The document discusses the principles of performing an investigation into issues with PPC accounts. It outlines a root cause analysis (RCA) process involving defining the problem, limiting the scope, and investigating through before/after analysis and comparing impacted vs non-impacted campaigns. The goal is to identify the root cause rather than just addressing symptoms. Key steps include understanding the trigger, problem definition, asking questions, and using data visualization and filters to find the "smoking gun".
5. We are the best* scientific
performance marketing agency...
* we are the best according to our mums and now that we’ve won an award as the Best small PPC agency in Europe, it’s official: mums are always right.
6. ...a complex way to say that we
love numbers and the scientific
method
14. RCA
● Root cause analysis (RCA)
● The root cause analysis process tries to find the source of
an issue, as opposed to simply addressing the symptoms.
16. Step #0: Trigger
● This is the part when you realize you have a problem
● It can be triggered by a customer email
● Or even better by one of your automation alerts
19. Step #1: Problem definition
● This is the part where you try to describe the problem
● Making sure you are delimiting the characteristics
○ what is the metric impacted
○ what dates
○ are there other metrics?
20. Step #1: Problem definition
● Something is
wrong, we get less
sales
● Something is
wrong, starting
last Monday we
noticed a sharp
decrease in Sales.
● Something is wrong,
starting last Monday
we noticed a sharp
decrease in Sales.
● Google Ads is the
only channel impacted
● The problem is limited
to the UK
😃
😞 😐
30. #2. Limiting the scope of the issue
● Visually look at the graph to see if there are variations in
the metrics impacted on a specific date
● Focus on higher spending (or converting campaigns). If
the problem is impacting the entire business chances
are the top campaigns are impacted
33. #2. Limiting the scope of the issue
● Try to see if the problem is affecting only a portion of the
account:
○ a certain campaign type (eg. Shopping only)
○ a certain geography
○ a certain device
48. Approach 1: Before/After analysis
● Compare dates to see when the performance
decreased:
○ If the problem is concentrated in time (2 weeks
before / after the event)
○ If the problem is not concentrated in time (2 weeks
that are clearly affected by the problem vs 2 weeks
when the problem did not exist)