2. PROBLEM
• Gauge the performance ofTV advertising for company XYZ:
• Website traffic that is attributable toTV advertising
• Metric for measuringTV ad performance
• Calculating metrics
• Visualization:
• Improving client dashboard XYZ
3. CONTENT
• Part I
• Metric
• Part II
• Client dashboard
• Part III
• Future work!
5. PART I
• What is Marketing Attribution ?
• Assumptions
• Data:
• Spot Data: Exploratory Analysis and Cleaning
• Traffic Data: Exploratory Analysis and Cleaning
• Baseline
• Metrics
6. WHAT IS MARKETING ATTRIBUTION?
• “… Marketing attribution provides a level of understanding
of what combination of events in what particular order
influence individuals to engage in a desired behavior,
typically referred to as a conversion...”
Source: https://en.wikipedia.org/wiki/Attribution_(marketing)
7. ASSUMPTIONS
• Website traffic through ‘direct’ traffic source during the
first ten minutes of airing an ad is considered attributable
toTV advertising
• It is acknowledged that there could be ads in a different
"program" within a second, meaning multiple ads are
attributable
• Therefore, Metrics were calculated on a daily basis
• Note:Visitors clicking the link on an ‘email’ are not considered attributable to the lift
13. EXPLORING SPOT DATA: TIME DIFFERENCE
• Shortest time difference between airing of an ad is a second
• Therefore more than one ad attribute to a lift in that 10mins window
14. EXPLORING SPOT DATA: FINDINGS
• Consist of 1456 rows and 13 columns
• Each airing on the East coast followed by one on theWest coast (local time)
• 160 entries with missing 'program' and 'duration' values
• 124 entries with missing 'duration' values
• Ads commenced at 2017-10-16 8:25am ET and lasted till 2017-11-13 5:53am
ET (US, Eastern).This is about 29 days.
• Interval between each airing on a single ‘program’ might be at least 31mins,
but with multiple ‘programs’, it could be as short as a second
25. EXPLORING TRAFFIC DATA: FINDINGS
• Consists of 63951 rows × 3 column
• 40,380 rows are ‘direct’ and relevant to this investigation as the visits are
made by typing the URL in a browser
• 23,571 are through clicking of an email and are unrelated
• Data collection started at 2017-10-16 3am ET and ended at 2017-11-13
02:59am ET (US, Eastern).That is about 29 days.
• This traffic data is collected every minute
29. METRIC
• Website traffic that is attributable toTV advertising is considered
to be within the first ten mins of airing an ad less the baseline
• Metrics are calculated per day
1) Lift =Value - Baseline
2) Spend = Sum of ‘spend’
3) Cost perView (CPV) = ‘spend’ / ‘lift’
4) Spots aired = Count of ‘time’ (or ‘spend’)
41. CLIENT DASHBOARD
• Approach to dashboard:
• Collect information from stakeholders
• Develop the specification
• Identify the technology stack
• Complete the required analytics
43. FUTURE WORK
• Computing Baseline
• Develop an advanced smoothing algorithm: e.g. a two
stage non-linear signal processing algorithm
• Marketing Attribution: Big Problem!
• Which ad attributed the lift is a big challenge
• Hidden Markov Model, ShapleyValue, Logistic Regression
and Classification
• Multiple ads within the chosen 10min window
44. CONCLUSION
• Metrics were computed following exploratory analysis within the stated
assumptions
• The “lift“ has followed the “spend” for the most part, validating the
computation
• Average CPV was $1.21
• Future work was identified:
• Developing a more accurate algorithm for baseline calculation
• An advanced attribution model
• Addressing multiple ads within the chosen 10min window