Monthly Archives: December 2014

How private marketplaces actually work

By Eric Picard (Originally Published on iMedia – December 13, 2014)

Recently Ricardo Bilton wrote an article for Digiday about the difficulties that publishers have had embracing private marketplaces (PMPs). The validity of his article is arguable, and he called out a few of the buy-side platforms as causing some of the difficulty — despite the massive and growing volumes those platforms are actually driving in the PMP world. So instead of rebutting his article, let me define how these things work, and what the scope and difficulties are in making use of private marketplaces, but also what benefits can come from them.

Before we get into it, let’s talk about complex vs. complicated. They actually mean different things. Complex implies that the difficulty of embracing something is unavoidable — some things are just complex, have lots of moving parts and lots of opportunities to implement. Complicated implies that the difficulty is avoidable, and could be designed around. Private marketplaces today are both complex and complicated. We need to remove the complications.

History

Back in the dark ages of the programmatic world, let’s say 2008, publishers were wary of the newly emerging programmatic landscape. In order to convince them to put their inventory into the proto-exchanges that existed, the concept of a private exchange or private marketplace evolved. Keep in mind that up to that point, mostly the inventory that flowed on the exchanges came from ad networks daisy-chaining their inventory together. But as publishers began participating, and SSPs entered the scene, these private marketplace mechanisms were rolled out to support publisher concerns about yield optimization — and especially cherry picking and cream skimming — buying strategies that were major concerns for publishers in those early days.

As a result, the first private marketplaces were fairly simple to understand, and were nice ways for publishers to get their feet (or at least their toes) wet in the programmatic space. The basic concept was simple: Publishers could expose some or their entire inventory to an exchange or SSP. They could hand-pick which advertisers were invited to come into the private marketplace. Only those invited to have access could bid on the inventory.

The problem with this early approach was that it missed out on some very important fundamentals of exchange-based buying and selling. One important fundamental is bid density: For every impression that is exposed to an auction, you need as many bidders (buyers) as possible competing for that inventory in order to have the price reach a reasonable amount — especially in a second price auction.

What is a second price auction?

It’s a pretty simple idea, really — if three people participate in a second price auction for an Apple, all three people put in the highest price they’re willing to pay for that apple. Person A bids $1.00. Person B bids $1.50. Person C bids $0.50. Person B would win the auction, but only pay $1.01 for the Apple. The reason for a second price auction rather than a first price auction (in the example above for a first price auction, person B would still win, but would pay $1.50) is to encourage the bidder to put their true price into the auction. Second price auctions are generally understood to have less “gaming” of the auction — since the high bidder is protected from overpaying.

But in a world where only one or two advertisers are bidding on the same impression, there’s often no second price to use. So private marketplaces by nature are problematic when it comes to bid density — and many early private marketplaces ultimately failed to succeed. There are mechanisms that can be tried — for instance using a first price auction for private marketplaces — but of course this can lead to rampant gaming of the auction — and rarely will a buyer put the actual price they’re willing to pay into a first price auction. Another mechanism is price floors –which protect the publisher from having the impression fall on the floor for close to nothing — but often a PMP price floor in those days became a price, rather than a floor due to the lack of bid density.

As our industry evolved away from the original exchanges and toward real-time bidding, a whole host of new complex issues were uncovered — but also amazing new capabilities. One of the key things that this drove in the PMP world were new innovations like dynamic floor pricing — where the SSP or even the publisher ad server was able to analyze demand across the ad server, the SSP, the PMP, and the open exchange and set the floor on a per-impression basis.

As publishers got over their initial fear of programmatic selling, they began to put their inventory into the open exchange and blend the private and open bids into the same auction. Publishers quickly realized that they needed to give the buyers that had private marketplace access a set of preferences so that they would continue in the PMP rather than bounce out to the exchange. This led to all kinds of mechanisms — across various systems that have brought us to our modern programmatic landscape for private marketplaces.

