Fighter Pilots, Feedback Loops And Programmatic Marketing

mike-driscoll-2Data-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 Mike Driscoll, CEO and founder of Metamarkets.

In 1980, retired Air Force Col. John Boyd walked into the Pentagon and pitched a concept that reshaped how the United States trains its pilots to win dogfights: the OODA loop. It stands for Observe-Orient-Decide-Act, and it’s a framework for decision-making that is used today by everyone from Olympic athletes to Wall Street traders.

One of Colonel Boyd’s hallmark insights with OODA was this: Moving quickly through the whole decision cycle is more valuable than speeding through any one part of it. The world’s fastest trigger finger can’t hit a target that’s no longer there. Over time, pilots that cycle quickly – reacting and reorienting with each loop – will dominate their competition.

Mapping OODA To Programmatic

The OODA concept can map to programmatic advertising – specifically to a single RTB auction. A buyer observes a publisher’s bid request, orients using data, decides whether to bid and acts with a priced response.

As an industry, we’ve spent a huge amount of time and effort getting transactions to execute in real time, but we’ve given short shrift to the other parts of the cycle. As a result, we’re just scratching the surface of the potential value.

For example, if it takes six hours between launching a campaign and observing the results, a buyer can make just four adjustments to that campaign per day. Reduce that latency to six seconds and there’s time to make dozens of informed optimizations. Even if each change is slight, the cumulative impact can be enormous.

This is true whether the buyer is a man or a machine. An algorithm works in real time, but it’s built by people who can only improve it as quickly as they can observe its performance. Today, many buyers and the algorithms they employ live in the punch-card era of programmatic, waiting hours or days between taking action and observing their performance in the marketplace.

New Technology, New Workflows

There are two reasons why real-time feedback hasn’t gone mainstream. The first is that it’s an enormous technical challenge. Programmatic technology firms are data factories, and converting their output from periodic to continuous is a significant undertaking. The reigning big data platform, Hadoop, is a batch-based system that’s not natively designed for real-time data flows. While there are several alternative open-source frameworks for real-time processing, these are still nascent technologies that require a dedicated engineering team to customize and maintain. Today, each company attempts this process from scratch, without the benefit of robust and universally embraced standards.

The second reason is market maturity. Continuous feedback is a new thing, and buyers, sellers and exchanges will have to experience it to understand the depth of its benefits. As these technologies become mainstream, new organizations, expertise and workflows will develop around them. It took almost a decade after the rise of electronic trading in the financial markets before real-time reporting became standard, and we’re seeing the process happen much more quickly in our industry.

What Comes Next

John Battelle believes that we’ll find applications for today’s programmatic infrastructure far beyond advertising, and that it will become a “real-time processing layer driving much of society’s transactions.” As programmatic markets mature, we’re witnessing the emergence of the digital nervous system that supports this infrastructure.

We’ve exited the first phase of programmatic, the “make it work” phase, which was defined by laying the initial pipes that connected buyers and sellers and enabled the basic functionality of a real-time auction. We are living in the second era, the “make it fast” phase, with an emphasis on automation and efficiency.

The third phase, the “make it smart” phase, is where we are now heading: an era of intelligence where real-time feedback finally matters.

It’s not a single transaction loop that matters – it’s visibility into the full battlefield that matters. Integrating what has happened across the full collective experience. The individual transactions are worthless on their own. The hard part is getting the full picture.

Follow Mike Driscoll (@medriscoll), Metamarkets (@metamarkets) and AdExchanger (@adexchanger) on Twitter.

Enjoying this content?

Sign up to be an AdExchanger Member today and get unlimited access to articles like this, plus proprietary data and research, conference discounts, on-demand access to event content, and more!

Join Today!


  1. Thanks for the OODA loop reminder.

    On the idea of making adjustments every six (or even one) seconds instead of six hours: it may not be the right thing to do. When you’re measuring and decisioning at that rate, it’s likely the signal to noise ratio has gone down. There’s a lot of jitter and noise introduced when you view a system and sample results at that high a frequency. In fact, frequent measurement of the data magnifies the noise. (Taleb, The Noise Bottleneck or How Noise Explodes Faster than Data, 2013)

    As an advertiser, I hate waiting to make decisions. As a scientific marketer, I know that frequency won’t necessarily improve my results. I certainly don’t want to make decisions based primarily on volatility.

    How do we get the right frequency of “Observe” in the OODA loop so that we’re making decisions based on signal and not just momentary noise?

  2. Hi Mark and thanks for the comment. I agree that increasing the frequency of observations can amplify the noise. We need not look further than our overflowing Inboxes for proof. Yet, there no question that instant communication enhances our ability to make faster decisions, and because of that no modern organization would consider banning email.

    When it comes to feedback, we need not respond to every jitter, but we should have that option. I know several media buyers for whom minutes translate directly into dollars: every minute a campaign is over budget, or under-delivering, or not performing.

    You’re correct that separating signal and noise to can put a heavy burden on marketers. The hope is that in the coming years, this work will be done not by people staring at live dashboards, but by machines and smarter algorithms — or put another way, programmatically.

  3. Denise Colella

    Excellent article, Mike. However, while the tech vendors are very much in the “make it smart” phase, what we are finding is that many of the actual users are still “making it work.” So many of the publishers out there are still trying to cobble together a stack while performance and CPMs suffer. To your point, more data is just noise if it is not being actioned. Instead of focusing on real-time measuring and decisioning, (optimization shouldn’t be real time–you need critical mass to make a qualified decision) publishers should be focusing on harnessing the value of the data they have and using it to improve campaigns now–the machines to Observe and smarter algorithms to Orient, Decide and Act programmatically are already available.