Media Optimization

I’ve been spending a lot of time over the past year or so trying to figure out where the advertising industry is headed andOptimization

while I still have many open questions, I’m quite confident that I have one specific area nailed down… and that’s around media optimization.

In context of this post, media optimization is when a media or ad ops team is changing/tweaking a media plan based on the performance of the sites based on a certain schedule.  If one site is performing well and other sites are performing poorly, the sites performing poorly would be manually called (yes, via telephone) or emailed and small tweaks would be made by the publisher – and if performance didn’t change, the site would be optimized “out” of the media plan.  Lost revenue for publisher.

Please note that in this context, media optimization is not media planning, or the upfront portion of media where the agency/brand selects the sites/audiences they want to use based on different research such as ComScore, Nielsen, QuantCast, etc.

Looking at the future of media optimization at the agency/client side, it’s heading away from human-lead to computer-lead interactions.   There are many reasons for this and here are a few of them:

  1. Humans are humans and are prone to error.
  2. Most media agencies don’t hire quantitative minded people, and many quantitative minded people don’t end up in advertising.
  3. Even the smartest human cannot process all of the information necessary to make an optimization quickly (i.e. 6 creatives * 12 sites * 24 hours in a day * 60 minutes in an hour * 13 audience segments * 4 geotargets * etc)
  4. CRM/databases are plugging into media systems and technologies will need to be bridged

The key note here is #3.  As we move into real-time buying for specific audiences on a custom list of sites, technologies are going to lead this.  Humans are going to drive the technologies by setting KPIs (key performance indicators) and managing the overall campaign, but once we tell the technologies how much we can bid per impression (how much we can vs. what we should bid is fundamentally different), they will buy/trade and secure placements.  Placements that do not work well after a statistically significant sample will be automatically optimized.

The adoption curve of making this happen within an agency or brand is fairly steep, however, there are multiple mid to large scale agencies who are starting to making inroads.  The digital media world has been trained for manual optimizations and it took years for our clients to understand this, so changing their mindset could take a bit of time.  There is a learning curve not only internally but for our clients as well, and with enough time, we’ll be having algorithms perform these optimizations.   I believe that this competency will be necessary as we roll into 2010/11 and exchanges proliferate media planners and strategists.   With media folks spending less time now on optimization, they will be able to spend more time on creating big ideas.

  • http://www.adexchanger.com/ AdExchanger.com

    Thanks for this, Darren. Absolutely agree.

    If you’re going to maximize efficiency in the media buying space, you will be buying audience in real-time at the impression level in the near future and leveraging technology.

    Moreover, you raise an important point about making time for “the big idea.” Though many think increasing automation takes the power away from the message and its craft, it’s quite the opposite as technology enables the optimization of the message like never before.

  • http://www.adexchanger.com/ AdExchanger.com

    Thanks for this, Darren. Absolutely agree.

    If you're going to maximize efficiency in the media buying space, you will be buying audience in real-time at the impression level in the near future and leveraging technology.

    Moreover, you raise an important point about making time for “the big idea.” Though many think increasing automation takes the power away from the message and its craft, it's quite the opposite as technology enables the optimization of the message like never before.

  • http://www.healthcare.com/ Howard

    Right on target. Lot's has been done in optimizing campaign, creatives, keywords, landing pages, etc. There are some very interesting technologies and companies optimizing at the media and audience level. The information on the users being passed through the media exchanges are great enablers for more efficiency and transparency. However, Darren, one thing I would put out is that media optimization isn't going to be led initially by the agencies for consumer brands. I suspect that the online performance marketers will be the ones who make full use of media optimization, before it trickles into more mainstream advertising. Any reactions? By the by, how's the growing family? =)

  • http://www.optimizeandprophesize.com/ jonathanmendez

    Great post. As you know I’m 100% invested in this optimization idea. A few thoughts:

    Interesting that you frame it around the performance of the sites and not the performance of the creative. Possibly some sites that are being optimized out of the mix could have outperformed if the creative was better targeted to their audience? I wonder about the complexity of systems that strive to do both. This should also be built so it can be optimized based on different metrics and even third party outcomes (your #4) though adding more complexity is tricky since ultimately it means more noise and/or less confidence in the data.

    The scenario you describe looks much like search – we should look there for insight and I’m not sure that fits into the idea of having to spend less time on optimization. While there is a ton of value in automated bid management and placement there is a huge human factor still involved in campaign set-up & management. A fully dynamic & realtime market requires a lot of hands on deck to monitor results make adjustments and test, test, test. All this technology is just helping us aim better. In terms FTE more dials might just mean you might need more peeps!