Private marketplaces today

Private marketplaces today are very confusing. They’re both complex and complicated. There’s no clear and simple definition of a PMP that means exactly the same thing to everyone because there are so many ways to implement one. And depending on your ad server, your choice of SSP and/or exchange, and the buyer’s DSP, it’s fairly impossible to know in advance how the PMP will instantiate itself. Literally if you took five impressions of a PMP and reviewed them, each could be delivered completely differently from the others.

For publishers looking to start using private marketplaces today — without any legacy configurations or expectations, there are some benefits of having waited. Today PMPs are really about giving the publisher control over the way their preferred customers get treated by the auction. As everyone knows, when you have a big customer, who spends a large amount with you annually, you probably want to give them some discounts and benefits for working with you. Private marketplaces today are evolving into sets of controls for protecting the relationship with the buyer, and often for giving them either a discount, or giving them better access and control over the inventory they want to buy. It is the latter scenario — giving the buyer control — that makes some publishers very nervous, but is the real benefit of the PMP in today’s market.

In this scenario, the publisher lets their big spending customers get some additional control over defining the audience and the inventory that they have access to. Sellers frequently will bundle this additional control with a larger overall buy, or with a high minimum CPM, or with a high minimum overall budget. And publishers are finding that this approach makes everyone on all sides of the deal much happier. Everyone wins, as long as the complexity required to pull this off is embraced.

One of the biggest innovations in the programmatic world, and one that causes a lot of the complexity behind the issues this space has been saddled with, is the Deal ID. Deal ID was supposed to solve many problems in the programmatic space, but they have added another layer of complexity. The trick is to embrace the complexity without structuring things in such a way that they become unnecessarily complicated.

What is deal ID?

It became clear that while RTB was a vastly superior way to buy and sell ads than anything else we’d seen as an industry — there were touch-points between the old systems and the new systems that were confusing. Nowhere was this confusion worse than when a buyer wanted to execute a guaranteed deal over the RTB infrastructure.

But that Deal ID mechanism has now been used in much more flexible ways than its original driving intent. Think of a Deal ID as a way to prioritize a buy against supply. And the features for how you prioritize the bid vary by ad server, by exchange, by DSP, and by SSP. Sometimes the combination of each of those things leads to a different set of capabilities.

If you’re feeling confused, you’re getting the picture. This isn’t simple stuff. But that’s okay, because with complexity comes opportunity. Here’s a complex, but powerful scenario that Deal ID opens up:

All of these bids are Deal ID bids — prices are CPM:

  • Advertiser A sets up a dynamic bid that lands at $5 for the impression. Publisher floor prices this advertiser at $7.
  • Advertiser B sets a dynamic bid at $6 for the impression. Publisher floor prices this advertiser at $4.
  • Advertiser C sets up a dynamic bid for $17 for the impression. Publisher floor prices this advertiser at $20.
  • Advertiser D sets up a dynamic bid for $1 for the impression. Publisher floor prices this advertiser at $3.

In the above scenario, Advertiser B would win the auction, and pay $6 CPMs for that impression. Since one of the Deal ID bids won the auction, the impression never makes it to the open exchange. If for some reason the publisher had set the floor price for advertiser B at $7, then this auction would have flowed through to open exchange, and then advertiser C would have likely won the auction (assuming nobody in the open exchange bid higher than $17). Advertiser C would end up paying whatever the next highest bidder was willing to pay, plus $0.01.

How was that for complex? Want it to be more complex?

Some ad platforms can support a Deal ID with a dynamic bid, or a fixed price with a priority cascade. So while price mattered a lot in the example above, if one of those bids (even the low bid) was a fixed price, it would have won the auction at the fixed price. That’s how you give your preferred advertisers ways to find their preferred audience while giving them a fixed price. The trick is to negotiate well on the price on both sides so everyone gets what they want.

So while Deal ID is just one mechanism that may or may-not be part of a private marketplace, the two concepts are becoming somewhat inextricably linked together. What is a private marketplace today? It’s a complex set of interacting tools, systems, mechanisms, and approaches that can be used to give the publisher control over the prioritization of their supply against the demand represented over the exchange. Easy to understand? No. Easy to configure? No. Easy to execute against? Not yet. But worth using? Absolutely!

