2019 Digital Marketing Outlook A data-driven perspective on key market forces that will shape digital paid media in the year ahead Last Updated April 1, 2019 Introduction This report is intended to provide a data-driven view into how digital ad buyers and sellers will transact in the year ahead. It draws heavily from a market forecasting model that the Jounce Media team maintains based on a combination of public company earnings reports, public disclosures from privately held companies, and our assumptions about the state of digital advertising. The model contains historic data from 2017 and 2018 as well as a projection of 2019 ad spend. In addition to quantifying budget inflows and outflows, this report provides our commentary on the primary drivers of these shifts in ad spend. We further assess the impact of these major market trends on the technology companies that facilitate digital advertising: buy-side bidding systems, sell-side yield management systems, data management platforms, and measurement and attribution systems. We organize our 2019 market outlook into 8 key trends: ● The Size Of The Internet: While overall non-search digital marketing investments will grow to $156B globally in 2019, most publishers will experience revenue declines. ● The Changing Shape Of The Internet: Google, Facebook, and Amazon have captured 100% of the growth of digital advertising in the past three years. In addition to their scaled owned-and-operated media properties, these three companies now also power 57% of ad spend for open internet publishers. ● Two Diametrically Opposed Sales Strategies: Unlike the six walled gardens, who each operate a closed, unified auction, most open internet publisher conduct upwards of 10 duplicate programmatic auctions for each available impression. ● Platform Fragmentation & Steps Toward Orchestrated Media Buying: Scaled enterprise marketers must operate programmatic buys through at least 7 buying platforms, creating both an operational tax and a marketing performance penalty. ● Navigating Shifting Auction Dynamics: To address auction duplication and the shift to first price auctions, marketers, agencies, and buy-side technology platforms are overhauling their advertising supply chains and forming strategic alliances with trusted sell-side partners. ● Pricing Models & The Ad Tech Tax: The standard take rate pricing model is under pressure, and scaled marketers are facing a growing list of upcharges and non-transparent fees. At the same time, subscale buyers are embracing performance-based pricing with non-disclosed margins from Google and Facebook. ● Audience Targeting Headwinds: Legislation, browser privacy initiatives, and walled garden policies will cause the total addressable market for third party audience targeting to decline from its 2017 peak of $11B. © Jounce Media, LLC 2019 1 ● The Squeeze On Third Party Attribution: The same headwinds that put pressure on audience targeting will limit the utility of third party attribution. Companies who operate trusted data cleanrooms will emerge as the category winners. The Size Of The Internet Marketers will spend $581B on paid media in 2019. This includes five primary channels: digital, TV, print, out-of-home, and radio. It notably does not include three “dark pools” of marketing investments. First, our data does not address the size of venue or event sponsorships (e.g., the branding rights for Citi Field). Second, our data does not address the size of trade spend (e.g., Tide endcaps at Walmart). And third, our data does not address the size of influencer marketing (e.g., DJ Khaled endorsing Stride Gum). While we believe these are all multi-billion dollar global categories, we are not aware of any data sources that would enable us to provide robust market sizing estimates. Among the five well-tracked paid media channels, digital is both the largest and fastest growing. Global TV budgets have grown just over 1% annually, lagging behind inflation. Similarly, print, out-of-home, and radio are all flat-lining or in decline. These trends, however, reflect a generous definition of digital advertising. We classify all internet-connected media as digital advertising. We classify Pandora, as an example, as digital advertising, not radio advertising. Hulu, by our standard, is digital advertising, not TV advertising. © Jounce Media, LLC 2019 2 Most research reports will identify three subsets of digital advertising: search, programmatic, and reservations. Search advertising reflects both text-based keyword listings (e.g., a Google search for insurance) and an image-rich format called product listing ads or shopping ads (e.g., a Google search for newborn diapers). Programmatic advertising reflects non-search auction-based advertising. And reservations reflect any pre-negotiated price and volume commitments between buyers and sellers. We think it is critical to split programmatic advertising into two more granular categories: walled garden programmatic and open internet programmatic. We define walled gardens as media companies who (a) capture at least $1B of global ad spend and (b) transact auctions through proprietary bidding systems. By this definition, there are six walled gardens: Google, Facebook, Amazon, Twitter, Pinterest, and Snap. These six companies capture the lion’s share of programmatic ad budgets and more than 100% of the growth in the programmatic category. The remaining 10,000+ media properties, which we call the “open internet,” capture a declining share of programmatic budgets. We use the term “display” to represent all non-search digital advertising. This includes banner, video, and native formats delivered on mobile, desktop, and connected TV screens. The display category is now larger than search ($156B vs. $124B) and growing faster than search (18% CAGR vs. 10% CAGR). The growth of the display category is described entirely by the success of the six walled gardens. Between 2017 and 2019, the walled gardens have collectively captured $42B of net inflows, growing from a $55B category to a $97B category. During the time, ad budgets allocated to open internet publishers have held flat at approximately $60B. © Jounce Media, LLC 2019 3 Additionally, during this period, the transaction model for open internet publishers has rapidly shifted away from reservations in favor of programmatic buying. The typical publisher on the open internet monetizes its inventory in two ways. Buyers can negotiate