From Digital Attribution to MMM
#MeasureCampNL – April 21, 2018
@mertanen
Using machine
learning for
explaining and
predicting user
behavior from web
analytics data.
2
Traditional Web
Analytics is dying...
Optimizing Promotions
4
Cookie-based 1:1 attribution produces superior
understanding of the conversion journey
Tradional measurement:
Customer journey:
Time Value
These clicks are ignored
Time Value
The last click gets all the credit
Assisted conversions
Attribution model comparison
Cookie-based 1:1 attribution produces superior
understanding of the conversion journey
Algorithmic attribution:
Customer journey:
Time Value
Model estimates value of each touchpoint
ValueTime
Attribution model comparison
Organic search Paid search Email Social media
Total conversion value - example
First touch
Last touch
Linear
Advanced Attribution
The 4-7 Ps of Marketing
10
The total effect of marketing actions over time
Direct Buy
Short Term Effect
Create Curiosity,
Seek online info
Intermediate
Effect
Change attitude
towards brand
Long Term Effect
Time
Total Sales Effect
+ + =
Campaign
Digital Analytics Digital Analytics + Research Marketing Mix Modelling
What is MMM?
Marketing Mix Modelling
Media Mix Modelling
Econometric Modelling
Media Attribution Modelling
It is used to understand all drivers of business
performance Offline Investment
Brand image
Market Dynamics
Operational factors
Digital media
TV media
Pricing / Promotions
PR
Competitors
Business
KPI
SALES!
Typical Data Requirements
KPI
Sales / Bookings / Quotes
Web visits / calls
Regional
Store level
By source (e.g. retail vs online sales)
Internal
Pricing
Promotions
Store level info
Launches / product changes
Retention
Media
Offline media: spend/ratings by
channel
Online media:
Impressions/clicks/spend by channel
Social media: Facebook and Twitter;
Likes/Clicks/Impressions
Media spend for all competitors by
channel
DM / Email etc volumes
Other
Competitor info
Weather
CCI, RPI, Inflation etc.
Legislation
Brand metrics
Data needs to cover a
wide range of factors
Data needs to be over
a long enough period
Data needs to be
weekly for best results
Modelling projects follow a similar process
Insights
Debrief writing
Results
Delivery
Kick Off
Data
Collection
Data
Processing
Analysis /
Modelling
Pulling
together
results
Several months
Typical questions MMM can answer
How effective is our
marketing and
advertising
investment overall?
How important is
pricing to my sales
and what is the likely
impact of future
changes I might
make?
Do I need to be
‘always on’ or are
there seasons when
investment is more
effective?
How are my
promotions
performing and which
ones deliver the most
revenue?
Which communication
channels are most
effective? Return on
Investment?
How do current
events and market
dynamics impact my
KPIs?
How does competitive
media impact my
business?
Should I spend more?
How many more
sales would that give
me?
Sales drivers - weekly basis
What is exactly causing the sales?
Degree of explanatory power: 58%
-1,000
-500
0
500
1,000
1,500
2,000
2010-1
5
9
13
17
21
25
29
33
37
41
45
49
2011-1
5
9
13
17
21
25
29
33
37
41
45
49
SALES
Base Seasonality Distribution Price TV Sales
88% 8% 8% 6% 2% 0% -1% -1% -10%
857.447
79.646
75.566
61.520 22.250
-0 -6.000 -13.366
-102.223
974,841
0
200,000
400,000
600,000
800,000
1,000,000
1,200,000
BASE Price & Promotion Tracking Media Distribution &
Launches
Unexplained Seasonality Competitor - Media Competitor - Price
& Promotion
Total
Revenue('000)
Promotions are the
biggest driver of sales.
Negative impact from
price increases
Media investments
drive almost as
much sales as
promotions
Successful product
launches have
grown the brand
Competitor price
reductions and
promotions have
had the biggest
negative impact
Sales drivers - weekly basis
What is exactly causing the sales?
Optimization recommendations
Changes in media allocations
TV Spot
Web-TV
Desktop Display
Mobile Display
Programmatic Display
Programmatic Mobile
Facebook
Chart Title
2015 budget
2016 optimized budget
Media Total effect
TV
ROI:
Sales effect:
3,3  3,7
+2 500 000 €
+20%
-37%
+20%
+20%
Change %Investment level
Mobil
Display
Web-TV
Programmatic
Mobil
Programmatic
Display
Facebook
-100%
+20%
-100%
000’
What kind of results you can expect?
+6%
ROI/CPA
10 % of clients are
making small
changes on media.
+12%
ROI / CPA
50 % of clients are optimizing media
mix and timings. They get benefits
from media synergies.
+28%
ROI / CPA
40 % of clients makes significant
strategic decisions regarding to
marketing actions and media budgets.
Pros & Cons of MMM
What’s good
• We can quantify objectively & fairly
- What is driving our KPI
- When it affects sales (short/medium term)
- How much it affects sales
- Factors outside of your control
• We can provide a starting point
- For strategic planning purposes
- For investment allocation
• Explains and quantifies your historic KPI trend
What’s difficult
• Data Limitations
- Incorrect/missing data
- Lack of variation
- Multi-collinearity: all media happens at the
same time
- Long data series required
• We can’t always tell you a positive story
• We can’t tell you the impact of something you
haven’t done before
- Or the impact of activity beyond levels of
experience
• Long project timeframe
Contact information
Petri Mertanen
Director, Digital Analytics
petri.mertanen@annalect.com
Mobile: +358 400 792 616
@mertanen

From Digital Attribution to Marketing Mix Modelling

  • 1.