    Another somewhat analogous existing space to look is onsite content targeting. There has been a lot of work and investment is collaborative filtering and other algo approaches and very little yield in the way of breakthrough performance results. I’m not sure what this says but there is much that can be gleaned from those involved in that space that could be applied here as far as audience clustering and dynamic, realtime results — especially in the areas of temporal targeting that should have a huge place in media optimization (probably second only to geo).

    My overall thought is that there must be systems in place that are woven into the publishers performance goals and objectives. This can’t be stand-alone buy side. Search works because Google can serve ads that are bid at a lower CPC but get a higher CTR. The resulting impact provides mutual benefits for the publisher, the advertiser and most of all, the visitor. Without new systems that can create and optimize this shared value I’m afraid no value will be created at all.

    • http://www.darrenherman.com dherman76

      Jonathan, thanks for the comments. I talked about “site” optimization specifically but also meant to say “creative.” All too often, too much emphasis is placed on sites – when it’s really the audience & creative having issues.

      I’d be curious to see what systems you think need to be in place- maybe a “future industry” diagram of sort?

  • http://www.optimizeandprophesize.com/ jonathanmendez

    Great post. As you know I'm 100% invested in this optimization idea. A few thoughts:

    Interesting that you frame it around the performance of the sites and not the performance of the creative. Possibly some sites that are being optimized out of the mix could have outperformed if the creative was better targeted to their audience? I wonder about the complexity of systems that strive to do both. This should also be built so it can be optimized based on different metrics and even third party outcomes (your #4) though adding more complexity is tricky since ultimately it means more noise and/or less confidence in the data.

    The scenario you describe looks much like search – we should look there for insight and I'm not sure that fits into the idea of having to spend less time on optimization. While there is a ton of value in automated bid management and placement there is a huge human factor still involved in campaign set-up & management. A fully dynamic & realtime market requires a lot of hands on deck to monitor results make adjustments and test, test, test. All this technology is just helping us aim better. In terms FTE more dials might just mean you might need more peeps!

    Another somewhat analogous existing space to look is onsite content targeting. There has been a lot of work and investment is collaborative filtering and other algo approaches and very little yield in the way of breakthrough performance results. I'm not sure what this says but there is much that can be gleaned from those involved in that space that could be applied here as far as audience clustering and dynamic, realtime results — especially in the areas of temporal targeting that should have a huge place in media optimization (probably second only to geo).

    My overall thought is that there must be systems in place that are woven into the publishers performance goals and objectives. This can't be stand-alone buy side. Search works because Google can serve ads that are bid at a lower CPC but get a higher CTR. The resulting impact provides mutual benefits for the publisher, the advertiser and most of all, the visitor. Without new systems that can create and optimize this shared value I'm afraid no value will be created at all.

  • http://www.darrenherman.com dherman76

    Jonathan, thanks for the comments. I talked about “site” optimization specifically but also meant to say “creative.” All too often, too much emphasis is placed on sites – when it's really the audience & creative having issues.

    I'd be curious to see what systems you think need to be in place- maybe a “future industry” diagram of sort?

  • http://adgear.com/ Vlad Stesin

    It all kind of comes back to Albert Wenger’s post and our comments there. I’ll chime in if you don’t mind.

    A few points to consider:

    - Multivariate testing and optimization are relatively easy to do based on clicks, where such procedure would follow the same type of methodology as used in Search campaigns. But since the effects of display advertising are much harder to quantify, chances are direct click-based optimization would yield odd results (“natural born clickers” etc).

    - From my experience, large campaigns from large advertisers on large networks/sites are difficult to turn around in terms of contractual obligations, paperwork, billing, etc. This is often what puts a break on optimization, not necessarily the technology as such.

    - Agree with Jonathat that If, by definition, you can “pull out” a placement easily, sites/networks would need a way to replace it back as easily. Hence the whole thing needs to be integrated for publishers to be on board.

    Let’s also keep in mind that no matter how much intelligence you put into optimization of creative and placement, there are many factors that make it completely irrelevant. For one, we routinely see campaigns not being properly geo targeted, served to “external” markets. Many impressions are never seen at all, served below the fold or otherwise out of viewport. Banner blindness also affects at least some percentage of online population.

    Auditing and universal tech allowing to keep a close eye on who the campaign is being served to and where is first step in this direction — and really, this can be considered as being the first stage of optimization.

