Why the 'Highest Bid Wins' Strategy Won't Raise Your Revenue

jordanmitchell“Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by Jordan Mitchell, vice president of product at Rubicon Project.

The concept of "holistic yield management" is in the front of many publishers' minds right now. It allows RTB demand to compete with a publisher's direct sold campaigns. The conventional thinking: By aggregating all demand within a single, unified auction, and awarding each impression to the "highest bid," a publisher makes the most money.

Sound familiar?

Well, it's the wrong approach. Awarding each impression to the highest bidder, without considering scarcity, does not make the publisher the most money.

Yield management is the process of understanding, anticipating and influencing demand in order to maximize yield or profits from a fixed, perishable resource. It's not just holding an auction and allowing price to be the sole determinant. Since the airline industry gave yield management its name, it may help to use airline seats as a parallel.

Masters Of Yield Management

Airline seats are a lot like ad impressions. The seller and the buyer value each seat differently based on factors such as time of day, destination and distance, when the seat was purchased and whether it's for business or personal travel. Airlines use multiple sales channels, and very carefully segment their customers to apply price discrimination; very few seats on the same flight are sold for the same amount.

But here's the difference. Airlines are very careful to allocate inventory to different buyers and channels at different prices, using probability models to evaluate scarcity. As a simple example, this is why they provide free first class upgrades. They've determined that the probability of selling another $300 seat in coach is  more likely than selling another $800 seat in first class, so providing a free first-class upgrade actually maximizes revenue per flight. Airlines make suboptimal decisions at a seat level, in order to optimize revenue per flight.

Translating Yield Management To Ad Impressions

In contrast, most ad servers don't consider scarcity at all in their decisions to allocate inventory to channels and buyers. Instead, they typically use a "waterfall" method of decisioning, whereby they first consider delivery schedules for direct sold campaigns, to ensure even pacing, with the remaining unfilled impressions allocated to buyers based on the highest CPM or bid. And in some cases, ad servers allow indirect buyers to compete with direct buyers on the basis of bid CPM. In other words, ad servers make optimal decisions at an impression level, which are suboptimal for overall revenue.

How is this suboptimal for overall revenue? They routinely forego scarce, high-value RTB demand in favor of direct sold campaigns, which could have been served to the next visitor. For example, if two people visit your website at the same time, when you as a publisher are 50% sold through, one of those users will be served the direct sold campaign and the other will be sent to the indirect channel. They're both worth $15 to the direct advertiser, although visitor A is worth $10 in the RTB marketplace while visitor B is only worth $1.

The publisher's ad server usually doesn't know how much each visitor is worth within the indirect channel, when considering its decision on whom to serve the direct sold campaign to. Even if it did, $15 still beats $10 and a standard auction ­– awarding the impression to the highest bid – would still result in the wrong decision, which is to serve the direct sold campaign to high-value visitor A and send visitor B to the indirect channel. The right decision is to send high-value visitor A to the indirect channel and serve the direct sold campaign to visitor B because it means the publisher earns $25 instead of $16.

To make that right decision, the ad server must recognize that a $10 impression is scarce within the indirect channel, and the publisher should take advantage of every one of those opportunities.


Awarding each impression to the highest bidder, without considering scarcity, optimizes your revenue for only that impression. It does not maximize your revenue overall.

Do's And Don'ts

Publishers should not allow ad-serving decisions to be based on price alone, or "schedule" alone. Rather, publishers need to consider all factors, including price, schedule and scarcity of supply and demand. Publishers generate the most revenue by sending scarce, high-value RTB impressions to their indirect channel, with the rest served by their direct channel, even if the RTB clearing price does not exceed the direct sold CPM.

Now many of you publishers may be asking yourselves: "If I bias the scarce, highest-value impressions within RTB to be fulfilled by RTB, doesn't that affect the performance or delivery objectives of my direct sold campaigns?"

The answer is no.

As for delivery, the core of ad server decisioning has always been around maintaining the pacing and delivery objectives of direct sold campaigns. When campaigns are "behind schedule," ad servers automatically adjust the priority until campaign delivery is pacing on schedule. That said, publisher judgment is still important in balancing RTB delivery and prioritization with direct sold campaign delivery and prioritization.

