Commentary

Smarter Than An Ad-Serving Platform?

Throughout January, many of the major technical publications in our space and most technical blogs have agreed that 2009 is shaping up to be the year of cloud computing. From lead articles in InfoWeek and CNet to events such as SoftwareAG's Cloud Computing Innovation Day, Under the Radar, AdMonsters and the DCIA's events, it seems that cloud computing will dominate discussions throughout our industry.

So what does cloud computing mean for the world of Interactive Advertising?

Ad-serving on the web began in the early-mid 1990s. Many display publishers still rely on systems that haven't changed a whole lot since then. This seems inconsistent because most people in the interactive ad business consider themselves technologists not marketers. Unfortunately, there haven't been many new options or technological breakthroughs in the last 15 years to change the ad serving landscape.

Ad serving has always been about supply and demand, but with so many variables, it takes some pretty fancy software and diligent sales staff to track everything. Even then, revenue is lost. Publishers have multiple constraints today: overlapping target audiences, campaign conflicts, inventory geographically or seasonally constrained. Add multiple remnant ad networks and ad exchanges to the mix, getting the most out of every impression and not leaving money on the table has become a very complex, labor intensive task. Media buyers and ad ops folks will be the first to tell you that there is no standard or easy way to handle inventory management, forecasting and optimization for premium guaranteed cpm display inventory.

Current ad serving models do not account for future inventory availability or demand which drives both pricing and of course, revenue. Most ad servers use historical data to determine what ads should be served. Historical modeling for risk and inventory allocation makes sense in some contexts, particular where there is a predictable pattern. But looking backward does not work well for interactive advertising for the very reason that demand, and factors like seasonal targeting, rely on future events. Inevitably then, display inventory allocation or pricing based on past events is not accurate.

Optimally, trafficking and operations staff would prefer to allocate order delivery with greater insight into what is actually available and what is likely to happen based on fact and not sampling or conjecture. To maximize revenue, publishers must make the most of every impression and allocate orders as efficiently as possible. It makes more sense to know in real-time how an ad server will deliver an order when scheduling new orders or making changes to existing orders.

That's where cloud computing can make a difference in today's ad serving world. For the uninitiated, cloud computing is a network of servers and connections collectively known as "the cloud." In cloud computing the network becomes a supercomputer. When a request is made to the cloud, tasks are assigned to a combination of connections, software and services within the network. So instead of being tied to a single limited server, cloud enabled platforms can access a very powerful network providing a clear path to manage scale, data complexity and speed.

With cloud computing technology, massive amounts of complex data can be accessed, processed and computed at one time, producing real time results. The cloud combined with proprietary algorithms, analytics and a special purpose database provides highly reliable revenue and sell-through forecasts at the product level and allows for granular inventory forecasts for many different products at the same time.

The immediacy of all this information enables advertising sales teams to make smart decisions about pricing and targeting, behavioral and contextual. Because a cloud computing analytics model can provide clear understanding of availability, under delivery and value, sales teams can take proactive steps to correct course and maximize yield. With this kind of proactive knowledge the sales teams can better communicate with clients and internal stakeholders to identify the best options.

Cloud computing is truly what's hot this year. But what is really important to interactive is that this new technology will finally move the needle for delivering value to marketers and publishers alike. Analysts covering the ad serving space have said for years that the old model is fatally flawed because it uses historical modeling to solve uniquely dynamic, forward-looking, multi-variant supply-demand equations. Having proactive modeling and more accurate inventory knowledge and forecasting won't solve every revenue problem for publishers in this troubled economy, but it will certainly enable them to maximize the opportunities that do exist.

For those of you in ad operations I suggest that you take a few steps to get educated and ensure your organization is taking full advantage of cloud computing:

1) Talk to your tech team and understand if you can leverage cloud computing to solve your internal data volume challenges

2) Take the time to understand how your data is gathered and processed; is it sampled or based on summary information that is missing granularity

3) Schedule meetings with your vendors and have them brief you on their use of cloud computing and how it could improve the reliability of their products

Recommend (2)
4 comments about "Smarter Than An Ad-Serving Platform?".
  1. Joe Fredericks , February 16, 2009 at 9:57 a.m.

    I could not agree more that high-powered computing is critical to maximizing revenue potential for the publisher - and the advertiser, too. This is the future for ad exchanges where all display ad inventory can be traded in real-time and yield the best deal possible for all parties.

  2. Adam Tuttle from Addroid , February 16, 2009 at 1:03 p.m.

    While I’d agree that most ad servers are based on archaic client server architecture and design . I'm not sure that the problem has to do with computational power as much as the general lack of predictive algorithms. The cloud is simply a distributed architecture it is not a magic solution to cull data from all the sources and then decipher the data to predict the future. It allows for infinite scale and efficient processing but the magic you talk about comes from advanced software design.

    Ad servers are generally not limited by processor or dB power but instead are limited by their ability to understand and predict. Having more servers simply accelerates the speed of delivery and decision making I don’t understand how or what the cloud has to do with creating a predictive delivery algorithm.

    Early in the development of inventory optimization Ad Server’s borrowed from the Airline Industries ticketing systems. Systems that were designed to use demand and availability to maximize the yield of a particular seat. Advances along these lines led to the rise of companies like Ad.com and RightMedia. Since that time a number of companies and technologies have been ported over from various industries and applied to ad delivery. There is no shortage of reactive analysis for delivery. What is still missing are the magic numbers that predict future behavior based on past actions.

    Adam Tuttle

  3. Jeremiah Budzik from Acquity Ltd. , February 17, 2009 at 5:13 a.m.

    Increased horsepower (databases, processors, memory, etc) will not 'solve' the problem of inventory forecasting. The two ad serving companies I'm most familiar with, DoubleClick and Atlas each have a tremendous amount of computing power (distributed across multiple systems) driving forecasting and we're still not completely satisfied with the results we get.

    As Adam points out in his comment, the difficulty comes in improving the 'proprietary algorithms' that Larry notes in his article. The algorithms weren't invented by the ad serving companies, rather they have been labored over by Phd statisticians for many, many years before being coded into the ad server. All of the cloud power from Amazon, Google and Microsoft combined won't improve your forecasting - it will only return the same results quicker! That is, until someone can revise the algorithms.

  4. Henry Blaufox from DragonSearch , February 19, 2009 at 2:08 p.m.

    Cloud computing is a fast, lower cost and lower maintenance method for handling process intensive applications, like the ones Mr. Allen describes. The common alternative has been to buy more servers and run the apps entirely in-house. Cloud computing eliminates that purchase and maintenance cost.

    The original "cloud" may have been all that horsepower Boeing and the like made available back in the early days of data processing, when it was called time sharing (yes, I am that old.)

    Henry Blaufox
    Oxclove Workshop