Good forecast of supply and demand is crucial for advertisement inventory management in the online advertising industry. Imagine that an advertiser is planning to purchase some inventory matching the following criteria: (age > 30) and (state = California) and (web_property = yahoo_news) from Dec 1 to Dec 31, 2010. Two questions need to be answered:

• How much inventory will be available to match the advertiser’s target?
• What is the cost to the advertiser?

There are three challenges in answering the first question. First, inventory forecasting at web-property level is very challenging. Many factors influence the future inventory, such as seasonal effects, user growth, economic environment, etc. Second, there are dozens of targeting attributes (including age and state) and the number of possible targeting criteria is in billions. Third, the response of the forecasting system must be in real time. Each inventory query should be answered within seconds even if it has never been seen before.

To address the second question above, we need to forecast advertisement demand as well, as pricing should be a function of supply and demand. Time series analysis and booking curve analysis are possible tools for demand forecasting.

Future research includes two areas:

• Improving forecasting of small inventories: Increasingly advertisers would like to buy highly targeted inventories, which tend to have small supply with huge variances. Accurate forecast of such inventories is very challenging, but is crucial for revenue optimization.
• From impressions to reach: Rather than purchasing ad impressions, traditional big advertisers would like to purchase user reach (percentage of US households who will be exposed to the ads). To support such programs. we need to forecast unique web users who match certain targeting criteria.