW
tel. +44 (0) 20 7650 9600
Disclaimer
This document is confidential to INSERT CLIENT NAME. Disclosure or circulation of this
document, or any part of it, may only be made to INSERT CLIENT NAME, its employees
who need to have knowledge of this matter and to its nominated agency. Unauthorised
copying, disclosure or circulation of this document (by any company or individual) represents
a breach of copyright and confidentiality. By accepting disclosure of this document you
agree to these terms and to indemnify Ebiquity if you breach any of these terms.
The results contained in this document are primarily derived from a comparison of INSERT
CLIENT NAME’s performance with the performances of Ebiquity’s other clients, whose
results are contained in Ebiquity’s databases. These databases contain the mean
performance results achieved by Ebiquity’s clients for a given format in a given period. The
results contained in these databases change over time, as new data is added to the
databases. We undertake no duty to notify or inform our clients as these databases change
over time nor to update any work previously performed by us in light of such revisions to our
databases.
30

Attribution or Misattribution? How to make digital measurement less awful - Nic Pietersma

  • 1.
  • 2.
    2 Attribution or Misattribution? Howto make digital measurement less awful
  • 3.
    INTRODUCTION TO EBIQUITY 3 Independentconsultancy helping clients make better Media and Marketing investment decisions
  • 4.
    IN THE ROOMTODAY 4 Ebiquity Effectiveness: Our expertise lies in measuring and improving ROI. We are commercial first! Not an attribution specialist Nic Pietersma Business Director
  • 5.
  • 6.
    VALIDATE AND VERIFY 6 Weneed one version of the truth The most harmful words…
  • 7.
    GEOTESTING 7 Every attribution managershould consider geotesting to validate the results coming out of their system
  • 8.
    Ratings/Adspend RegionalQuotes(Indexed) Ratings/Adspend Media Upweight Regions(Test Index) Media Downweight Regions (Control Index) GEOTESTING 8 This is what a successful geotest looks like; the size of the gap can be quantified to calculate an ROI Testing PeriodIndexing Period Re-Indexing Period Quantify Incremental BenefitEstablish Regional “Match” Establish “Tail”
  • 9.
    Ratings/Adspend RegionalQuotes(Indexed) Ratings/Adspend Media Upweight Regions(Test Index) Media Downweight Regions (Control Index) GEOTESTING 9 If you have anchored your ROI expectations exclusively on digital attribution, you may be in for a surprise… Testing PeriodIndexing Period Re-Indexing Period Quantify Incremental BenefitEstablish Regional “Match” Establish “Tail”
  • 10.
    VALIDATION RESULTS 10 In ourexperience, some media lines (e.g. digital display) consistently look worse from a geotesting perspective First Touch Last Touch Machine Learning Geo Test Conversions By Measurement Method Machine Learning Geo Test Zooming In…
  • 11.
    PLACEBO TESTING /WHITE SPACE TESTING / AD SWAP TESTING 11 Placebo testing is a way of validating the tracking framework by tagging “as if” advertising is your product We have seen placebo tests in the past where the client creative works no better than the placebo!
  • 12.
    Verification Tools • Geotesting •Econometrics • Matchback Analysis • Placebo Testing • Etc. Remedies • Change attribution window • Only feed viewable impressions into ML • Use a multiplier ( x 0.2) to calibrate • Only use attribution for “within channel” budget decisions We do not need to blindly trust attribution technology, there are many tools that can help us verify the numbers VALIDATION OPTIONS 12
  • 13.
    13 2. OPTIMISE THERIGHT THING
  • 14.
    OPTIMISATION KPI 14 Sales Revenueby Channel In Store E-commerce Realtime machine learning optimisation sounds great… but is it right for a business that mostly sells offline? This business might be better off buying campaigns on the basis of cost effective reach rather than optimising to conversion
  • 15.
    OPTIMISATION BUDGET 15 Media Budget Onlinein Attribution Online Not In Attribtion Offline Attribution may also ‘see’ less of your marketing budget than you think (esp. if you factor in platforms that do not allow tagging) An attribution framework only helps measure a small share of this client’s marketing budget
  • 16.
  • 17.
    DO THE MATHS 17 £12.52 £1.14 DigitalRadio Regular Radio Cost Per Thousand Attribution ROI needs to be sense checked against media costs; how plausible is your ROI estimate in light of big cost premiums? £29 £6 VOD TV x 10x 4.9
  • 18.
    DO THE MATHS 18 53% 73% 88% 94%97% 99% 100% 100% 100% 100% 14% 25% 41% 53% 67% 80% 89% 95% 98% 100% <= 1 <= 2 <= 5 <= 10 <= 20 <= 40 <= 80 <= 160 <= 320 <= 3285 %ofDevices/Impressions Cumulative Devices Cumulative Impressions VOD: Your realtime optimised campaign might not look great from a reach and frequency perspective! THE PROMISE: “The complexity of our optimisation methodology… has thousands of possible attributes with literally billions of potential combinations from multiple data partners”
  • 19.
    DO THE MATHS 19 If90% of your volume has a single touchpoint, you may want to reconsider what role attribution plays in your decision making! We often find that media lines such as PPC and Affiliates have only one touchpoint (90%+)
  • 20.
  • 21.
    DATA HYGIENE 21 Attribution &media platforms often don’t talk to the IT environment that runs the rest of the rest of the business COMMERCIAL & OPERATIONAL WORLD Transaction ID Transaction Value Basket Details / SKUs Conversion Rate Customer ID Customer Age Customer Gender Customer Purchase History Delivery Address DIGITAL MEDIA WORLD Affiliate Reference ID First Touch Source Last Touch Source Campaign ID Checkout Page ID
  • 22.
    Campaigns that arenot tagged in the first place cannot feed into any useful analytics, let alone cutting-edge attribution DATA HYGIENE 22 Someone failed… TIME SERIES TAGGING DATA
  • 23.
    DATA HYGIENE 23 Is theproduct classification & hierarchy used in the media tagging sensible? DISPLAY ADVERT FOR CORDLESS DRILL SALE OF THIS FINE GARDEN SHED
  • 24.
  • 25.
    SOCIALISE THE KNOWLEDGE 25 Thewhole marketing team should understand how Attribution framework is set up and how ROI is calculated In practice, we often find only 1 or 2 people know how their attribution framework works!
  • 26.
  • 27.
    Our Total ViewAttribution is based on the idea that businesses have different needs and should use the right tool for the job Digital Online Shareofsales Share of media Econometrics + Digital Attribution Digital attribution Digital attribution (indirectly) Econometrics 27
  • 28.
    EBIQUITY Head Office Citypoint, OneRopemaker Street LONDON, EC2Y 9AW tel. +44 (0) 20 7650 9600 Disclaimer This document is confidential to INSERT CLIENT NAME. Disclosure or circulation of this document, or any part of it, may only be made to INSERT CLIENT NAME, its employees who need to have knowledge of this matter and to its nominated agency. Unauthorised copying, disclosure or circulation of this document (by any company or individual) represents a breach of copyright and confidentiality. By accepting disclosure of this document you agree to these terms and to indemnify Ebiquity if you breach any of these terms. The results contained in this document are primarily derived from a comparison of INSERT CLIENT NAME’s performance with the performances of Ebiquity’s other clients, whose results are contained in Ebiquity’s databases. These databases contain the mean performance results achieved by Ebiquity’s clients for a given format in a given period. The results contained in these databases change over time, as new data is added to the databases. We undertake no duty to notify or inform our clients as these databases change over time nor to update any work previously performed by us in light of such revisions to our databases. 30

Editor's Notes