    From Digital Attributionto MMM #MeasureCampNL – April 21, 2018 @mertanen
  • 2.
    Using machine learning for explainingand predicting user behavior from web analytics data. 2
  • 3.
  • 4.
  • 5.
    Cookie-based 1:1 attributionproduces superior understanding of the conversion journey Tradional measurement: Customer journey: Time Value These clicks are ignored Time Value The last click gets all the credit
  • 6.
  • 7.
  • 8.
    Cookie-based 1:1 attributionproduces superior understanding of the conversion journey Algorithmic attribution: Customer journey: Time Value Model estimates value of each touchpoint ValueTime
  • 9.
    Attribution model comparison Organicsearch Paid search Email Social media Total conversion value - example First touch Last touch Linear Advanced Attribution
  • 10.
    The 4-7 Psof Marketing 10
  • 11.
    The total effectof marketing actions over time Direct Buy Short Term Effect Create Curiosity, Seek online info Intermediate Effect Change attitude towards brand Long Term Effect Time Total Sales Effect + + = Campaign Digital Analytics Digital Analytics + Research Marketing Mix Modelling
  • 12.
    What is MMM? MarketingMix Modelling Media Mix Modelling Econometric Modelling Media Attribution Modelling
  • 13.
    It is usedto understand all drivers of business performance Offline Investment Brand image Market Dynamics Operational factors Digital media TV media Pricing / Promotions PR Competitors Business KPI SALES!
  • 14.
    Typical Data Requirements KPI Sales/ Bookings / Quotes Web visits / calls Regional Store level By source (e.g. retail vs online sales) Internal Pricing Promotions Store level info Launches / product changes Retention Media Offline media: spend/ratings by channel Online media: Impressions/clicks/spend by channel Social media: Facebook and Twitter; Likes/Clicks/Impressions Media spend for all competitors by channel DM / Email etc volumes Other Competitor info Weather CCI, RPI, Inflation etc. Legislation Brand metrics Data needs to cover a wide range of factors Data needs to be over a long enough period Data needs to be weekly for best results
  • 15.
    Modelling projects followa similar process Insights Debrief writing Results Delivery Kick Off Data Collection Data Processing Analysis / Modelling Pulling together results Several months
  • 16.
    Typical questions MMMcan answer How effective is our marketing and advertising investment overall? How important is pricing to my sales and what is the likely impact of future changes I might make? Do I need to be ‘always on’ or are there seasons when investment is more effective? How are my promotions performing and which ones deliver the most revenue? Which communication channels are most effective? Return on Investment? How do current events and market dynamics impact my KPIs? How does competitive media impact my business? Should I spend more? How many more sales would that give me?
  • 17.
    Sales drivers -weekly basis What is exactly causing the sales? Degree of explanatory power: 58% -1,000 -500 0 500 1,000 1,500 2,000 2010-1 5 9 13 17 21 25 29 33 37 41 45 49 2011-1 5 9 13 17 21 25 29 33 37 41 45 49 SALES Base Seasonality Distribution Price TV Sales
  • 18.
    88% 8% 8%6% 2% 0% -1% -1% -10% 857.447 79.646 75.566 61.520 22.250 -0 -6.000 -13.366 -102.223 974,841 0 200,000 400,000 600,000 800,000 1,000,000 1,200,000 BASE Price & Promotion Tracking Media Distribution & Launches Unexplained Seasonality Competitor - Media Competitor - Price & Promotion Total Revenue('000) Promotions are the biggest driver of sales. Negative impact from price increases Media investments drive almost as much sales as promotions Successful product launches have grown the brand Competitor price reductions and promotions have had the biggest negative impact Sales drivers - weekly basis What is exactly causing the sales?
  • 19.
    Optimization recommendations Changes inmedia allocations TV Spot Web-TV Desktop Display Mobile Display Programmatic Display Programmatic Mobile Facebook Chart Title 2015 budget 2016 optimized budget Media Total effect TV ROI: Sales effect: 3,3  3,7 +2 500 000 € +20% -37% +20% +20% Change %Investment level Mobil Display Web-TV Programmatic Mobil Programmatic Display Facebook -100% +20% -100% 000’
  • 20.
    What kind ofresults you can expect? +6% ROI/CPA 10 % of clients are making small changes on media. +12% ROI / CPA 50 % of clients are optimizing media mix and timings. They get benefits from media synergies. +28% ROI / CPA 40 % of clients makes significant strategic decisions regarding to marketing actions and media budgets.
  • 21.
    Pros & Consof MMM What’s good • We can quantify objectively & fairly - What is driving our KPI - When it affects sales (short/medium term) - How much it affects sales - Factors outside of your control • We can provide a starting point - For strategic planning purposes - For investment allocation • Explains and quantifies your historic KPI trend What’s difficult • Data Limitations - Incorrect/missing data - Lack of variation - Multi-collinearity: all media happens at the same time - Long data series required • We can’t always tell you a positive story • We can’t tell you the impact of something you haven’t done before - Or the impact of activity beyond levels of experience • Long project timeframe
  • 22.
    Contact information Petri Mertanen Director,Digital Analytics petri.mertanen@annalect.com Mobile: +358 400 792 616 @mertanen