Our CEO, Bart Muskala, presented at the CommPass symposium. He discussed the challenges we came across moving from single touch attribution to multi touch. Proving drive to store impact for multiple as opposed to one channel is quite challenging - the slides document the steps we took and are still taking to further refine our model.
4. Hi, I’m Bart.
Nice to meet you.
car via E40 coffee - passed 7 billboards
👋 HI THERE!
5. We help improve media decision making
via real-world consumer insights
5
6. visit
City worker
Carrefour customer
Car commuter
Has kids
Heavy fitness user
Received door-to-door
Billboard interaction
Turning location data of
160K consumers
into media insights
home / work interaction
7. Understand consumer
behaviour & their journeys
Evaluate long term impact
& category benchmarking
Drive-to-store impact &
improve media mix
Solving many typical use cases for media
DIFFERENT MEDIA CHALLENGES ARE ANSWERED VIA ACCURAT’S INSIGHTS
11. Did the campaign generate incremental visits?
BILLBOARD
CAMPAIGN
WAS THERE AN UPLIFT IN VISITS?
All visits to Quick
incremental visits?
12. Is it that easy?
BILLBOARD
CAMPAIGN
IGNORING IMPACTING FACTORS
All visits to Quick
13. In comes exposed vs control audiences
MAKING AN OFFLINE MEDIUM DATADRIVEN
14. In comes exposed vs control audiences
IMPACTING FACTORS
Control group visits to Quick
Exposed group visits to Quick
All visits to Quick
15. Historic visits help understand influencing factors
IF CONTROL AUDIENCE VISITS GROW, THERE ARE INFLUENCERS AT WORK
Control group visits to Quick
Exposed group visits to Quick
+10%
Predicted visits
All visits to Quick
16. As a result: the bar is raised for expected visits
THE BAR IS RAISED AT A SIMILAR LEVEL
Control group visits to Quick
Exposed group visits to Quick
+10%
+10%
All visits to Quick
17. Incremental visits can be attributed to the medium
A SMART MODEL IS BORN...
All visits to Quick
Control group visits to Quick
Exposed group visits to Quick
incremental visits!
19. Incremental visits = total exposed visits - corrections
ACTUAL CONTROL GROUP VISITS x HISTORIC CONTROL GROUP BEHAVIOUR
The result: a simple & smart attribution model
21. Not all media are addressable
Not all interactions are equal
Exponential growth of control audiences
Different metrics per medium
The order of touchpoints
The wear-out effect
Different messages per touchpoint
Local context variables
Non-measurable context
Ecosystem incompatibilities
Multiple channels
introduce complexity
on different levels
22. Exponential growth for control audiences
Data-driven
control groups
YET FEASIBLE PROVIDED THERE ARE SUFFICIENT RESPONDENTS
exposed
control
1 control
group
TP 1 TP 2 TP 1 TP 2 TP 3
Touchpoint 1
3 control
groups
7 control
groups
23. Not all media are addressable
Probabilistic
modelling:
potential
interaction
/zone
YET INSIGHTS ARE AVAILABLE
24. Letterboxes
Reach/zone
Different metrics per medium
GRP
Reach & Frequency
CTR
Impressions & Clicks
Readers
Reach
YET ALL CAN DERIVED TO A PROBABLE INTERACTION
Normalise to
‘Interactions’:
Banner view,
banner click,
radio ad
interaction,
folder read...
25. The order of touchpoints
Keep track of
interactions:
timestamps
YET POTENTIAL TIMESTAMPS CAN BE INCLUDED
26. Post-campaign WK1
The wear-out effect of media
ML
modelling of
time decay:
more data
will generate
better insights
YET MODELS CAN BE CONSTRUCTED OR GENERATED BASED ON RESEARCH
Campaign WK1 Campaign WK2 Post-campaign WK1
29. HOW THE IMPROVED MODEL WORKS
Let’s start from your media mix
Panels used
Budget
Campaign period
Interactions
Budget
NIS/BD-zones
Validity date
Budget
GRP & Reach
Budget
Campaign period
GRP & Reach
Budget
Campaign period
Impressions & Clicks
Budget
Zones
aggregated
granular
The system is fed
with touchpoint input
(when, what, where, budget…)
30. HOW THE IMPROVED MODEL WORKS
Over to our panel to define exposed audiences
We start from our
real-world panel of
160.000 individuals
Panels used
Budget
Campaign period
Interactions
Budget
NIS/BD-zones
Validity date
Budget
GRP & Reach
Budget
Campaign period
GRP & Reach
Budget
Campaign period
Impressions & Clicks
Budget
Zones
aggregated
granular
31. HOW THE IMPROVED MODEL WORKS
First up: trackable interactions
We plot each
trackable interaction
when it happens
Panels used
Budget
Campaign period
Interactions
Budget
NIS/BD-zones
Validity date
Budget
GRP & Reach
Budget
Campaign period
GRP & Reach
Budget
Campaign period
Impressions & Clicks
Budget
Zones
aggregated
granular
WK1 WK2 WK3
32. HOW THE IMPROVED MODEL WORKS
Probabilistic modelling of other touchpoints
Panels used
Budget
Campaign period
Interactions
Budget
NIS/BD-zones
Validity date
Budget
GRP & Reach
Budget
Campaign period
GRP & Reach
Budget
Campaign period
Impressions & Clicks
Budget
Zones
aggregated
granular
WK1 WK2 WK3
We model
non-addressable
media for location 8.477 unique users
84.774 inhabitants
8.477
84.774
= 10%
probability
33. WK1 WK2 WK3
HOW THE IMPROVED MODEL WORKS
Defining expected behaviour via historic data
For each individual we
know their historic &
current store visits
34. WK1 WK2 WK3
Correcting results for time decay & others
HOW THE IMPROVED MODEL WORKS
Interactions are
corrected for time
decay & others
35. WK1 WK2 WK3
HOW THE IMPROVED MODEL WORKS
Measuring campaign impact per medium
The control & exposed
model calculates
impact of each medium
36. Incremental visits = total exposed visits - corrections
CONTROL GROUP x HISTORIC BEHAVIOUR AS THE BASIS REMAINS
A refined model, on top of its base concept
vastly improved
38. A summary of
reach, impact
(e.g. lift,
incremental
visits) and
ROAS metrics
Let’s top it of with a case
A MULTITOUCH ANALYSIS FOR A REALWORLD CAMPAIGN
39. Detailed
insights for
reach & the
selectivity for
different
profiles of your
visitors
Let’s top it of with a case
A MULTITOUCH ANALYSIS FOR A REALWORLD CAMPAIGN
43. The future is data driven
2020 2021
WEATHER CAMPAIGNS ROAD INFO
HISTORIC DATA CURRENT DATA
Day by day
WEATHER CAMPAIGNS ROAD INFO
TO UNDERSTAND HISTORIC IMPACT TO PREDICT FUTURE IMPACT
BETTER EVERY DAY
MEASURED STORE TRAFFIC PREDICTED STORE TRAFFIC
44. Making more channels data-driven
BETTER EVERY DAY
...and more to be announced soon