As for performance, it's important to note that direct buyers have different objectives and value inventory differently compared to real-time buyers. All buyers aren't chasing the same minute audience segment.

Direct buyers buy for audience reach and performance. Real-time buyers are often looking for very specific impressions and users, as well as the ability to only bid when they see something they like, allowing them precision, scale and performance. If you as a publisher accept click-through rate (CTR) as a proxy for performance, then consider this: Multiple analyses have demonstrated no correlation between CTR and the most important factors driving RTB bid value.

In other words, a holistic yield-management strategy that intelligently allocates impressions to the best sales channel does not affect performance of direct sold campaigns. Everyone wins.

Publisher sales channels continue to evolve as automation increases between buyers and sellers. As a result, many are considering the benefits of holistic yield management. The conventional approach offered by vendors today involves aggregating direct and indirect demand into a single system, with the impression being awarded to the highest bid.

The "highest bid wins" approach, unless it considers scarcity, does not provide publishers with optimal revenue. All buyers value inventory differently, and the scarcity of supply and demand must be considered in each decision to optimally allocate an impression to a sales channel.

Publishers that are willing to make suboptimal decisions on an impression-level basis can achieve optimal overall revenue.

Follow Jordan Mitchell (@kickstand) Rubicon Project (@RubiconProject) and AdExchanger (@adexchanger) on Twitter.

 [u1]Insert image here


  1. Good article! Rarely do we get to talk about the nitty-gritty but super important details here.

    One reason this is difficult is that the ad server rarely takes into account more than one or two simultaneous dimensions. There are many ways that the problem you describe (I particularly like the graphic you created showing the $15 vs $10 problem (what you're calling scarcity).

    This is why its important for the people in charge of yield (whether explicitly, as in a yield operations person, or implicitly - as in a sales operations or ad operations person) need to understand how ad servers make "pick" decisions.

    The problem you describe isn't just one for dynamic allocation situations where the exchange bids need to compete against guaranteed direct buys. Even within the direct buys there needs to be more intelligence applied.

    Rare Crowds can help with these things by breaking the demand down more granularly and managing price against an average for a line item - effectively treating the sub-components of a guaranteed line item like much more differentiated (and numerous) sub-line items. Thereby increasing competition, thereby increasing yield.

    This is a very different approach to standard yield optimization technologies that do this very differently - and therefore are compatible with the approach we bring to bear. i.e. We work together.

    The dynamic allocation problem you describe needs to take into account scarcity - which should be addressable in their algorithms pretty easily. There are many ways to solve it - depending on the ad server and the dynamic allocation technology being used. The stupid simple one is to bid boost rare demand in the auction such that those rare (scarce) bids basically always win. Great that you called this out - and I'm sure that all the players in that space will adjust their technology soon, given how easy it is.

  2. Interesting article, Jordan. The math makes perfect sense. We've been mediating between direct, guaranteed campaigns and RTB-driven exchanges for three years, and it's been tempting at times to subordinate the direct side, as you suggest, to optimize the exchange spot buys. The challenge with that approach, however, is that the guaranteed campaign buyer is paying a premium for the fill and CPM commitment (along with position), because they explicitly state or implicitly assume they get first look at all users, especially the better or more relevant users. If those better audiences are optimized to the exchanges first, the direct campaign buyers will arguably see their performance metrics decline and simply stop making the guaranteed commitments.

  3. Hi John, appreciate your comment and feedback. I think the challenge you point out, while certainly conventional thinking and easy to understand, involves some fallacies. First, it implies that all buyers are after the same impressions because they value inventory the same way. We know that not to be true. Second, it implies that "performance" is a zero-sum game, which we also know not to be true.

    Let's say someone sent this article URL to you and your VP of Technology via email, and you both clicked on the article at the same time. You were shopping on Amazon this morning and your colleague just wiped his cookies. If AdExchanger's ad server understood that and served the direct sold campaign to your colleague, so that you would see a retargeting ad, are you absolutely sure the direct buyer would see a performance decline or be denied better audience?

    Our belief is that with appropriately designed algorithms, that better match supply and demand instead of merely holding a price-based auction, buyers and sellers BOTH win.


Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>