Kia Motors Australia implemented a new digital attribution strategy using Google's DoubleClick platform to gain better insights into how their digital channels were working together. They consolidated tagging and reporting, developed a single view of the customer, and used attribution modeling to move beyond last-click attribution. This revealed that over half of conversions came from 1-2 interactions, search delivered most unassisted conversions, and display played a key role in assisted conversions. Results varied significantly by attribution model and publisher. The new strategy provided a holistic view of performance and more reliable decision making. Next steps include custom modeling and deeper analytics.
2. Don‟t worry about all the notes
Rhys Williams
Head of Media Technology
Solutions, AU/NZ, Google
@rhysmwilliams
Andrew Hughes
Director : Innovation & Technology,
Reprise Australia
@y0z2a
Shortcut to this presentation: http://goo.gl/otRUll
Questions on twitter: #atds
4. Our collective thought space
• Measurement is the foundation
• The technology is there
• We have the right smarts
• We want to increase digital investment
• A solid foundation is the solution
• Less conversation, more action
• So, where were we?
Image Credit – Frue : https://secure.flickr.com/photos/frue/2377169864/sizes/o/
5. Snapshot: Kia‟s tech journey
2012
Many
Many
Many Tags
Multiple Ad
Networks
AdTech
Ownership
Varied
Success
Metrics
Lack Of Agility
Discrepancies
Reliant on
GA
6. Snapshot: Kia‟s tech journey
2013
Digital
Measurement
Focus
Tag Management
Consolidated
50 Tags
Platform
agnostic
conversion
New CMS &
Standardisation
Mobile
Enabled
Clear Tech
Roadmap
7. 3 key problems to solve
1. Tech & Data
in silos
SearchDSP
SearchAnalytics
SearchSearch
SearchDisplay
2. Conversion
management
• Difficult to deploy
• Tag management
• Discrepancies
• Duplication
• Campaign Agility
3. Last Click
Reporting
0%
0%
0%
100%
• Display recognition
• Lack accountability
8. Many unanswered questions
This was leading to questions
• What is the average path length of the user journey?
• How often does upper-funnel activity drive last-
touch Search Conversions?
• What display channels are the most effective in
assisting other conversions?
• What are the leading paths to conversion?
10. c
Chefs don‟t use blunt knives
And the best chefs invest in their knives,
and are trained how to use them
Image Credit – Fruit Ninja
11. Consolidate measurement and
tracking in a unified platform
DFA
DBM
DFA/DBM
DS
Google
Analytics
Holistic view
of audience
and cross-channel
customer insights
True,
deduplicated
conversions
Unified
reporting for
clear attribution
Data
integrity from
a common source
Conversion
tracking
with Floodlight
Developed a single view of the customer
Site Analytics
Search
Video
DSP
Display
Confidential
12. Developed a strategy for conversions
Kia deployed a single floodlight tag
The most appropriate container tag solution
Map out all the conversion points
(take time to do this properly)
Have reporting in mind - this determines want
to track and where (natural + paid keywords,
display placements, site paths, order value)
Deploy tags and test
Confidential
13. Got the right resources onboard
And appoint a central “owner”