This complexity means power, but the complexity leads to confusion and complications. So when we have people who aren’t practitioners writing articles about very complex systems and how they are used, and then going to sources for quotes about adoption of these complicated scenarios, the answer is going to either be vague (not quotable) or clear (not accurate.) And these clearer quotes, which aren’t really very accurate in many cases, paint a picture of the space that looks like it isn’t working.

Private marketplaces give publishers control over the prioritization of buys coming from the programmatic channel. As an industry, we’re still figuring private marketplaces out — but vast and growing dollars are being spent over them in the meantime, and those buyers and sellers willing to take the time and effort to understand the complexity are winning. Yes, we need to make the execution of private marketplaces less complicated. It would be nice if we could also make them less complex, but only if we don’t lose the power that comes with the complexity. And in the meantime, the channel is growing and productive.

Advertisements

Don’t Believe The Lies About Digital Media

By Eric Picard (Originally published on AdExchanger Monday, December 1st, 2014)

For years, there has been a series of bad memes spreading throughout our industry. Some of the big ones have caused a huge amount of misunderstanding in our space.

Here’s my favorite: “There is infinite supply of ad inventory. With this overabundance of supply, the cost of inventory will be driven to zero.”

This way of thinking caused the publisher side of the industry to fear and block adoption of programmatic buying and selling until the last year or two, when it was proved to be false.

A simple truth: All ad impressions (in a nonfraudulent world) are created by a person seeing an ad. I estimate that there are 5 trillion monthly digital ad impressions in North America. If we divide this against roughly 300 million active Internet users in North America, assuming they spend on average five hours a day consuming content on the Internet, that breaks down to approximately 100 ads per hour.

So we as an industry have 100 chances an hour, or about 500 chances per day, to reach each person in North America with a digital ad. Of the 300 million people we can reach, advertisers only care about a sliver of the total audience.

Once you break down the audience to the desired number of people to reach, with the relevant targeting, the question becomes: Of the 500 daily opportunities, how many times do you want to reach that group of people? It comes down to several factors: what mechanisms you, as a buyer, can use to identify them and deliver an ad to them, the format of the ad and how effective you believe your opportunity to reach them will be.

So, no – the number of impressions is not infinite. And if we believe that some percentage of the ads in that 5 trillion monthly statistic are fake, meaning fraudulent or simply not viewable, then the number of chances to reach consumers could be much smaller, from 100 to as low as 50 ad opportunities per hour.

Suddenly the lie is turned on its head and it becomes more about maximizing the opportunity with your target audience. And for those opportunities, the cost is definitely not heading toward zero.

Not True: Ad Inventory Can Be Defined By The Publisher And Divided Into Pools Of Undifferentiated Impressions

Ad inventory is made up of a group of individual, unique ad impressions. Every impression has hundreds of points of data surrounding it. The problem with this belief is that it assumes limitations that don’t exist. Publishers define inventory in broad, relatively undifferentiated buckets, which are the lowest common denominator from a complex media plan sent with a fairly detailed RFP by the buyer.

For instance, buying a million impressions of “soccer moms” from a publisher creates a very limited view of that inventory. The range of income, interests, product ownership or geography is broad for the individuals behind those impressions. And some “soccer moms” may be worth more than others depending on an advertiser’s campaign goals.

In a world where inventory is publisher-defined, this lowest common denominator approach was the only way to operate a scale media business. But that’s no longer the case. Buy-side decisioning allows the buyer to define the inventory – and that inventory definition by nature can be more complex than ever.

So what is an impression? Ultimately an impression is a human being engaging in a monetizable experience via a computer or digital device, in a certain modality. By modality, we mean that they’re either passively consuming content (such as watching a video), actively consuming content (reading an article or email), actively participating in an interruptible interactive experience (playing a game with breaks between levels) or actively participating in a non-interruptible interactive experience (writing an email or engaged in a video chat).

