by Mark Thielen | Oct 15, 2012
Publishers constantly strive for maximum utilization of their ad inventory to optimize cash flow. Their “stock” of available ad impressions however is directly tied to their site’s traffic, as well as to campaign priorities and target criteria. Therefore it’s subject to constant changes. This makes the task of looking into available inventory a rather complex one that can only be handled with the latest data-processing technology.
The days when using spreadsheets was sufficient to identify sales potentials are long gone. Today, publishers who don’t want to waste inventory depend on sophisticated forecasting engines that operate with highly complex algorithms to calculate future availability. The industry offers publishers a few choices when it comes to selecting the right forecasting engine to help them do the math. Publishers can rely on:
- their ad server's in-built forecasting capabilities
- a technology solution as part of a workflow/order management tool
- a stand-alone technology solution for digital inventory management
Since the implementation of a forecasting solution is a daunting project that often involves a lengthy and costly set-up process, the decision for the right solution obviously is not one to make lightly.
Here’s what to look out for when choosing a forecasting engine:
1 . Go for the solution that gives you maximum freedom in the use of targeting criteria
Big media brands increasingly insist on quantifiable, quality audiences, meaning that campaigns with complex targeting combinations have gone from being the exception to being the rule today. To serve their clients’ growing needs, publishers need to be able to offer them guaranteed impressions taking into account any possible set of booking parameters that constitute the advertiser’s campaign delivery goals. Often this involves dynamic targeting such as key value targeting or third-party audience targeting.
From a publisher’s perspective, the highest flexibility in defining audiences and selling them as guaranteed impressions contains the most revenue potential. However, only a few forecasting engines in the market are actually able to do key value forecasting. Even those who claim to offer key value forecasting sometimes show considerable limitations when it comes to using all possible key value operators in availability requests. For your own benefit, a good forecasting engine should give you no restrictions in the definition of key value sets, be able to integrate third-party data and allow any combination of campaign booking parameters you use in defining your products.
2. Aim for the highest possible forecasting accuracy
While looking at the marketing messages of companies offering forecasting solutions to publishers, you will realize that nearly everybody claims to have come up with the most accurate forecasting system in the industry. So what defines accuracy in this field? How do you differentiate a provider who truly fulfills his marketing promises from one who is hiding the fact that his solution merely produces good guesses?
a) Recent live data should power the model
The foundation of an accurate forecast is always good historic data. A technology solution which uses a data model based on your network’s recent live data will potentially give you more accurate results than one which only relies on a probabilistic model. To get reliable results, making a good guess based on probabilities is simply not enough.
b) Granularity drives accuracy
The granularity of the data model influences the quality of the forecasts. An ad server usually stores very detailed logs with all incoming information of an ad request, thereby generating a huge amount of raw data that a forecasting engine can utilize in its calculations. The way the forecasting engine looks at this data plays an important role in the reliability of its predictions. Those systems that manage to look at all relevant campaign parameters, keep them in context and connect them in multiple ways will generate better results than those that rely on factor-based calculation methods, which handle campaign parameters isolated from each other. On top of this, the aggregation level of the data used will impact results. A system which aggregates data for every hour is more precise than one dealing with daily or weekly averages.
c) The algorithm should mirror traffic patterns
The flexibility of the forecasting solution in letting you choose the calculation method that best fits your needs should be one aspect in your buying decision. Leaving money on the table simply because your inventory forecast doesn’t reflect your actual traffic patterns isn’t what a smart publisher should aim for. As a publisher you know that your site’s traffic is influenced by seasonality, bank holidays or any other unique event relevant to your business. Some of you might even notice a constant growth in traffic due to your site’s growing success. A forecasting system whose architecture doesn’t allow the usage of multiple algorithms or can only adapt the calculation method through cost-intensive and lengthy customization processes will be a stumbling block on your way to success.
My suggestion is to go for a solution provider who lives and breathes data. After all, it’s a matter of getting close to reality, and only those companies who best utilize technology to make your inventory’s potential visible will achieve the best results for you.