Project
Management
Overall vision, strategy,
implementation
13
Creative Agency
Dynamic, data driven
creative, on site tagging
implementation, CMS
Ad Technology
Consulting, Support and
implementation of
DoubleClick Digital Marketing
suite
Client
Responsible for approval
of approach, alignment to
business objectives
Search and Analytics
Ownership
Understand the meaning
and relevance of each
page in real time
Exchange Buying
RTB, DSP buying,
optimising
MBW Ad Ops
Central AdOps team ,
responsible for
implementing
Datalicious
Container tag solution
14. Ask: How do my channels work together?
Top Paths with Channel Grouping
Remarketing, Generic Search and Display consistently supported branded search
Confidential
15. Used the attribution tools available
Attribution modeling built right in to DoubleClick
16. Compare up
to three
models
simultaneously
See all your
most important
channels
Select your
baseline model
Simple metrics to
quickly evaluate
differences
between models
Moved beyond the “last click”
Compare models simultaneously
17. Choose from six built-in models in DFA
100%100%
First Click or Impression
All value assigned to first interaction
Last Click or Impression
All value assigned to the last click
20% 20% 20% 20% 20%
Linear
All value assigned evenly across clicks
0%
10%
15%
25%
50%
Time Decay
Value assigned by how close in time
click was to conversion
35%
10% 10% 10%
35%
Position-based
Value earlier and later interactions more heavily
100%
Floodlight
Use as a baseline for comparison
Based on “last-click” wins
18. Or create your own models
• Create granular modeling
rules based on interaction
type, position and time
• Instantly see the impact of
changes
• Save and manage custom
models
21. Top line results and insights; in summary
Key take-outs
• Half of all conversions came from 1 or 2 interactions
• Search delivered 87% of unassisted conversions
• Organic search contributed to 35% of unassisted conversions
• Display delivered 52% of assisted conversions
• Not all publishers are equal; Eg. 90% of one publisher‟s assisted
conversions were click-through
22. >50% of conversions are from 1-2 touches
• Path Length Report reveals the number of interactions a user had with
KIA ads before converting.
• Interactions can be impressions (i.e. views of ad) or clicks.
• An impression followed by a click is counted as 2 separate
interactions
42.60%
13.10%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
1 2 3 4 5 6 7 8 9 10 11 12+
% Conversions
23. 87% of unassisted conversions were from search
By Channel
12.10%
35.10%
51.40%
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
Display Natural Search Paid Search
% Unassisted conversions
• 51.4% of Total Paid Search Conversions were unassisted by other
channels
• i.e. 1 interaction clicked paid search ad and converted
• Natural Search drives 35.1% unassisted conversions
Note – paid search impressions are excluded from the model by default
24. 52% of assisted conversions were from display
By Channel
• Display (Image and Simple Flash) drove over 51.8% of total assisted
conversions
• Natural Search (SEO) – 13.0% of total assisted conversions
• Paid Search – 35.2% of total assisted conversions
• Evidence demonstrates the Display Channel playing a critical role in
assisting conversions
52%
13%
35%
0%
10%
20%
30%
40%
50%
60%
Display Natural Search Paid Search
Assisted Conversions
25. 90% of one publisher‟s assisted
conversions are click through
• Publisher #1 drove the most click-through assisted conversions
(14% more than Publisher #9)
• Publisher #1 also performed well in (last interaction model)
generating clicks which may lead to conversions
90%
88% 88% 88%
84% 84%
79%
77% 76%
65%
70%
75%
80%
85%
90%
95%
Publisher
#1
Publisher
#2
Publisher
#3
Publisher
#4
Publisher
#5
Publisher
#6
Publisher
#7
Publisher
#8
Publisher
#9
Click Through Assisted Conversions
27. Performance summary by channel
• Overall display results all under 1.0, which could result in a loss of
<0.49% conversions if we spend 1% less in that channel
• Natural Search marked the highest return in all models
• Performance of Paid Search return exceeds investment in three
models and breaks even in the other three
28. Site Floodlight Last Interaction First Click First Interaction Linear Time Decay
Publisher #1 1.28 1.28 1.24 1.45 1.17 1.12
Publisher #2 0.99 0.87 0.92 0.78 1.02 1.06
Publisher #3 1.22 0.98 0.91 0.69 0.96 1.01
Publisher #4 0.80 0.77 0.82 0.82 1.30 1.33
Publisher #5 0.74 0.81 0.78 0.71 0.75 0.76
Publisher #6 1.43 0.99 1.17 1.07 1.29 1.33
Publisher #7 2.09 2.00 1.69 1.69 2.95 3.06
Publisher #8 1.71 1.49 3.59 2.08 2.04 2.12
ROAS
Performance by publisher (Display)
Channels Not Performing
Publisher #2
Publisher #3
Publisher #4
Publisher #5
Channels Performing Well
Publisher #1
Publisher #6
Publisher #7
Publisher #8
We are also able to conduct this level of analysis at a site, pan-
campaign or creative level
30. Key learnings
1. Define The scope
Start simple, then iterate
Create a roadmap and stick
to it
Timing is important
3. Tagging strategy is
critical to success
Don‟t just whack tags on
the site
Do plan the conversion
strategy
4. Understand the tech
Invest resource to
understand the capabilities
Tech moves fast
Test and Iterate
2. There are many
stakeholders
Assign an owner
Get buy-in from stakeholders
Set expectations – its an
ongoing process
OVERALL: it will take longer than you think, but its worth it!
