PMG Digital Made for Humans

The Nitty-Gritty of Programmatic Trading

8 MINUTE READ | January 10, 2019

The Nitty-Gritty of Programmatic Trading

Author's headshot

Abby Bollinger

Abby Bollinger has written this article. More details coming soon.

In this second piece of our 3-part series on Programmatic Advertising, we build on the basics covered on Part 1 and dive deeper into how the bidding process works and the role machines and people, or “traders”, play in running and optimizing programmatic campaigns.

As mentioned in the previous blog post, the most important part of the programmatic ecosystem for advertisers, in particular, is the demand side platform or DSP. This is where billions of impression opportunities are offered up and where advertisers customize parameters to target their audience at the most efficient price.

Once the campaign is set live, the DSP will automatically respond to bid requests sent from ad exchanges based on the specifications put in place.

The ad exchange facilitates this process between the advertisers — or agencies that manage this for advertisers — and the publishers. This process is often referred to as real-time bidding (RTB) and occurs as an “auction” within a matter of milliseconds, during the time a web page loads. RTB can be summarized in the following steps:

  1. A page begins to load in a person’s web browser

  2. Information about the user and the page is passed back to an ad exchange, where publisher ad impressions are pooled together for advertisers to bid on

  3. A bid request is sent from the ad exchange to the DSP for each impression opportunity

  4. Specifications put in place by the advertiser rules whether or not the DSP places a bid response to the exchange. For example, if you have an opportunity for someone who is determined to be male, browsing from Canada and in the market for athletic shoes, the DSP would not bid if the campaign is only targeting females.

  5. DSP(s) place a bid on behalf of advertisers, usually using data and algorithms to determine the right price to bid, depending on the campaign goals

  6. Highest bidding advertiser wins

  7. Highest bidding advertiser’s ad then loads on the page

But, how much do advertisers actually end up paying for each impression?  It’s one thing to bid $3 CPM through the auction, and quite another what the impression actually clears for, in other words, what you actually end up paying for the impression. How much the impression ultimately clears for depends on the type of auction being used. There are two primary types of auctions — second price auctions and first price auctions. Other auction types include hybrid and fixed auctions.

In a second price auction, the advertiser only pays $0.01 above the second highest advertiser’s bid. So for example, if you bid $5.00 and the second highest bid was $0.50, you would only pay $0.51. Second price auctions are how programmatic ad auctions have historically operated and are preferred by DSPs because it is less costly for the advertiser and easier to implement at scale.

In a first price auction, the winning advertiser pays exactly what they bid. If you bid $5.00 and you’re the highest bidder, then you pay $5.00, even if the next highest bid was at $0.50 like in the example above. First price auctions are popular amongst ad exchanges and SSPs because they mean more money for publishers, but also argue that first price auctions promote transparency and prevent abusive bidding just to win the impressions.

Many large ad exchanges like OpenX and Index Exchange have shifted from second price auctions to first price auctions gradually over the past year.

There also various other “creative” ways exchanges mix and match several auction types. For example, exchanges can submit two bid requests for the same impression, one marked as a first price auction, and another marked as a second price auction. If the second price auction comes in higher, they will use the 2nd highest bid as a clear price, regardless of which advertiser or auction type that is.

For example, the DSP submits $4.00 for the second price auction and $3.50 for the first price auction. The second highest bid in the second price auction is $3.00, so it would normally clear at $3.01, but since a $3.50 first price bid was also submitted, that impression would clear at $3.51. So yes, a bit sneaky, but again, likely being done to prevent gaming and abusive bidding practices.

While something of a misnomer, “fixed auctions” are basically guaranteed deals where a specific number of impressions are agreed upon at a set CPM. This not only guarantees what advertisers will spend and how much each impression will cost but also allows advertisers to curate a list of publishers they will be guaranteed to serve against, normally for their higher quality inventory.

As we’ve hinted a bit at already, each demand-side platform implements their own version of “automated bidding” in which the platform uses algorithms to find what drives performance. Machine learning is implemented to process large amounts of data in order to improve campaign performance. These algorithms look at dozens of data dimensions and can action off of these in real time to gain efficiency.