    Good post.

    • http://www.darrenherman.com dherman76

      Vlad, thank you for the comment and feedback.

      You raise some great points to consider. Here are some questions/thoughts based on your notes:

      1. Multi-variate testing and optimization is great in theory, but you need to truly separate out specific audiences in order to make it most effective. You raise a great point about “click-based optimization” – clicks vs. view-thru’s are different but treated either exactly the same or similar with many brands.

      2. Contracts and paperwork get in the way of some optimization, but I’ve found that Ad Networks are able to optimize within their infrastructure without much paperwork, if any. It’s up to the yield manager and publisher/network to make a campaign “work.”

      3. Agreed that auditing and universal tech allowing a close-eye on campaigns and companies like DoubleVerify are leading the charge here and providing some forward thinking agencies and clients.

    • http://greghills.com greghills

      Vlad

      As Darren mentioned, you make a great point with the view-through vs. click-through conversion. Even if you optimized off a specific action, instead of clicks, and allowed for view-through and click-through conversions, you would have to consider the view-through delay. If most of your conversions came from view-throughs, and the average view-through conversion took 7-10 days, it would greatly limit your ability to optimize.

      Interesting to see how you and Darren think differently about paperwork getting in the way of optimization…..in my experience, it really depends on how specific the contract is. If you wanted, you could issue a contract with specified impressions by day, by creative size, and by section/target. Or you could just list one line item for the entire campaign. It depends on how much freedom the planner wants to give the account manager to optimize the campaign without having to consult the planner and revise the paperwork.

  • http://shiftmarket.com vstesin

    It all kind of comes back to Albert Wenger's post and our comments there. I'll chime in if you don't mind.

    A few points to consider:

    - Multivariate testing and optimization are relatively easy to do based on clicks, where such procedure would follow the same type of methodology as used in Search campaigns. But since the effects of display advertising are much harder to quantify, chances are direct click-based optimization would yield odd results (“natural born clickers” etc).

    - From my experience, large campaigns from large advertisers on large networks/sites are difficult to turn around in terms of contractual obligations, paperwork, billing, etc. This is often what puts a break on optimization, not necessarily the technology as such.

    - Agree with Jonathat that If, by definition, you can “pull out” a placement easily, sites/networks would need a way to replace it back as easily. Hence the whole thing needs to be integrated for publishers to be on board.

    Let's also keep in mind that no matter how much intelligence you put into optimization of creative and placement, there are many factors that make it completely irrelevant. For one, we routinely see campaigns not being properly geo targeted, served to “external” markets. Many impressions are never seen at all, served below the fold or otherwise out of viewport. Banner blindness also affects at least some percentage of online population.

    Auditing and universal tech allowing to keep a close eye on who the campaign is being served to and where is first step in this direction — and really, this can be considered as being the first stage of optimization.

    Good post.

  • http://twitter.com/zachcoelius Zach Coelius

    Great post, as you know I completely agree. A lot of the concerns being raised by the other commentators are addressed with exchanges that provided scale, liquidity and transparency. We are finally at a point technically where where we can put all the pieces together and execute effective buyside targeting and optimization at scale. Think SEM with more variables, data, and levers for optimization.

  • zcoelius

    Great post, as you know I completely agree. A lot of the concerns being raised by the other commentators are addressed with exchanges that provided scale, liquidity and transparency. We are finally at a point technically where where we can put all the pieces together and execute effective buyside targeting and optimization at scale. Think SEM with more variables, data, and levers for optimization.

  • http://www.darrenherman.com dherman76

    Vlad, thank you for the comment and feedback.

    You raise some great points to consider. Here are some questions/thoughts based on your notes:

    1. Multi-variate testing and optimization is great in theory, but you need to truly separate out specific audiences in order to make it most effective. You raise a great point about “click-based optimization” – clicks vs. view-thru's are different but treated either exactly the same or similar with many brands.

    2. Contracts and paperwork get in the way of some optimization, but I've found that Ad Networks are able to optimize within their infrastructure without much paperwork, if any. It's up to the yield manager and publisher/network to make a campaign “work.”

    3. Agreed that auditing and universal tech allowing a close-eye on campaigns and companies like DoubleVerify are leading the charge here and providing some forward thinking agencies and clients.