31. “It was worth the effort just to get a
better view into what was really
happening off the back of our digital
investment.
Everything in this space is
measurable, we just had to decide
what was important to our business
and align everything to discover what
we wanted to know.”
Gerrit Walters – Brand and Advertising Manger, Kia Motors Australia
32. What do we differently now?
Reliable, data informed decision making
• Viewing the whole picture
• Consider substantiated alternatives
• Apportioning appropriate credit
• Incorporating these learnings ongoing
• Campaign tracking tweaks
• We‟ve moved beyond last click
34. Next steps from here
• A custom digital attribution model
• Deep dive into individual conversion events
• Apply weighting and value to events and conversions
• Re-define the customer purchase funnel
• Incorporate creative performance
• Ad-Visibility to conversion
• Cross device conversion??
• We will carry on testing & learning…
Image Credit: http://takingtheyoke.blogspot.com.au/2011/04/guest-post-leaving-our-home-church.html
We are both aware that there are many of these sessions that don’t really give you lots of take outs – our whole focus here is to share with you our process and what we achieved, as well as giving you the learnings.
Before (12 months ago)150 tags P.A.Multiple ad networksAd tech ownership blurredDigital success metrics variedOver reliance on Google AnalyticsLost or slow data / reporting discrepanciesSlow, and lack of agility to implement change
NOW50 conversion tagsNew SiteCore CMS = greater standardisationMobile enabledCentralised approach to digital measurementConsolidated platform agnostic conversion pointsTag Management in place = greater agilityClear technology roadmapAttribution
Tech & data in silos:Multiple technologies, all collecting dataResulting in multiple data sets, and multiple views of the same customerVery difficult to gain cross channel insightsTagging had gotten a bit messy – No tag management meant it was difficult to get tags on siteLast Click reporting We cognitively excuse display from delivering optimum performance due to this reporting flaw.The problem is this effect is assumed universal across all publishers, removing accountability. This relinquishes responsibility and invites campaign cruising.ALL RESULTED IN
Last click leaves so many unanswered questions…..What is the average path length of user journey?How often does upper-funnel activity drive last-touch Search Conversions?What Display channels are the most effective in assisting other conversions?What are the leading paths to conversion?
Thanks Andrew, I’m gonna take you through what the technology vision was for this project and an idea of some the tools we used.
I’ve always wanted to use an image from Fruit Ninja! I mean who hasn’t, right? I hope you all know fruit Ninja – its one of the first apps my kids ever got on the iPad, and aside from being a lot of fun, you get choose some really sharp samuari swords and cut up lots of fruit. The point is, that you need to use the best tools for the job, and you need to invest time and money in how to the most out of them, and that they stay sharp
Conversion strategy done with search and display in mindDataliciousSupertag allowed quick and easy deploymentDetailed map of conversion paths and points
Having access to the right resources is critical, and having everyone one the same page is critical. There were many stakeholders, and many moving parts as new technology gets s=isntalled, people are trainedWas critical to appoint a “project manager” – in this instance it was Andrew, who was leading the overall strategy and implementation. You can see however, as the media agency of record, Initaitive media took the lead overall with this project, using the Mediabrands resources of Reprise, Cadreon, Airborne and MNET, Innocean involved from a creatibve build and reporting Point of view, and underpinned by DoubleClick technology and consultants.
