In this presentation, Andrew Covato talks about the uses of attribution modelling and Big Data within a large company like Google. Covato introduces himself and talks about his own working background before going on to talk about the ins and outs of marketing and digital marketing. As well as giving insight into attribution modelling with his experience.
Neo4j - How KGs are shaping the future of Generative AI at AWS Summit London ...
Attribution Modeling and Big Data, Google
1. Google Confidential and Proprietary 1Google Confidential and Proprietary 1
Attribution Modeling and
Big Data
Andrew Covato
June 21, 2013
2. Google Confidential and Proprietary 2Google Confidential and Proprietary 2
What’s on the menu for today?
Who am I?
Marketing 101: The Basics
Marketing 201: Digital Marketing
Whet your appetite for attribution
Conclusion: Digital marketing
today = finance in the 1980s
3. Google Confidential and Proprietary 3Google Confidential and Proprietary 3
Who am I? (i.e. am I at least somewhat credible?)
4. Google Confidential and Proprietary 4Google Confidential and Proprietary 4
Ready to go?
Let the fun
begin!
5. Google Confidential and Proprietary 5Google Confidential and Proprietary 5
The Basics
Marketing 101
6. Google Confidential and Proprietary 6Google Confidential and Proprietary 6
“Half the money I spend on advertising is wasted;
the trouble is I don’t know which half.”
--John Wanamaker: Retailer, business man, innovator,
“advertising genius”
7. Google Confidential and Proprietary 7Google Confidential and Proprietary 7
The question marketing
execs are seeking to
answer is…
8. Google Confidential and Proprietary 8Google Confidential and Proprietary 8
“What is the dollar value
of sales generated by
each of my marketing
initiatives?”
9. Google Confidential and Proprietary 9Google Confidential and Proprietary 9
“Traditional” offline marketing and advertising is
impossible to track at an un-aggregate level…
10. Google Confidential and Proprietary 10Google Confidential and Proprietary 10
MARKETING MIX MODELING
…but in the 1960s, big CPG firms developed a
framework for measuring marketing ROI
11. Google Confidential and Proprietary 11Google Confidential and Proprietary 11
The Basic Principles of Traditional Marketing Mix
Modeling (MMM)
Inputs
Controls
Response
Examples…
often other
metrics or
dimensions
included
Incremental Sales Volume
12. Google Confidential and Proprietary 12Google Confidential and Proprietary 12
MMM is great for describing macro behaviours, but
sales attribution to specific channels is dubious at
best—especially with new advertising technology
13. Google Confidential and Proprietary 13Google Confidential and Proprietary 13
Digital Marketing
Marketing 201
14. Google Confidential and Proprietary 14Google Confidential and Proprietary 14
What is this funnel you
speak of?
16. Google Confidential and Proprietary 16Google Confidential and Proprietary 16
Digital marketing allows
advertisements to align
with a specific person’s
instantaneous intent
This makes ads hyper-
contextual
18. Google Confidential and Proprietary 18Google Confidential and Proprietary 18
There are many more
dimensions by which to
accurately target an audience
at various stages in the funnel
19. Google Confidential and Proprietary 19Google Confidential and Proprietary 19
Ads (and advertising dollars) can be tailored
across a variety of such dimensions
20. Google Confidential and Proprietary 20Google Confidential and Proprietary 20
Taking bite-sized nibbles from a buffet of complexity
Whet your appetite for attribution
21. Google Confidential and Proprietary 21Google Confidential and Proprietary 21
Remember what we are
trying to quantify (ROI by
marketing channel)…now
let’s set up the problem
23. Google Confidential and Proprietary 23Google Confidential and Proprietary 23
Today, many marketers employ “last-click”, “first-
click”, or other weighting schemes to value their
channels
24. Google Confidential and Proprietary 24Google Confidential and Proprietary 24
But assigning weights will
not show you the true value
of each channel…the
weights need to be derived
using the data!
25. Google Confidential and Proprietary 25Google Confidential and Proprietary 25
Some things can
complicate this
26. Google Confidential and Proprietary 26Google Confidential and Proprietary 26
Probability of conversion
is path-dependent
People are in various
stages of awareness
when they interact with
an ad
Lots of exogenous
variables (e.g.
competitors’ response to
marketing)
27. Google Confidential and Proprietary 27Google Confidential and Proprietary 27
Technology is a further
complication—you need
very detailed data to
create a proper attribution
model!
28. Google Confidential and Proprietary 28Google Confidential and Proprietary 28
For each user, we
need a time-
stamped record of
each interaction with
each
advertisement…
29. Google Confidential and Proprietary 29Google Confidential and Proprietary 29
We also need
environmental data
(some sort of
competitive index)
30. Google Confidential and Proprietary 30Google Confidential and Proprietary 30
Oh, and we need to
develop a model to
project the users’
instantaneous
awareness state
31. Google Confidential and Proprietary 31Google Confidential and Proprietary 31
If you have all this,
however…
…you wield great power
(and responsibility)
32. Google Confidential and Proprietary 32Google Confidential and Proprietary 32
While respecting privacy, you can:
Calculate, in real-time, each user’s
instantaneous probability of converting
Determine the best creative to show a
user at 20% conversion probability to
bump her up to 80%
Understand how “upper-funnel” ads
impact sales farther down the funnel
Ultimately calculate a dead-
accurate ROAS for each digital
channel!
33. Google Confidential and Proprietary 33Google Confidential and Proprietary 33
How? Let’s get technical…
We can use a high-order Markov
model to calculate conversion
probabilities at each path
Or a first-order MM could be used to
move visitors between states
(Abhishek, Fader, Hosanagar)
Alternatively (and possibly
preferably), some sort of additive
model with a time-decay for touch-
points
34. Google Confidential and Proprietary 34Google Confidential and Proprietary 34
Who will develop the magic model?
Digital Marketing today is like Finance in
the1980s