  • http://greghills.com greghills

    Great post Darren. Your post reminded me of a quote from an X+1 whitepaper: “What Is Optimization” (http://bit.ly/Qq6aC)

    “What does it mean to say we ‘optimize’ online marketing? To many, it connotes some sort of ‘improvement’, but it is far more than a fancy word for ‘improvement’. ‘Improvement is a lnear, single step in making something better, while ‘optimization’ is a mathematically rigorous evaluation of the full range of practical options leadin gto the selection of the best available choice.

    Here is a good example: Painting a house is not the same as optimizing its resale value (though it helps). If no attempt is made to evaluate the full range of alternatives for spending time and effort on the house, you’re merely improving (without ‘optimizing’) the price your house will bring.”

    The distinction between improvement and optimization is important. Given the complexity of point #3, I would say that humans can improve, technologies can optimize.

    Media is already optimized to a certain degree, since ad networks already have the algorithms to optimize. So individual network buys are optimized in silos, but the overall plan is only improved. Optimizing across networks will bring huge gains, especially if you get a user-level view of interactions across the multiple high-overlap networks with multiple creative messages. As Jonathan and Darren mentioned, the creative is important too.

    • http://www.darrenherman.com dherman76

      Love the analogy you used.

  • greghills

    Great post Darren. Your post reminded me of a quote from an X+1 whitepaper: “What Is Optimization” (http://bit.ly/Qq6aC)

    “What does it mean to say we 'optimize' online marketing? To many, it connotes some sort of 'improvement', but it is far more than a fancy word for 'improvement'. 'Improvement is a lnear, single step in making something better, while 'optimization' is a mathematically rigorous evaluation of the full range of practical options leadin gto the selection of the best available choice.

    Here is a good example: Painting a house is not the same as optimizing its resale value (though it helps). If no attempt is made to evaluate the full range of alternatives for spending time and effort on the house, you're merely improving (without 'optimizing') the price your house will bring.”

    The distinction between improvement and optimization is important. Given the complexity of point #3, I would say that humans can improve, technologies can optimize.

    Media is already optimized to a certain degree, since ad networks already have the algorithms to optimize. So individual network buys are optimized in silos, but the overall plan is only improved. Optimizing across networks will bring huge gains, especially if you get a user-level view of interactions across the multiple high-overlap networks with multiple creative messages. As Jonathan and Darren mentioned, the creative is important too.

  • greghills

    Vlad

    As Darren mentioned, you make a great point with the view-through vs. click-through conversion. Even if you optimized off a specific action, instead of clicks, and allowed for view-through and click-through conversions, you would have to consider the view-through delay. If most of your conversions came from view-throughs, and the average view-through conversion took 7-10 days, it would greatly limit your ability to optimize.

    Interesting to see how you and Darren think differently about paperwork getting in the way of optimization…..in my experience, it really depends on how specific the contract is. If you wanted, you could issue a contract with specified impressions by day, by creative size, and by section/target. Or you could just list one line item for the entire campaign. It depends on how much freedom the planner wants to give the account manager to optimize the campaign without having to consult the planner and revise the paperwork.

  • http://noahrobinson.wordpress.com Noah Robinson

    Hi Darren,

    Totally agree that algorithms will ultimately do a much better job optimizing than humans — but only if the media strategists/analysts set up the media plan to test and optimize against the right variables.

    I think the biggest obstacle towards future-perfect optimization is the lack of supply-chain integration between clients and buying partners. We need clients to share realtime sales data with media buying agencies/exchange partners. To achieve better integration we’ll need serious security, confidendiality, and privacy protocols in place.

    Once we get to that point, the game changes dramatically.

    But to get there, we need humans, not computers :-)

  • noahrobinson

    Hi Darren,

    Totally agree that algorithms will ultimately do a much better job optimizing than humans — but only if the media strategists/analysts set up the media plan to test and optimize against the right variables.

    I think the biggest obstacle towards future-perfect optimization is the lack of supply-chain integration between clients and buying partners. We need clients to share realtime sales data with media buying agencies/exchange partners. To achieve better integration we'll need serious security, confiendiality, and privacy protocols in place.

    Once we get to that point, the game changes dramatically.

    But to get there, we need humans, not computers :-)

  • http://www.darrenherman.com dherman76

    Love the analogy you used.

  • William Taylor

    Darren,

    RE: NYTIMES.com article

    I hate to break it to you, but advertising metrics have been around a long time. In the 1930′s advertisers would use split runs in newspapers to generate the exact same type of metrics. There is always some smug “numbers” guy spouting off the same assured speech about how this is route to success…until it turns out that it isn’t. The “big idea” and “catchy phrase” the article so condescendingly refer to are persuasion. Persuasion is the business of advertising and it happens to be an art, not a science. The difference between a price off and a free T-shirt is a false choice. The real choice is imagination or the same old tire tricks that everyone else is doing. If you are going to work in advertising, you need to understand the basics.

