Forecasting: how to get the most reliable estimates

This article outlines some things you can do to get the most out of Triton’s Inventory Availability forecasting tool.

Understanding the Forecasting Algorithm

Our algorithm assesses the activity in your network or specific stations/podcasts (if you have applied the Station filter) from the past 28 days. We monitor impressions delivered (regular, filler, and programmatic) as well as "missed opportunities" (ad requests that were not fulfilled for various reasons).

We then calculate an average by day of the week for the past 28 days. When forecasting within a given time period (defined by the date range), we multiply each past average for Mondays, Tuesdays, Wednesdays, etc. For example, when forecasting for November, we multiply each past Tuesday and Wednesday average by 5 and other days of the week by 4, and add them all together to acquire the Total number in the result of the query. Additionally, we observe all bookings within the specified date range and use them to estimate contending impressions per flight, taking into account all targeting parameters (flight date range, content running on etc.). Subsequently, we subtract the contending impressions from the total to estimate the availability of impressions for your query.

CPM vs. CPS Forecasts

CPM and CPS both work on the same algorithm; the only difference between impression- (CPM) or spot-based (CPS) forecasts is that we have to convert impressions to spots.

Usage note: CPM forecasts include CPS flights in contending, just like CPS forecasts include CPM flights in contending. Both include legacy Tap flights as competing flights.

Filters and Dimensions

Consult the TAP User Guide to see the full list of available filters and dimensions.

Ad Duration and IAB Category

When pulling a forecasting query without any filters, the results rendered are "raw" numbers. These numbers are the sum of all impressions and missed opportunities in the past 28 days. They do not take into consideration that in a break, only a specific number of ads from an advertiser or category is allowed. (Setting page > Ad Separation: default is 1 ad from the same industry/break). Nor does it take into account the length of an ad break and how many ads can play. (Example: a two minute ad break could run four 15-second ads and one 60-second ad, or four 30-second ads).

Ad duration (only for mid-rolls and In-stream publishers)

Mid-roll forecasting is based on impressions, not spots (ads), even when spots (CPS) is selected as the pricing model. When you select an Ad Duration, the forecast takes the total avails and divides it by the selected ad duration to determine the forecasted impressions. In other words, ad duration is not really a filter; it is a recalculation of avails based on the selected ad duration.

• When pulling a forecasting query for mid-rolls, use the ad duration filter to get most of the Total inventory calculation and hence your available impressions.

• Total inventory in your forecasting result is the sum of played ads and missed opportunities in the last 28 days.  

• When forecasting on Instream stations, use ad duration when possible, to make the most of the missed opportunities.

How missed opportunities are calculated for In-stream stations

When the first unfulfilled ad request comes in, duration from that time to the end of the break is sent and converted into 1 missed opportunity with that duration. Whether there are 30 seconds or 120 second remaining until the end of the ad break, it will be counted as 1 missed opportunity. For more information, see Understanding Missed Opportunities.

Including ad duration in your filter will ensure you get all the missed opportunities, as the calculation will then take into account the time of the first unfulfilled ad request and the time left until the end of the break as per the marker information.

Ad Separation

As mentioned above, without filters the query results are "raw" numbers. To make sure forecasting takes into account the industry separation or ad separation rules for specific advertisers as you have set them, use IAB category as a filter (or Advertiser, as the advertiser selected will automatically pull that advertiser’s IAB category) for a more reliable forecast.

• If the Ad separation setting exceeds the number of ads in a break, the settings will not be considered in forecasting. (E.g.: Advertiser ad separation is set with ad interval of 4 in a break, but a break only has three ads.)

• Frequency capping represents a threshold ("do not serve an ad from this flight more than twice in a break to a unique listener"). It will not override ad separation settings. If a flight is set at frequency capping of 2 ads per break and settings for ad separation were left at defaults (1 ad of same industry per break), the flight will not be eligible more than once per ad break.

Example of the effect of filtering by IAB Category:

Click to expand and see the results filtering by relevant IAB categories.

What Not To Do

• Commit to an impression goal without checking your inventory availability.

• Take the full available inventory as-is and commit to this impression goal. These are educated projections based on historical activities, not guarantees.

• Keep a flight as paused when it has been cancelled by the advertiser. A paused flight will still appear as a contending flight.

• Use both Station and Station tag filters thinking it’s using OR logic. It will use AND logic (between groups = AND, within a group = OR). If the station does not belong to the station tag you will get no results.

Troubleshooting

Inventory availability in Tap is an estimate of future availability not a guarantee, and that estimate is based on an analysis of at least 28 consecutive days of listener data.

If you are a new user of Tap and you have not been in production for at least 28 days, avails reported to you will be be inaccurate and unreliable; this is normal and expected.

If you do have 28 days of consecutive listener data and you're seeing zero avails, you should check your avails report's targeting criteria. It's possible you've targeted too aggressively by being too precise, or perhaps you've created a targeting paradox such as including New York City while excluding New York State. Or perhaps you have targeted a geographical area where you simply do not have any listeners. (Also see What Not To Do, above, for more tips.)

If you have at least 28 days of consecutive listener data and targeting or the other items mentioned above are not the problem, contact Customer Support and we will work with you to find the problem.