Design of multichannel attribution model using click-
stream data MeasureCamp Prague 2015
Lucie Šperková
Everything you need to know (about me)
I used to work in bank.The only language I can use is SQL.I have never worked directly with GA, just extract the data from it.
Somebody said: “Without data you are just a person with an opinion”I say in addition:“… but with data, which are messy nad shitty, you are a clear liar.”
Data overload?Lack of data -> incomplete decisionsToo much data -> overload and still lack of knowledge (What I should focus on?!)
Basement / garage problemI store big volume of data just for case, but will probably never use it. -> Ask yourself why you will need them (have a target)
Why?costs and revenues
expenses and benefitsincome and spending
profit and loss
customer loyalty/satisfaction
TargetCreate exponential model that takes into account all the inputs into
the conversion funnel.
With the use of AdForm metadata: for every cookie (user) on the particular trackingpoint calculate number of interactions for the
particular time period and assign weights to campaign channels.
What I do / will do with the data...- calculation of the weights and share of channels in conversions
- budgeting the total cost to the individual channels according their share
- visualize the shares of the channels
- drill down the channels - to medium, campaign,...
- slice according to refferer type, device type, customer segments …
- find the right campaign mixture (how to achieve particular number of conversions for the lowest price)
- prediction of the future development and setting the right campaign mixture
- observe the conversional / non-conversional rates (how many interactions didn’t lead to conversion)
- intregration of data from other sources (GA, sklik, CRM, budgets, etc.)
- revenues from conversions
- customers data
- ...
seen the banner 1
seen the banner 2click
PRclick PPC
click Organic
click banner1
Web - Conversion1point 1point2points 2points2points 3 points
Weights assigned according to:
basic division:
conversion click (triggered the trackingpoint)
last impression (triggered the trackingpoint)
direct entry
click
impression
refining the weights: ● by mouse overs, mouse over time, visibility time,
refferer type, medium etc.
● on the web there are many trackingpoints cookie has visited (not interested about the move through websites)
● focus on conversion points or points foregoing conversions (e.g. where customer left the action)
Trackingpoint A
Trackingpoint B
Trackingpoint C
Trackingpoint D
Trackingpoint Econversion
metadata
calculationsextract
extract
transform
Process of basic transformation
data cleaning- delete robotic transactions- transactions, which happened in less than 30 minutes from the last transaction (same cookie, same
trackingpoint, same session) - avoid refreshjoins
- for every cookie at the trackingpoint find all interactions which happened during the time between trigger of the last trackingpoint and today’s trackingoint (for more conversions of single cookie)
- every cookie can have interaction with different campaign: calculation for every campaign (avoid multipletimes counting of the same add - banners etc)
- the campaign of the conversion interaction is known (higher weight)weights calculation and refining
!
Predictionscosts
conversions (revenues)
more investments to this campaign mix won’t help
right campaign mix for acceptable price
100 300 330
Thanks. Let’s talk!
mail: [email protected]: https://cz.linkedin.com/in/luciesperkovatwitter: https://twitter.com/pihatka