How are publishers to plan their ad inventory?
Forecasting website traffic is necessary for the publisher for many reasons; and two of them can be considered primary. First of all, inventory and audience planning will ensure that the contract with the advertiser will be fulfilled to the full extent. Secondly, it is essential at the stage of shaping one’s mid-term and long-term budget.
Inventory planning is based on the amount of total and available traffic for the billing period. The accuracy of the prediction of the number of impressions per day, month, or any other period depends on the quality and completeness of the data, as well as on the calculation methods, of which we will dwell upon three:
- calculations in Excel;
- built-in advertising server forecasting tools;
- dedicated forecasting systems.
The approximate total volume of future traffic for specified parameters can be calculated manually based on reports or log files of an ad server. To do this, one will need to determine the average amount of impressions for given targeting types over several weeks (arithmetic average, moving average, linear trend, etc.). This will be the forecast for the future period, simple but at the same time inaccurate. More reliable results will require inclusion of additional factors:
- Natural fluctuations in the network. As a rule, the volume of traffic changes during a week, month, year. E.g., if there is an influx of users at the start of the working week, and on Saturday the traffic drops, it makes sense to calculate the average traffic by data on the corresponding days of the week.
- Traffic spikes. Sporadic spikes in historical data distort the typical picture, so it is better to smooth them.
- External events affecting audience behavior. An Ad Ops Manager can estimate the expected change in the volume of traffic associated with any events (New Year’s holidays, World Cup, etc.) and include them into the forecast.
These are some basic examples of additional variables, without which the forecast and real figures may diverge drastically.
To calculate the available amount of traffic, one needs to know what part of the total traffic will be occupied by running and planned campaigns. Calculation principles remain the same, but in this case data on the specific advertising campaigns is analyzed. Given that the campaign settings are regularly changing, the calculations become obsolete already by the end of the day. Predicting the delivery of a planned campaign is even more difficult, since there are no statistics on it, and therefore one has to use statistics on similar ad campaigns.
Despite the relative simplicity and the minimum necessary tools (just Excel and ad server reports), this method is far from being effective. Even for an experienced specialist, one calculation takes 30-60 minutes; its accuracy is extremely low, and forecast on complex overlaps of targeting parameters depends directly on the ad server. Without sufficient report, one has to combine those that are available, hoping that their experience would not fail them and the heuristics have been applied correctly. Inclusion of additional factors, such as frequency capping for example, ultimately turns calculation into divination.
Some ad servers provide forecasting tools as built-in functionality (for example, Google Ad Manager) – one has just to make a request and the ad server will perform all the calculations. However, the results in this case will also be far from perfect. Since forecasting has never been the primary function of an ad server, few of them have developed it at the proper level, and even those have certain limitations:
- When building a forecast most ad servers use statistics for the last 28 days. If an ad network volume changes significantly over a shorter cycle or is highly dependent on seasonality, the period of statistics from an ad server cannot be changed accordingly, and so the quality of the forecast will suffer profoundly.
- Not all targeting overlaps can be emulated in the request.
- Usually, the results of the forecast are provided in the form of summary numbers, without the possibility to dive into the details on the slices of interest.
- Not all ad servers can take into account external events or smooth one-time traffic spikes.
- As a rule, advertising servers cannot forecast for unique users.
In general, ad server forecasting modules is a much more advanced and handy tool, and it shows satisfactory results for high-level forecasts, unless you are going to include these figures into the next year business plan for your investors.
Functions that lacked the attention of ad servers were eventually developed in other software products, explicitly focused on the most accurate forecasting. Examples include Yieldex, ShiftForward, and Inventale. Such systems determine the amount of future total and available inventory taking into account traffic spikes and external events in a few seconds with an accuracy of 90% and higher.
The focus of such systems on in-depth data analysis allows the user to identify hidden patterns and recognize fraudulent traffic. They are also very flexible and adjust to specific business processes – letting the user choose any period necessary for analysis, integrate with several ad servers and additional data sources (DMP, CRM, SSP, etc.)
Unlike ad servers, dedicated systems can forecast taking into account different platforms (web, mobile, SmartTV) and any targeting types – audience, custom, keywords, frequency cap and unique users, as well as their various overlaps. This allows them to identify under-delivered and overbooked inventory, wherever it is hidden, and to determine the fair CPM for the most valuable inventory. One has to just press the button – and the system will do everything in several seconds.
Such systems usually offer some additional functions that significantly save time of Ad Sales and Ad Operations Managers and make their life easier. As an example, Inventale offers monitoring for the progress of active ad campaigns with forecast on their final delivery; inventory yield management module; monitoring for the sales plan fulfillment; inclusion of planned and booked campaigns into the forecast. Moreover, the system removes the risk of selling the same inventory twice.
Whether only high-level figures on the traffic volume are enough, or the detailed picture is essential, whether investments in specialized systems will be justified, or using traditional methods is more cost effective – each publisher decides on their own. However, modern tools allow publishers not to choose between accuracy, convenience, and efficiency – so why not take advantage of it?