All the data about the person behind the impression is captured in first-party (buy side and sell side) and third-party data platforms. And all the data about the content being consumed, and the user’s modality, belongs to the publisher. It is by matching both types of data that we can truly unlock the value of inventory. The more open and transparent we make things, the more value we unlock. By giving buyers access to the unfettered truth of inventory and the ability to peruse and pay for their desired inventory, price ultimately tends to go up, not down. The old world of limited, siloed and blocked data is responsible for this lie.

As we’ve opened up inventory sources and unlocked access to audience and modality data, the market responded by equalizing prices. As a result, publishers can make just as much money from programmatic channels as direct sales. Publishers that are allowing demand from programmatic sources to compete directly with guaranteed inventory are becoming pleasantly surprised with the results. Publishers that started early on this path are gaining some significant advantages that could be sustainable over the long haul.

Not True: Publishers Don’t Let Buy-Side Systems Access Inventory Because Of Potential Data Leakage

This is an old misconception. Back around 2005, some publishers invested in technology to enable creation of “publisher first-party” audience targeting data. They tracked individual audience members’ activity on the publishers’ sites and put these content consumption behaviors into behavioral targeting segments. They could then sell these segments as inventory definitions, rather than just selling locations.

This was very useful when publishers had small pools of valuable inventory that would sell out, such as auto-related inventory that would sell out months in advance. So publishers tracked users who read articles in the “auto content” bucket and created a segment called “auto intenders” so they could sell ads targeted to those users when they were browsing other pages of the publisher’s content. If they charged $15 CPMs for auto content ads, they would sell behaviorally targeted “auto intenders” for $10. They’d deliver those ads on pages that were probably selling for $5 CPMs, more than doubling the yield on those impressions.

The problem is that only the largest publishers with big user bases that consumed lots of their content could assemble enough valuable behavioral data. The small window into a person’s web-surfing behavior that any one publisher had access to was not enough to really create sustainable value. However, that data was created, sold and valued by the market, although it was less valuable to buyers because it originated from user activity on just one publisher, rather than data pooled across publishers.

Right around that time, the first behavioral ad networks, typified by Blue Lithium, figured out that they could supercharge their behavioral targeting segmentation by buying guaranteed targeted buys from publishers and stealing segmentation data from the publishers. They maximized reach by keeping the frequency cap as low as the publisher would allow, then dropped their own cookies on those users and added the publisher’s targeting definition to their own.

For instance, say a buy of “auto intenders” from Yahoo had a frequency cap that was set to one. If the ad network bought 1 million impressions for a $10 CPM, they would add 1 million unique users to their own cookie pool of “auto intenders” for just $10,000. They could then find those same users on cheaper sites, and eventually buy the inventory over ad exchanges for less than $1 but sell it for $8, allowing them to arbitrage the market. Since these networks could turn small expensive direct buys into feeders of their behavioral targeting pools, and then extend those buys to cheap inventory sources, publishers obviously became very concerned about this “data leakage.”

Most publishers, other than the very largest, stopped investing in their own first-party behavioral data technologies, leading to the creation of the lie that publishers are deathly afraid of data leakage. But what people missed is that most publishers simply gave up fighting this battle. Instead they partnered with third-party data providers that paid publishers for the rights to collect behavioral data.

They would then push the behavioral segments back into the publisher’s ad server so they could sell the data as part of their direct buys. The data leakage problem led to the creation of the third-party data marketplace and the tracking of users across publishers, which marketers find more valuable.

The real value that publishers can provide is not in turning the behavior of their audiences into targeting data. Instead publishers can give buy-side decisioning systems access to data about the content being consumed (category-level data) and what users are doing on those pages (modality data). They can also enable the buyer to bring their own first-party data, which is far more valuable for buyers than anything the publisher could assemble.

