Slide1: Hosted by Decisioneering, Inc.
April 24, 2007 I'll Be Back: DVD Sales Forecasting
To listen to the session on your phone, follow the instructions in the 'Join Teleconference' pop-up dialog box which will appear in a few moments. To listen to the session on your computer speakers instead of your phone, follow the instructions in the 'Join Internet Phone' pop-up dialog box which will appear in a few moments. Please Do Not join both, as this is redundant. Guest Speaker
Michael Lieberman President, Multivariate Solutions Register Now for the 2007 Crystal Ball User Conference!
http://www.crystalball.com/cbuc/index.html I'll Be Back: DVD Sales Forecasting: I'll Be Back: DVD Sales Forecasting Michael D. Lieberman
President
Multivariate Solutions
http://www.mvsolution.com
[email protected]
The Ruse-Scheme Business Model: The Ruse-Scheme Business Model Ruse-Scheme is a DVD distribution company that houses up to 30,000 titles.
Ruse-Scheme has a membership approaching 6,000,000.
Ruse-Scheme charges its members a monthly fee for different levels of memberships.
Bona Fide
Preferred
Platinum
Ruse-Scheme collects customer information when members join:
Each member has a member ID (e.g., email address)
Demographic information
Age
Income
Ethnicity
Education
Favorite Film Categories of movies
Today: Today Ruse-Scheme wants to be able to assess the number of DVDs it must purchase in order to begin renting to its customer database.
MVS is tasked to develop a Conditional Demand Analysis (CDA) to test customer needs.
Tasks—
How to use History Data to assess future Ruse-Scheme database.
How to pinpoint factors that are most important when Ruse-Scheme makes DVD inventory decisions.
The blending of database mining, market research, and Monte Carlo simulations using Crystal Ball®
Producing Demand Model
The 'Freakonomics' of follow-up.
Assessing time related factors (e.g. promotions)
Business additives for Ruse-Scheme
An added benefit of assessing the ‘expected value’ of each level of membership
The Ruse-Scheme Conundrum: The Ruse-Scheme Conundrum How many DVDs do they need to order for new releases or historical flicks that come to DVD?
If Titanic, then order a lot.
But what if?
Limited-release indie films
Documentaries: Supersize Me
A flop: Basic Instinct 2
Old Movies: East of Eden
Foreign Films: : Les Invasions Barbares (The Barbarian Invasions)
The Ruse-Scheme Database: The Ruse-Scheme Database The Ruse-Scheme Database is able to provide a list that permits matching demographic variables to titles rented.
Rentals to key demographic groups.
Indicated favorites.
Time lag between release of a DVD title and it’s a rental request.
Accepted bumps in promotions.
Developing a Model of Trial Renting: Developing a Model of Trial Renting Start at the individual level, then aggregate
What are the individual-level behaviors?
Time
Segmentation Approach
Test ‘accepted’ industry knowledge
For example, male 10-13 viewers have a higher likelihood to view Ninja Turtles then soccer moms 26-40. True?
Divide the membership list into a set of homogenous segments.
Test membership response by emailing to a random sample of each segment.
Segmentation results can then be correlated with membership type
As time passes, and historical data becomes longer, an expected value of membership by key demographic groups can be developed.
Asked how many DVDs members might watch in a given time period.
Weekly in the preferred time assessment.
Life changes can affect DVD rentals.
Blending Techniques: Blending Techniques The Role of History (Data Mining)
Ruse-Scheme uses it historical database.
Market Research
Ruse-Scheme regularly communicates with its membership by email.
Recently sent an email survey to its members with the upcoming list of DVD releases and asked them to check off the ones they intended to rent within the next month.
The CDA (Conditional Demand Analysis) is a blend of the above techniques.
Model Input: Model Input Demographics
Using input from Ruse-Scheme executives and theater sales, we were able to examine key demographic groups. Regression analysis will show a link between these groups and rental sales.
Intentions
For a given film, if a respondent indicates he intents to rent, the model regresses this against actual rental data (three months later).
The Role of Film Category
Are fans of one film category more predictive than fans of another?
Sci-Fi vs. Chick Flick
Looking at historic rental sales, sorting by Film Category, it became clear that Film Category is a key indicator of rental intentions.
Model Input—Apply Weights to the Core CDA Regression: Model Input—Apply Weights to the Core CDA Regression Demographics
The Initial model uses regression analysis. Demographic weights are assigned.
Ruse-Scheme provides the film usage percentages.