One all your data is in spot, you are able to use the reporting tools availabel to start to answer some ofthose questions we had at the start, like what is the influence my awareness display or ewmarketing display is having on my search campaigns?
We were able to use the rtools
So once you’ve implemented, and you have all data in one place, you can then start using the reporting and attribution tools available.DFA attribution tools allow to compare multiple models
For example, you may want to give more 10x credit to search clicks to conversions compared to display view through conversions, or 5x or 20xOr you might want to see what the affecdt of shortening your lookback window from 7 days to 48 hours looks like,m and see which channels are affected in the conversion picture.
For reference, here’s a very quick sample of the creative that was in market at the time.With the move in platform, we were able to migrate to dynamic creative for our Cadreon DSP marketing activity, maximising the effectiveness of our prospecting and remarketing activities, and presenting the most effective creative throughout the users journey through the funnel.Obviously, the output that you are to see is not flavoured, polarised, or intended to push a particular channel’s success – it’s agnostic, and is limited to a particular time frame.
Over half of total conversions come from 1-2 interactions (55.7%)Search delivered 87% of unassisted conversions, 35% of which coming from natural searchDisplay delivered 52% of assisted conversions - 68% of which coming from two publishers90% of one publishers assisted conversions, were post click not post viewOrganic Search performance was included in broader media attribution
Paid Search% of conversions in three models (First Click, Last Interaction, and Floodlight) are higher than % of media cost for Paid Search and other three models are showing almost same % as cost %.This shows Paid Search is a relatively cost effective channel in those three models.Natural Search% of conversions in every model shows much higher than % of media cost.Natural Search appears to perform best in all channels and even better than Paid Search from a ROAS perspective.Image and Simple FlashDisplay (Image and Simple Flash) showed the highest % under First Interaction, Linear, and Time Decay models.This suggests Display channel is actually doing its one of key rolls in assisting conversions, which hasn’t been seen under Floodlight model.However, % of conversions under Floodlight, Last Interaction, and First Click models are lower than the share of Display media cost which shows Display channel is relatively expensive compared with other channels.
This isn’t intended to be a promotion of a particular channel, I know what you’re thinking – he’re’s the search guy, selling search as the best channel.Natural Search marked the highest ROAS in all models, which means we could have more conversions if we increase spend to this channel. For example, Under Floodlight model, if we increase spend on Natural Search by 1%, we could expect 2~4% more conversions.ROAS of Paid Search is showing green (more than 1.2) signs under three models and this suggests Paid Search campaign is performing well although we’re seeing yellow (between 1.2 and 0.8) signs at the same time.Display channel (Image and Simple Flash) marked ROAS almost under 1 point, which means we could lose maximum 0.49% conversions if we spend 1% less for Display channel. This suggests we could consider shifting budget carefully from Image and Simple Flash to Paid or Natural Search channel.
Channels Performing WellThese sites show relatively high ROAS and we could expect 1.13%~5.83% increase in conversions when spending 1% more.Channels Not PerformingThese sites are showing pretty low ROAS.For a 1% spend reduction in these sites, we would only lose approximately 0.59%~0.87% in conversions.
Define the scopeStart out simple, then iterate - You could make this the most complicated program everDon’t get caught up the in the technologyGet the timing right – tying it into a site migration can assist2. Create a working group, and assign a single ownerCreate a roadmap and stick to it – tech changesSet clear expectations – there are still limitationsTest and iterate – you wont get it right the first timeAttribution is not a one-off project – its ongoing 3. Define the requirement first – this is keyAssess available technologies against requirementsAim for flexibility and agility4. Internally at KiaInternally at MBAccess from the tech vendor
Looking at the whole picture – PTC, Interactions, MoM changes to refine execution and inform strategyLooking at alternative models to last click when judging performance of baseline and campaign activityApportioning appropriate credit to medium end publishers2014 digital planning is incorporating these learningsChanges to paid search campaign to distinguish in reporting between brand, Model, category (Eg. “family car”) and generic termsWe have moved definitively beyond only last click attribution