Everybody wins when publishers open up competition between decisions made by a demand-side or buy-side ad platform and direct buys booked through their sales force. When given the opportunity, buyers are willing to pay similar or higher rates for access to this inventory, compared to what they’d pay for publisher-packaged inventory that only offers publisher-based ad decisions. The two methods competing with one another increases the value of the inventory and maximizes the yield for each impression. It’s a “win-win” or a non-zero-sum game – a good thing for everyone.

How To Use RTB For Targeted Reach Instead Of Retargeting

By Eric Picard (Originally published on AdExchanger)

I was recently told by an executive in a position to know that 70 to 80% of revenue in the RTB space comes from retargeting. I found that stunning because it basically tells us that the RTB space is incredibly immature. If the vast majority of revenue in the space is retargeting, then nearly all the spending comes from ecommerce companies.

That means we have huge upside in this space because ecommerce companies certainly don’t make up anything near the majority of advertising spending.

Nearly 90% of advertising spend “all-up” is done on a targeted reach basis. In other words, the advertiser has come up with an ideal marketing persona (or series of marketing personas – many brands have five to 10 defined marketing personas) and their media plan is designed to reach people matching that persona. Using old-school methods, such as Nielsen or comScore, they find publishers with audiences matching their marketing personas, and that’s where they’ll buy impressions.

The problem is that this is extremely inaccurate, and wastes budget by spreading it across the whole audience that visits this publisher. On one hand, it’s wasteful because it pushes the message on audiences that don’t match campaign goals. On the other hand, it’s OK if there’s some “waste” in media spending because there’s value in getting the message in front of slight target mismatches.

Case in point: I don’t have cable at home. We watch Hulu, Amazon Prime and Netflix when we consume TV content. But recently, while traveling, I saw a few hours of TV in my hotel each night. I was shocked by the vast number of pharmaceutical ads on broadcast television – especially on the news (which I hardly watch anymore).

Most ads related to conditions I’m not facing today – so in a sense those ads were wasted. But should I ever contract one of those conditions, I’ll likely remember those products exist. Or should one of my close friends or loved ones get stricken with those conditions, I’ll recall that a medication exists and engage in conversation with them.

So yes – this broadcast brand strategy certainly does have some value. As I’ve said before: There’s value in the fact that I know Dodge Ram owners are “RAM Tough.”

On the other hand, we can be much more precise now than in the past — if you can find the data. And if you believe in the methodology that created the data, there are ways to more precisely reach your target personas and target audiences of all flavors.

Find The Right Tools

Using demand-side platforms and social media marketing tools, including the self-service tools within Facebook, it’s now possible to find your target audience in a variety of ways. You can be very narrow or very broad. You can control exactly which sites on which you’ll reach that audience, or you can simply specify on which sites you don’t want to reach your audience.

For brands that are very particular about running ads only on approved content, there is the white list – a specific list of domains matching against publishers that you specifically approve to run ads on. This does limit scale, but there’s no limit on the size of the white list you can create.  And there are vendors like Trust Metrics that you can use to build a custom white list for you, which both hones the targeting to sites that match your brand safety metrics and massively reduces fraud.

Or if you want, you can use private marketplaces to execute buys only on the sites you specifically negotiate with for access to their audiences over RTB. This has a lot of value for pharma and marketers that are extremely sensitive to running ads on sites that match their brand values.

If you want to specify a tightly targeted user base, one that is so targeted that it limits the audience size to only a few thousand users, you can do that using tools like Facebook’s advertising that lets you specify many different elements and tells you how limited the size of your audience is.

Or there are tools like Optim.al, which hones the audience and offers ways to expand or contract it. Or tools that let you find audiences similar to your targeting with less targeting but greater impression volume. (Disclosure: My company Rare Crowds does this.) Or you could use MediaMath’s built-in features to automatically find the right audience that performs best for your campaigns.

Nearly every company playing in the RTB space has functionality designed to meet the needs of advertisers that want to reach specific audiences, not just retarget people who visited your website or who are existing customers. There is the potential to reach people you haven’t reached before, find new customers and prospect for them.

The biggest growth sector for RTB this year is clearly going to be brand advertisers and those that use RTB for targeted reach — just like 90% of all media spending.