The prior regression analysis indicates which demographics should be weighted.
Weights are created
Percentage Actual/Percentage Indicated (in response to email survey)
Validity Weights
Do customers do as they say?
Matching rental intentions with actual customer rentals is a necessary, and useful, market research application.
Model each film’s rental using a finite mixture model with known intention weights.
Rental Intention Differences: Rental Intention Differences Data altered to protect confidentiality
Dry Run - Known RentalsLassie, The Puppy Years: Dry Run - Known Rentals Lassie, The Puppy Years Rentals of the kiddy flop, Lassie, The Puppy Years are known.
Inputs for Lassie, The Puppy Years are entered into Crystal Ball Excel Sheet
Lassie, The Puppy Years—Comparison: Lassie, The Puppy Years—Comparison Gray Line: There is an 80% chance that 138,000 rentals or more will take place. Blue Line: Expected Value; 50% chance of 140,000. Red Line: Ruse-Scheme is advised to order 142,000 copies. Actual 3 month rentals 138,612
Rollout the Model: Rollout the Model Film
Horror Film: Second Term
Details
Summer flick
Flop despite massive spin
Quick release into DVD (before approval ratings drop further)
More popular in Red states
Model Particulars
Mixed Model
The Monte Carlo mix will be higher among key demographic groups.
Regional considerations
Slide15: Below represents a sample input of one of the mixes in the model
Inputs are the CDA regression.
*Conditional Demand Analysis (regression analysis) Input altered to protect confidentiality
Unknown Rental Sales: Unknown Rental Sales The fearsome flop, Second Term, is going to DVD
Tiered Inputs for Second Term are entered into Crystal Ball Excel Sheet
“Second Term” Recommendations: 'Second Term' Recommendations Red Line: Ruse-Scheme is advised to order about 76,000 copies
Comments: Comments The model is used on a per film basis.
Individual movies are modeled
Model can be modified to encompass ‘Categories’.
An average rental scheme for a group of movies
Optimization of purchasing within budget using OptQuest
As the model evolves, use industry expertise to tighten the measurements.
Sub-models per category or by market.
West Coast vs. Chicago metropolitan area.
Follow-up data-mining not only determines the validity of member rental intentions, it provides additional insights.
Market Research Application: Market Research Application As a by-product Ruse-Scheme executives can examine average differences
Average gap per flick
List of flicks for demographic groups
For example, ‘Divorced men, graduate degree, 35-44 years old, no kids’.
The 'Freakonimics' of Follow-up
Do actual box office (theatre) sales correlate with Ruse-Scheme membership rental patterns?
Time of year—Do holiday promotions and summertime flicks impact Ruse-Scheme membership intentions?
Is a combination of the two, when applied to film category, able to produce a validity model?
Pre-Post Application to Advertising: Pre-Post Application to Advertising Does media advertisement bump rental sales and/or Ruse-Scheme membership validity?
If so, can Ruse-Scheme sell this idea to particular films who want add to their rental revenue?
For example, 'Second Term' is in the red. Can a joint advertising venture with Ruse-Scheme boost membership rental activity, and close the financial gap?
Should Ruse-Scheme allow individual studios to purchase their membership list to advertise their films?
Future Research with Time: Future Research with Time As further data comes in, Ruse-Scheme will be able to assess the value of membership.
Given the tiers of membership, how many flicks do the various members rent?
The stability of parameters across cohorts and film category can be assessed.
Attrition
Upgrading
Death
Further pricing information may emerge.
Compare this model to more complex models (with some help)
Fit of the forecast
Parameter validly
Cost/benefits of inventory assessment
Conclusion: Conclusion We used historical and behavioral data to parameterize the return distributions
Manage risk
Produce tiered models
The mix of techniques produces not only the desired goals, but a side dish of valuable information
Crystal Ball enables risk analysis of uncertain inventory controls Thank you for attending the Web Seminar I'll Be Back: DVD Sales Forecasting: Thank you for attending the Web Seminar I'll Be Back: DVD Sales Forecasting Michael Lieberman
President
Multivariate Solutions
E-mail:
[email protected]
Phone: 1.646.257.3794
Website: http://www.mvsolution.com Hosted by Decisioneering, Inc.
April 24. 2007 Decisioneering, Inc.
1515 Arapahoe St., Ste 1311
Denver, Colorado 80202
303-534-1515
www.crystalball.com Register Now for the 2007 Crystal Ball User Conference!
http://www.crystalball.com/cbuc/index.html