Business Intelligence – Taking the sting out of forecasting

“I Go to Extremes,” is the title of a 1989 Billy Joel hit. It also describes something Thomas Tileston can never afford to do in his job.

Predictions are Tileston’s stock-in-trade. As vice-president of global forecasting at Warner Home Video Inc., Tileston has to accurately estimate the demand for new movie titles, so the right number of DVDs and videos can be shipped to retailers across the world. And he and his team need to get it right, and have one shot to do that.

“The window of opportunity is short,” Tileston says. “With a major release, 45 per cent of the total video sales happen the first week, and 75 per cent within the first month.”

In such a volatile market, over-forecasting would mean inundating retailers with excess inventory, which would be returned to Warner Home Video for storage or disposal. Under-forecasting, on the other hand, means vacant shop shelves and missed sales opportunities. Striding the middle ground between these extremes is a big challenge for the Burbank, Calif.-based video distributor.

And Tileston and his team of professional prognosticators are also keenly aware of another big challenge associated with forecasting new release DVDs: that brand loyalty is limited and studio loyalty almost non-existent. “No one would go into a store thinking: I’m going to buy a Warner DVD. That’s not how it works,” says Tileston.

Other variables include rapidly changing consumer demand, and the fact that products are highly differentiated, so finding reliable historical precedents is difficult. Then there’s the fact that drivers for DVD sales change pretty quickly. “When a disk is released, sales are initially title driven, but very soon they become media driven,” says Tileston.

A dramatic change in forecasting fortune
Reporting in to the Warner Home Video exec are several analysts in four global cities — Burbank, Toronto, London and Paris — whose job is to accurately forecast around 300 titles a week. Three years ago this just wasn’t happening. In 2003, average forecasts at the company were off the mark by around 40 per cent.

Then suddenly that situation changed dramatically. In 2004, some smart forecasting helped Warner Home Video sell 68 million DVDs of the three Harry Potter movies, and lead the industry with sales of $4.75 billion, according to Video Business magazine.

How were Tileston and his team able to pull off this coup? By cleverly combining the strengths of two key data mining/data warehousing technologies — from SAS Institute and NCR’s Teradata division. Tileston described the impact of this magic mix at a SAS 2006 Business Forecasting Conference held in Cary, NC. His presentation was appropriately titled ‘Have your cake and eat it too.’

In forecasting feature film DVD sales, he said, several factors must be considered: how the picture fared at the box office, its ratings, popularity of the cast and director, the genre (family, thriller, mystery and so forth), and the time interval between the theatrical release and the DVD launch.

Tileston’s team looks carefully at all these variables, and also attempts to find past movie titles similar to the ones they are currently trying to forecast. “Say we want to forecast [the movie] Poseidon. We look back and investigate historical titles in the same genre: expensive disaster pictures.”

The team also takes point-of-sale data collected from retailers and matches that to title attribute data. All of that has to be aggregated and then a ‘decay curve” built to identify weekly percentage drop in sales. These then are the key data tasks Tileston and his group have to accomplish in a new release theatrical forecast:

• Identify and combine similar titles (genre, box office, rating);
• Merge point-of-sale with title attribute data and analyze historic performance;
• Aggregate merged data to different levels — for example, by weeks of release;
• Build decay curves that identify weekly percent drop in sales;
• Depreciate Box Office dollars of older titles; and,
• Perform statistical procedures — such as regression and cluster analysis

Warner Home Video’s data warehouse holds a trillion pieces, and more than a billion rows of data, which have to be sifted and analyzed to accomplish the tasks listed above. For Tileston the fundamental question is: where should all this data work occur — in Teradata or SAS?

Which tool?
SAS’s strength, he said, is that it offers the best data mining tool. SAS also provides flexibility in coding, and has the largest, most robust stats package available, he added. On the other hand, Teradata SQL (structured query language) harnesses the power of the system, and parses, analyzes and extracts data fast. “Aggregations and sorts are also easy and efficient in Teradata, while large data sets are simple to manipulate,” he said.

After much experimentation, Tileston and his group eventually decided to split the required tasks between the two systems. Tileston’s rule of thumb for divvying up the work was: if it’s data mining, do all of it in SAS; if it’s data manipulation or management, choose one or the other, depending on the specific task. For instance, it makes sense to do all data joining and heavy processing in Teradata, while using SAS functions and procedures instead of the Teradata equivalent, he said.

When Warner Home Video implemented this Teradata/SAS split in its data forecasting process, the impact was decisive and dramatic. Previously, new release forecasts, done in a SAS (PC)-only environment, took 36 hours for around 300 titles. “By splitting tasks between Teradata (90 per cent) and SAS (10 per cent) the same new release forecast now takes an hour and 15 minutes. That’s a huge win,” said Tileston.

When it came to the “talent ranking model”, he said, the SAS/Teradata blend provided similar gains. “In a SAS-only environment it took two hours, but splitting the tasks between Teradata (95 per cent) and SAS the same application took one minute.”

What is SAS’s take on this model that uses its technology in a very specific capacity — for data mining — while relying on Teradata for all the heavy data joining and processing? “We’re flexible, because customers have different needs and skill sets and want to harness those to their best advantage,” said Anne Milley, director of technology product marketing.

However, she added, Warner Home Video’s case was unique in some respects. “For them efficiency was the bottom line, and they had a rather unique combination of skills. I’m not aware of very many people who know both SAS and SQL to the same level that Thomas Tileston does.”

Milley also claimed it wasn’t an “apples to apples” comparison between the two systems, as Teradata was on a six-node box, and SAS was on a two-node PC.

The SAS executive also argued that “efficiency” should be looked at more broadly. “Yes data storage has got cheaper, but when you consider all the factors — the skill set that needs to be optimized, technology that can be brought to the table and over what time horizon that needs to be done — I still think that the SAS platform has an excellent story.”

QuickLink: 067521

Joaquim P. Menezes is Web Editor of IT World Canada. He can be reached at

Related Download
3 reasons why Hyperconverged is the cost-efficient, simplified infrastructure for the modern data center Sponsor: Lenovo
3 reasons why Hyperconverged is the cost-efficient, simplified infrastructure for the modern data center
Find out how Hyperconverged systems can help you meet the challenges of the modern IT department. Click here to find out more.
Register Now