    • http://www.darrenherman.com dherman76

      William, thanks for the note. Historically, we’ve always used persuasion based on the entire media buy – or the same message to a large “audience.” When audiences can be segmented and precise deliverance can be guaranteed, persuasion now happens almost on a 1-1 level and metrics get amplified. If we didn’t understand the basics, we would not have gotten here.

      The business is not all science – and that’s the beauty of it. We still need art to create a message that resonates with specific audience segments. A world where creative directors paired with data analysts is not that far off, and in some cases, it’s here today.

      • Zev

        Re William Taylor,

        The basics, the “art, not science” is not what Darren can add. He can add rigorous research too the contributions of the artists. The art produced can and often does have unintended effects beyond the dreams of artists. What Darren can find, is what that effect is, how strong, significant, and predictable it is, and whom it most effects. That knowledge is powerful, and totally outside of the realm of human artistic abilities.

        I can not do what you do, Mr. Taylor, but I do know that the human responses to it can be measured. We can never know the true response, or effect, but we can measure it with a surprising accuracy. (at lease, some of us can).

  • William Taylor

    Darren,

    RE: NYTIMES.com article

    I hate to break it to you, but advertising metrics have been around a long time. In the 1930's advertisers would use split runs in newspapers to generate the exact same type of metrics. There is always some smug “numbers” guy spouting off the same assured speech about how this is route to success…until it turns out that it isn't. The “big idea” and “catchy phrase” the article so condescendingly refer to are persuasion. Persuasion is the business of advertising and it happens to be an art, not a science. The difference between a price off and a free T-shirt is a false choice. The real choice is imagination or the same old tire tricks that everyone else is doing. If you are going to work in advertising, you need to understand the basics.

  • http://www.darrenherman.com dherman76

    William, thanks for the note. Historically, we've always used persuasion based on the entire media buy – or the same message to a large “audience.” When audiences can be segmented and precise deliverance can be guaranteed, persuasion now happens almost on a 1-1 level and metrics get amplified. If we didn't understand the basics, we would not have gotten here.

    The business is not all science – and that's the beauty of it. We still need art to create a message that resonates with specific audience segments. A world where creative directors paired with data analysts is not that far off, and in some cases, it's here today.

  • Zev

    Re William Taylor,

    The basics, the “art, not science” is not what Darren can add. He can add rigorous research too the contributions of the artists. The art produced can and often does have unintended effects beyond the dreams of artists. What Darren can find, is what that effect is, how strong, significant, and predictable it is, and whom it most effects. That knowledge is powerful, and totally outside of the realm of human artistic abilities.

    I can not do what you do, Mr. Taylor, but I do know that the human responses to it can be measured. We can never know the true response, or effect, but we can measure it with a surprising accuracy. (at lease, some of us can).

  • http://robertsandie.com sandieman

    This is why it’s difficult being a middle-market site. I am finding the ad networks tend to do a much better job of consistently selling. The question is when do you turn serious for selling direct to advertisers? 20M monthly uniques?

  • http://robertsandie.com/ Rob Sandie

    This is why it's difficult being a middle-market site. I am finding the ad networks tend to do a much better job of consistently selling. The question is when do you turn serious for selling direct to advertisers? 20M monthly uniques?

  • http://www.suhailabbas.com/ Suhail

    Darren, A great thought provoking post!

    Well, I almost agree. Except that it is really, really tough to make an automated system that takes decisions for optimizing the “site” and the “creative” part of the campaign with out the involvement of humans. I feel there are way too many parameters with varying degree involved in this. The system should submit its best possible solutions and leave the final step of decision making whether to follow them or not to the humans, like Jonathan said, “helping us aim better”.

    Suhail
    Secondary sales tracking

  • http://www.suhailabbas.com/ Suhail

    Darren, A great thought provoking post!

    Well, I almost agree. Except that it is really, really tough to make an automated system that takes decisions for optimizing the “site” and the “creative” part of the campaign with out the involvement of humans. I feel there are way too many parameters with varying degree involved in this. The system should submit its best possible solutions and leave the final step of decision making whether to follow them or not to the humans, like Jonathan said, “helping us aim better”.

    Suhail
    Secondary sales tracking