For example, if the algorithm finds that return on ad spend is 50% higher for a particular campaign when bids are won in a certain metro or at a certain time of day, etc. bids can be increased to win more impressions in those successful pockets. With so many variables to consider, only machines would be able to accurately assess all the different permutations in real time to make a more informed bid. This is why leveraging each DSPs algorithmic bidding can be so helpful.

The evolution in bidding types that we talked about earlier highlights a very important dynamic within the programmatic ecosystem — where just like the laws governing our physical realm — for every action, there’s an equal and opposite reaction. For years, DSPs took advantage of second price auctions by bidding higher than what they were actually willing to pay, in order to win that impression, and then count on the second price for them to end up paying less in the end.

Exchanges, in an effort to gain a higher share of publisher’s bid opportunities, started playing with auction types, hence the recent surge in first price auctions within the ecosystem. And now, in this game of back and forths, DSPs are in the midst of adjusting their bidding algorithms to detect the auction type and modify the final bid accordingly.

The Trade Desk has made the most progress on that front, having actually productized a feature called “Predictive Clearing Price” which automatically lowers the bid for first price auctions while claiming to maintain win rates. What’s great is that they actually report on the savings being generated on behalf of the advertiser, but of course, monetizing the feature by taking a cut of those savings (20%). Other DSPs have implemented similar features without the transparency, but usually at no additional cost.

Short answer — and of course, a bit self-serving but totally true — is, definitely not! Although machine learning and the use of algorithms can be useful for an “always on” look at granular data signals and real-time optimizations, humans are still very much needed in order to understand, communicate, and action on macro trends. People, or “traders” as those running the campaigns on the DSPs are referred to, are a necessary and underutilized piece of the programmatic landscape, primarily due to the lack of training in the trading space.

However, with proper training, an experienced trader can find creative solutions to capitalize on pockets of opportunity and in turn continuously evolve the advertising strategy which is something an algorithm, at least at this stage in the industry, is unable to provide. An algorithm is able to work with the parameters in place and bid up and down on the levers available, while a human can pivot the parameters or completely re-calibrate an entire campaign strategy to align with the results found.

They also have something a machine currently cannot assess (yet!). And that is, context.  People have a sense of all the subjective pieces of information that take place within an advertiser’s organization. For example, a person might know that while a campaign’s goal is to net a $5 ROI, the company might be willing to compromise performance for reach, and are happy to serve more impressions even if the ROI is actually $4. Or, they might know that competitors have been making a big advertising push lately, which might explain a sudden drop in performance.

For all these subjective circumstances, traders really need to think outside the box and develop creative ways to modify campaigns in order to properly address them…

Should the goal be manipulated within the platform to spend more while achieving a $4 ROI instead of a $5 ROI? Should an additional platform be leveraged to maximize reach? Should the ad creative be changed to better highlight competitive advantages?

All of these questions will always require people to address, so worry not, brave trader, as long as you keep a strong pulse on macro trends and context, you’ll be okay. At least for now.

Stay in touch

Bringing news to you

Subscribe to our newsletter

By clicking and subscribing, you agree to our Terms of Service and Privacy Policy

Happy Trading!

Related Content

thumbnail image

Consumer TrendsSEO & Paid SearchAI & MarketingData & TechnologyPlatforms & MediaDigital Marketing

Google Doubles Down on Automation, Consumer Privacy at Google Marketing Live

5 MINUTES READ | June 4, 2021

thumbnail image

Consumer TrendsStreaming & VideoData & TechnologyStrategyPlatforms & MediaDigital Marketing

Preparing for Streaming’s Growth & The Future of TV Buying

7 MINUTES READ | December 11, 2020

thumbnail image

The Potential Return to Context Targeting

2 MINUTES READ | April 24, 2020

thumbnail image

Quibi: State of the Landscape

1 MINUTE READ | April 22, 2020

thumbnail image

COVID-19 and Programmatic Advertising

6 MINUTES READ | April 21, 2020

thumbnail image

Social eCommerce is The Darling of Cyber Weekend

4 MINUTES READ | December 2, 2019

thumbnail image

Staying Proactive in the Current State of Privacy

4 MINUTES READ | October 31, 2019