2. Agenda
● How to optimise marketing budget with attribution modelling and data
integration
● Case Victorinox: How to connect all data and what to consider when doing
attribution modelling
● Case: How marketing agencies can benefit from automating reporting and
multi-touch attribution
● Discussion and questions
3. Quick intro: Who is talking?!
Niklas Kolster
- From Finland but speaks Swedish and German
- Founded Finland’s first VoIP Company in 2003
- After that focused for a while on quantitative trading and HFT.
- Was Lead data-scientist at Pryte, acquired by Facebook in
2014
- Headed Accenture’s data-science team in Scandinavia
- Siroop, eBay, windsor.ai
4. Challenges in marketing currently
● Data in silos
● Missing performance figures along the full customer journey
● Cost data not integrated
5. Data in silos
Teams go and check individual reports -> cannibalising conversion numbers and incomplete overview
Revenue tracked not realistic as it might contain cancellations and returns or quality of leads not taken into account
6. Missing performance figures along the full customer
journey
● Most used analytics tools are centered around last click
● Results in measuring clicks, bounce-rates etc.
● Optimising marketing based on low quality leads if the CRM is not connected
7. Cost data not integrated
● Multi-touch attribution only gives the attributed “conversion credits” to be
actionable this has to be connected to the cost. To get a ROI/ROAS for that
channel.
● A version of this is that someone compiles data manually
8.
9.
10. Solution
● Connect all data together
● Measure the impact of channels across the whole customer journey
11. Then it can provide you an overview of all channels
real performance
12. More about attribution modelling
The two widely used data driven attribution models: https://www.windsor.ai/shapley-value-vs-markov-
model-in-marketing-attribution/
15. Ausgangslage
Marketing is focusing on two objectives:
1. Awareness (benefits all sales channels)
2. Own performance marketing affecting e-commerce in multiple countries (run
by agencies)
Marketing channels:
- Google SEA, Shopping
- Bing, Facebook, Affiliates etc.
17. 17
Setup of integrations
Analytics and Ad Server Data
Channel cost data
Email and CRM-Data
Google
Analytics
Google
Campaign
Manager
MarketoSalesforce
Understand the whole student
journey from first interaction to
client
Give credit to each touch-point
using a machine learning model
Budget allocation scenarios
All media-spendings
in one place
18. We arrive at a data driven ROAS for all
countries, channels, keywords and
content
19. Attribution modelling and having a data-driven
ROAS on a keyword level can bring lots of
transparency
- Previously the business did not have so much insight into the performance
- Now with transparency it is easier to steer the focus and optimise where the
impact is the biggest.
20. Once there is ROAS based on multi-
touch attribution it can serve as platform
for both high-level decisions and tactical
optimisations
21. Challenge 1: Googles automation is good for google
Example:
- This can for example manifest it-self through expensive broad matches
capturing non-relevant traffic -> low relevance and quality scores ->
unnecessary costs
What is good for google is not always good for you.
Solution: Automation and highlighting the low-hanging fruit trough machine-
learning.
Still lots of manual work to decrease CPC’s from broad matches and creating ads
with good relevance scores.
22. Challenge 2: Automate going from broad matches to
exact increases relevance and decreases costs
28. 28
eClicks at a glance
● Mid-size agency based in Melbourne, Australia
● Google Premier Partner
● Strong focus on performance marketing (paid
and organic search strategies, remarketing,
conversion rate optimisation, analytics and
social media marketing)
29. 29
eClicks Goal:
Integrate all data and use insights to save
money and time!
● Transparency
eClicks manages large media buys on multiple channels but making decisions on business outcomes was
impossible with Google Analytics data alone.
● Double counting of conversions
Within paid social and paid search the numbers looked good but did not match with the clients numbers.
Multiple platforms take credit for the same conversion.
● Heavy reliance on last-click
Besides looking at intra-channel data, optimisations were done manually and based on last-click Google
Analytics data.
31. 31
Initially media buys were optimised based on
form fills
Customer Journey to form fill
Cost per lead
32. 32
Setup of integrations
Analytics and Ad Server Data
Channel cost data
Email and CRM-Data
Google
Analytics
Google
Campaign
Manager
MarketoSalesforce
Understand the whole student
journey from first interaction to
client
Give credit to each touch-point
using a machine learning model
Budget allocation scenarios
All media-spendings
in one place
33. 33
The media buys can be optimised on
real business outcomes
Marketo Salesforce
Customer Journey to opportunity
Customer Journey to enrolment
Cost per
opportunity
Cost per
enrolment
34. 34
eClicks can now optimise based on cross-
channel conversion rates to opportunity
36. 36
Results and outcomes (after 6 weeks)
● Cost per acquired customer and cost per opportunity were
measured for the first time down to the keyword and ad level
● Optimisations reduced overall the cost per CRM opportunity by 19%
● eClicks automated 50% of a FTE by using the automated media
optimiser
37. 37
Results and outcomes (after 6 weeks)
● Cost per acquired customer and cost per opportunity were
measured for the first time down to the keyword and ad level
● Optimisations reduced overall the cost per CRM opportunity by 19%
● eClicks automated 50% of a FTE by using the automated media
optimiser
Teams go and check individual reports -> overlap and missing overview
Revenue tracked not realistic as it might contain cancellations and returns or quality of leads not taken into account
Show journeys:
-stats
-3D
Taking it one step further is the budget optimiser
Not going into these details now
See what to increase and what to decrease
For product we have a free trial at onboard.windsor.ai