Forecasting Page Views

1
Forecasting Page Views Ad Impressions for
Bullseye Ad Campaigns

• David Feng
• Statistical Analyst Intern
• GEL, Trax

2
What is Bullseye?

• Cross-game advertising
• A clients ad campaign runs exclusively on a
  selected game space for a set amount of time.
• Example
• I want to advertise for a new football game I
  created.
• I feel that Madden players would be interested in
  my game.
• I buy a Bullseye campaign on Madden NFL 09, since
  I am targetting Madden players.

3
Existing Bullseye Procedures

• Sales team has two options
• Propose a predicted number of ad impressions to
  the client (but how much?)
• Charge a flat rate on the Bullseye campaign (Halo
  3 Balls of Fury)
• Problems
• Under-delivery must compensate client in some
  other way make goods
• Over-delivery lost opportunity for revenue
• Flat-rate pricing lost opportunity for revenue
  on AAA games

4
Bullseye
5
The Bullseye Predictive Model

• Objective To provide a more objective means for
  predicting the number of advertising impressions
  in a game space.
• 4 phases
• Data-collection and processing
• Predict number of page views for a game space 30
  days, 60 days and 90 days in advance
• Find the relationship between page views and
  advertising impressions
• Test validity of models and develop business rules

6
The Bullseye Predictive Model

• Objective To provide a more objective means for
  predicting the number of advertising impressions
  in a game space.
• 4 phases
• Data-collection and processing
• Predict number of page views for a game space 30
  days, 60 days and 90 days in advance
• Find the relationship between page views and
  advertising impressions
• Test validity of models and develop business rules

7
I. Data Collection

• Gathered data on all games released from 1/1/06
  to 5/31/08 using Trax
• Data pulled in Mon-Sun increments
• 3782 games, 2700-2800 after removing irrelevant
  games
• Computed lifecycle data for all games on the
  weekly level

8
II Predicting Page Views

• Predict page views because it should be the
  strongest determinant of ad impressions
• Determinants

9
Heteroskedasticity Big Games Big Errors
10
Constructing the Forecast Interval

• Two sources of uncertainty
• Standard Error of Regression (SER)
• Coefficient Uncertainty
• SER
• Future is uncertain (see figure)
• Reduce through stratification (e.g. separate
  models for X360 games, casual games, Q4, high
  PV-growth)
• Coefficient Uncertainty
• Uncertain coefficients, so uncertain dependent
  variable
• Ignored
• Sample size not large enough
• Software limitation

11
Solution

• Sell Bullseye ad campaigns on at least 3 SKUs to
  reduce risk of underdelivery
• Just need to reduce the SER to improve models
  goodness-of-fit

12
Summary of Results 30 Days
13
Summary of Results 60 Days
14
Summary of Results 90 Days
15
Testing Validity of Models

• Use games released after 5/31/08
• Test Interface Final Interface
• Interface will eventually be automated into a new
  tool, so that the sales rep only needs to put in
  the game ID, and all data will be pulled from
  Trax
• A new Bullseye tool and/or an internal version of
  Trax could be created to ease testing and usage
  of the models

16
Next Steps

• Test models
• Finalize relationship between page views and ad
  impressions (need 2007 Bullseye campaign data)
• Generate models for 120-day and 180-day windows
• Establish business rules using the results of
  testing
• New rules on Bullseye ad campaigns (e.g. at least
  3 SKUs)
• Ability to revise projections at the 90-day,
  60-day, 30-day windows
• All Bullseye ad campaigns can convert to the CPM
  model

17
Acknowledgements

• Sara Borthwick (manager)
• Entire Trax Team (Anne, Christian, Christie,
  Colina, Marisa, Matt, Maura, Shanna)
• Product Marketing (Andrew, Nina)
• Statisticians (Erika, Phil)
• Excel Master (Jared)