Advertising Yield Management: Ad Inventory Optimization

Executive Summary

  • Company: Large Media and Entertainment Network
  • Results and Impact: Millions of dollars of commercial inventory can now be sold, i.e., increased ad revenue, rather than given away for free as Audience Deficiency Units
    • CapEx: $0
  • Problem: Advertising Inventory Management: Increase Annual Advertising Revenue
    • Which ad-spots to grant to which advertisers for which audiences for which shows on what dates at what times of what durations? (there are millions of possibilities)
  • Solution: An optimization-based recommendation engine integrated into the company’s existing advertising sales software
    • This engine prescribes an exact allocation of ad spots that maximizes the projected sales value of remaining inventory left to sell at the click of a button

Context

TV Advertising: Inventory Management

TV Networks sell ad spots, which constitute perishable inventory. Specifically, once a show airs the ad spots during that show are gone and can no longer be sold. This is analogous to airlines selling air tickets. An empty seat on a flight cannot be sold once the plane takes off.

The Broadcaster’s ad inventory is further complicated by the fact that ad creatives come in different durations. For example, one advertiser might only have made 30 second commercials, while another advertiser has only made 45 second commercials. Consequently, it is not the case that any ad can be sold for any ad spot.

The majority of ad inventory is committed before the TV season starts in the form of guaranteed deals. Specifically, the advertiser pays $x in exchange for y total views (“impressions”) of its ads throughout the season. The guarantee mandates that the TV Network MUST deliver y views by the end the of season, referred to as the audience guarantee. In other words, the TV Network is truly selling audience impressions, but must allocate these impressions in the form of commercial ad time (i.e., ad spots).

Taken together, the total ad spend and audience guarantee imply a net price per impression. This rate is the financial term of the deal, which is actually stated as a price per thousand impressions in the advertising industry (CPM - cost per thousand). Reaching an agreement on the net price per impression is only the first step of a guaranteed advertising deal.

At this point, the advertiser’s commercials must be assigned to ad spots throughout the Broadcaster’s shows. This allocation of ad spots is guided by the advertiser’s requirements (e.g., specific shows, time of day), and must be agreed-upon by both parties. Each ad spot delivers some number of impressions of each audience segment.

But not all audience impressions are treated equally. In fact, only one audience demographic counts toward the audience guarantee. Since audiences across many demographics watch any given TV show, this concept of how audience credit is attributed to a deal has a major impact on the profitability of the Broadcaster’s ad inventory.

For example, suppose an advertiser spends $30 million on a guarantee of 1 billion impressions of women ages 25-54 throughout the TV season. Further, suppose an ad spot assigned to this advertiser delivers the following audiences:

  • 1 million women ages 25-54
  • 2 million men ages 25-54
  • 2 million women ages 65+
  • 1 million men ages 65+

Since the audience guarantee is for women ages 25-54, only the 1 million impressions of that demographic count towards the guarantee. The other impressions do not count at all, for any deal. It’s as if they were thrown away.

Therefore, it behooves the Broadcaster to:

  • Forecast the number of viewers of each show for every possible guaranteed audience segment. These forecasts are then imputed to a viewership figure for each ad spot.
  • Create a new set of projections, each of which is a slight reduction of its forecast counterpart.
  • Assign inventory to advertisers according to the reduced projections as well as advertiser requirements (e.g., show, time of day, time of year, commercial durations).

Why is inventory assigned using a projection below the forecast? Because audience over-delivery cannot be corrected, whereas audience under-delivery can. Specifically, if the TV Network delivers 1 million impressions over the guarantee then those impressions cannot be recovered. The advertiser simply receives additional impressions it doesn’t have to pay for.

In other words, some advertising inventory is wasted. However, audience under-delivery can be addressed using additional commercial inventory to fill in the gaps, which is discussed in the next section.

Audience Deficiency Units (ADUs)

But forecasting audience impressions is hard, especially for new shows. Consequently, broadcasters often come up short. This shortfall is referred to as an ad deal’s audience liability, expressed in audience impressions1. The nature of the guarantee means that TV Networks are still on the hook to deliver the guaranteed audience, and therefore have to grant additional advertising inventory, for free, to make good on their guarantees.

These free ad spots are known as Audience Deficiency Units, abbreviated as “ADUs.” It’s important to note that ADUs constitute lost revenue, since inventory granted as ADUs cannot be sold.

Consequently, allocating inventory for ADUs in a way that achieves the following tenets means more ad revenue for zero cost:

  • Match Audience to Deal: The densest audience segment should guide the audience for which the inventory is granted. For example, if an ad spot is projected to deliver 3 million impressions of audience segment A and 1 million impressions of segment B, then the inventory is more valuable if granted as an ADU for segment A.
  • No over-delivery: It’s possible that granting additional inventory to a deal with audience liability could bring that deal to impressions above the audience guarantee. Preventing this from happening recovers additional inventory that can now be sold.
  • Projected sales prices taken into account: Some inventory will be sold at a premium price, aside from impressions implications. For example, certain shows or dayparts (e.g., Prime Time) will sell at a premium. Accounting for the projected sales value means high value inventory will not be unnecessarily granted as free ADUs.

