Table of contents for BIWA Summit 2008
- Live from BIWA Summit – Competing on Analytics
- Powering Next-Generation Predictive Applications with Oracle Data Mining (ODM)
- Critical Success Factors for successful BI and analytic implementations
- From Data Warehousing to Strategic Data Assets
- Getting to the Right Price with Oracle Data Mining
- Oracle’s BI Strategy
- Intelligent OLAP: Data Mining and OLAP
- Fraud Detection with Oracle Data Mining
Usama Fayyad, previously Chief Data Officer of Yahoo, presented a keynote on From Data Warehousing to Strategic Data Assets – Case Studies on the Web: Social Networking, Direct-Response Marketing and Understanding Customer Behavior.
The number of users on the web continues to grow rapidly, approaching 1Bn. The data created as these users move around the web is huge. If you consider the events per day, SABRA has 50M events/day, VISA has 120M, NYSE has 225M and Yahoo!, for instance, has 14,000M events/day! Since 1998 the largest data warehouses has gone from a few Terabytes to many Petabytes. However, although performance is OK, price-performance is not. However, not all the data is needed for every operation, so you can manage less and less data as you deliver higher business value and performance. More expensive storage manages less data, for instance. This is all focused on the technical matters of the data warehouse. This is exaggerated by the explosion of data from the web and a worldwide customer base that has overwhelmed straightforward approaches to analytics. But what about the business need, KPIs, data strategy?
Rather than focusing on price, performance etc, Yahoo! developed a data strategy to prioritize business needs and have that drive innovation. Data strategy is a change of mindset from data as a mundane, historical record to one where it is a strategic asset. This is made more difficult because the business side of the house does not understand the technologies and capabilities. Data strategy is about mapping the capabilities to the business needs and drivers.
In Yahoo!’s business model there is brand advertising at the awareness end of the funnel (about $200B in the US) and direct response marketing at the purchase end (about $250B). The two groups involved are very different and don’t get on – brand marketers all driven by emotion while direct response all driven by the numbers. But what about the middle – the consideration phase? No products available because historically had no access but these people are still using the web while considering so Yahoo! could target them too.
He spent some time talking about the need to focus on your actual task. For instance, one could keep searching for information on coffee machines before you bought one, while buying one, while learning to use it and finally while trying to do advanced maintenance. Understanding this so you could target these searches differently is going to be next.
Take the consideration phase. Moving up the funnel first: Can use searches from the recent past to drive banner ads. Essentially use search information to bias banner ads for the 48 hours after a search. This generated much higher response rates and a lot of extra revenue. Also found that if you keep adding non-search behavior from elsewhere on the Yahoo! network then the increased rates of click through continue to rise. Can also move down the funnel because banner ads and search ads interact. Banner ads increase brand awareness, favorabvility and purchase intent, as you would expect. Good results but not great. But if you study the search engine behavior of the people who saw the banner ads you see a dramatic increase in the rate at which the advertisers paid search ads! Clearly seeing the banner ads changed the behavior of those who saw them.
He ran out of time at this point but he talked a little about the impact of social media and user generated content on search. For instance, searches for Jaguar return pictures of cars, animals, macs and aircraft. Separating these out algorithmically is very hard but the user generated content does it easily.