Achieving the above means more inventory, expressed as either impressions or commercial time, can now be recovered to be sold. Specifically, a broadcaster can increase its ad revenue by reducing the value of inventory granted as ADUs.

Advertising sales teams at TV Networks do not wait until the end of the season to grant ADUs. Instead, they grant inventory to each deal on an ongoing basis throughout the season, multiple times each week. Typically, sales teams allocate inventory that does not address a deal’s liability 100% completely, but in small portions at a time. The reason for this is twofold:

  1. Forecasting (Un)Reliability: Forecasting audience impressions is difficult, and viewership behaviors can change throughout the season. Assuming the current forecast of viewership will be valid months into the future may not be realistic. Therefore, the broadcaster could have an inaccurate view of the audience liability addressed by the inventory granted.

  2. Over-Delivery Mitigation: Difficulty in audience forecasting means over-delivery is possible. For example, if a show is more popular than predicted then individual ADUs could over-deliver on audience. If the target liability to address is 100%, then entire deals could become over-delivered. However, if target liability is lower (e.g., 20%) then it is unlikely that a deal would become over-delivered, even with ad spots with viewership higher than projected.

Unequivocally, the most profitable allocation of ADUs is one that minimizes the sales value of inventory granted, which is forgone revenue.

To attain such an allocation the Broadcaster must answer the following question:
Which ad-spots to grant as ADUs to which advertisers for which audiences for which shows on what dates at what times of what durations?

There are literally millions of combinations of deal, show, date, time, and ad spot duration. Now layer in the advertiser requirements across all advertisers doing business with the TV Network. This is immensely complicated, with millions of dollars at stake.

This complexity far exceeds what can be handled by manual processes, spreadsheets, general advice, and adages. Further, this type of problem is not a fit nor appropriate for typical software engineering and machine learning approaches.

Solution

The ADU allocation problem lends itself to a mathematical optimization approach, since the problem can be expressed in the following way:

  • Decisions: How to allocate currently unsold inventory for ADUs, by deal, show, date, time, and ad spot duration
  • Goal: Minimize the projected sales value (forgone revenue) of inventory granted as ADUs
  • Constraints:2
    • Address Audience Liability:3 The ADUs granted must deliver some minimum audience to each deal under consideration.
    • No Over-Delivery: No deal can become over-delivered on its guaranteed audience as a result of the ADUs granted.
    • Available Inventory: Limits on available commercial time must be respected. For example, if a particular combination of show, date, and time has 60 seconds of ad time available, then no more 60 seconds of inventory can be granted as ADUs for that combination.
    • Advertiser Deal Terms: Deal terms constrain permitted shows, times of day, and times of year. For example, if the deal terms do not allow a particular show then no inventory associated with that show may be granted as ADUs.
    • Valid Durations: ADU durations must correspond to the durations of the advertiser’s commercials. For example, an advertiser with 60 second commercials could not be allocated a 30 second spot.

This was productized as an ADU Recommendation Engine4 integrated with the company’s existing advertising sales software. Simply put, this engine was provided to the advertising sales team as a button. Clicking that button triggers a suite of optimization models and algorithms5 that prescribe exact inventory to grant as ADUs to all deals with audience liability.

Not only is an optimization-based system revenue-generating, but it is also extremely fast. In contrast to a multi-hour process, the engine recommends inventory in just a few minutes. This speed is meaningful because ad sales representatives grant ADUs to address audience liability multiple times each week.

But before launching the ADU Recommendation Engine, it was imperative to ensure all business requirements were reflected. Iterative model and system development was possible with stakeholder partnership. Specifically, stakeholder feedback and testing informed new model constraints based upon deal terms, without which the engine’s recommendations could not be implemented in practice.

Additionally, stakeholder feedback indicated that the engine’s solution6 must be viable in whole or in part. For this reason, the optimization models consider all deals with audience liability at once. In other words, it must be possible to grant the recommended inventory for one deal all the way up to all deals. In practice, this means that no inventory is allocated more than once (i.e., to two or more deals) if multiple sales reps click the button at the same time.

Results and Impact

This TV Network can now recover millions of dollars in commercial inventory to sell every year that would previously be granted for free as ADUs.

This stable revenue increase does NOT require:

  • Any change in the quantity of ad inventory within the TV season. I.e., there is no need to increase the ad time viewers watch, which could harm the viewer experience.
  • Any change in headcount on the advertising sales team.
  • Any capital expenditure. Optimized advertising inventory management requires zero CapEx.

In other words, the ADU Recommendation Engine dramatically increases advertising revenue without incurring any cost. It is a prime example of optimizing a company’s existing resources.



Footnotes

  1. E.g., an audience liability of 20 million impressions of women ages 65+↩︎

  2. These are basic, example constraints. The complete list longer, more complex, and confidential.↩︎

  3. More precisely, address the target portion of audience liability.↩︎

  4. This is referred to as a “recommendation” engine because the inventory is recommended to the company’s advertising sales representative. The sales rep does not have to accept the recommendation, and could grant different inventory.↩︎

  5. Also known as mixed integer programming models↩︎

  6. i.e., recommended allocation of inventory to be granted as ADUs↩︎