a reservation (often called a direct buy or an insertion order) in which the marketer and publisher agree to a fixed volume of ad impressions as a pre-agreed price. Alternatively, buyers can participate in auctions that occur in real time as each ad becomes available. To guarantee access to scarce inventory, buyers may rationally engage in a reservation, but buyers preference has quickly shifted to non-guaranteed auctions. We estimate that the average open internet publisher saw reservations decline by 37% in 2018 and will see reservations decline again by 20% in 2019. There are of course exceptions; Hulu has a booming reservations business. WebMD will continue to sell reserved access to its allergies page to endemic pharmaceutical brands, but the general trend is a rapid fall in reservation revenue. The decline of reservations does not mean that advertisers are spending less money on open internet media companies. For every dollar that is allocated away from reservations, we see a dollar allocated toward open internet programmatic buying. The net result is flat open internet ad spend at approximately $60B per year over the past three years. Although ad spend allocated to the open internet is flat over the past three years, we estimate that the average publisher’s ad revenue is down. When open internet publishers transact through a reservation, they capture 100% of the marketer’s spend as revenue. But when these publishers transact through programmatic channels, technology intermediaries collect transaction fees that © Jounce Media, LLC 2019 4 create a wide delta between advertiser spend and publisher revenue. If we assume that the average “tech tax” has held steady at 50% over the past three years, we find that while advertisers have increased their open internet investments by $2B, publisher revenue has fallen by nearly $4B. We see publishers addressing this tech tax issue in two ways. First, media company executives compare the programmatic technology costs with reservation selling costs. In our experience, it is not at all uncommon for publishers to operate at a 50% fully loaded cost of sales for hand sold reservations. In a reservation-first world, publishers capture 100% of marketer spend as revenue and are then burdened with a 50% cost of sales. In a programmatic world, publishers experience a 50% technology tax, netting the same gross profit potential. The fallacy in this line of thinking is that (a) media companies cannot scale down their sales costs quickly enough to keep pace with declining reservations and (b) programmatic transactions do require human sales support, often from highly compensated technically literate account executives. The second way that some media companies address the tech tax issue is through programmatic guarantees, which blend fixed price characteristics of reservations with the technology-enabled execution of programmatic transactions. Buy-side and sell-side systems are increasingly moving toward a multi-tier pricing model in which the buyer and seller benefit from reduced transaction fees when operating one-to-one deals. Programmatic guarantees promise to give the buyer and seller workflow efficiencies, streamlined reporting, and unified frequency control at a significantly reduced fee relative to open auction transactions. In our experience, however, buyers and sellers compare the economics of programmatic guarantees to reservations, not to open auctions, and they often conclude that a classic reservation is more economically attractive to both parties than a programmatic guarantee. We are doubtful that the hype currently associated with programmatic guarantees will © Jounce Media, LLC 2019 5 materialize into significant publisher revenue. Short of the emergence of a major new media channel (e.g., connected TV), we expect reserved media buys to continue to decline in the coming years. The Changing Shape Of The Internet Display media (all non-search digital advertising) is a $156B category, which is now larger than search marketing. But unlike search budgets, which are highly consolidated among a few major search engines, display budgets are fragmented across at least seven regions of the internet. The illustration below is drawn to scale to represent the amount of global ad spend that we estimate marketers will allocate to each region of the non-search internet in 2019. The six walled gardens each operate multi-billion dollar closed programmatic systems, and these companies will collectively capture 62% of total display budgets in 2019, up from 49% in 2017. What is most striking, however, about the changing shape of the internet is the growing role of the walled gardens in monetizing the open internet. Google, Facebook, and Amazon have become the dominant revenue engines for most open internet publishers. The region above labeled “The Google Internet” represents two distinct parts of Google’s non-search advertising business. First, Google operates several owned-and-operated properties that are classic walled gardens. We estimate that YouTube, Gmail, and Google Maps (the dark green region) will collectively capture $18.8B of global ad spend in 2019. To deploy an ad campaign on these properties, Google requires the marketer to use a product called Google Ads (formerly AdWords). Unlike the open internet, where marketers can use third party bidding technology (a DSP), Google © Jounce Media, LLC 2019 6 requires using a closed system (Google Ads) to access its owned-and-operated inventory. In addition to providing access to Google’s O&O properties, Google Ads enables marketers to deploy campaigns across the open internet. Specifically, marketers who have a Google Ads account can deploy their campaign to a curated list of thousands of websites and apps that participate in the Google Display Network. Accessing the Google Display Network through Google Ads is operationally simple and carries no spend minimums, but it is also a relatively simplistic media buying system that does not fully meet the needs of enterprise marketers. For these more sophisticated marketers, Google offers a DSP called Display & Video 360. We estimate that Google Display Network will be a $21.1B business in 2019, and DV360 (excluding YouTube line items) will be a $3.3B business in 2019. The light green region above represents the combined spending power of these two businesses, and it represents 41% of total open internet ad spend. “The Facebook Internet” in the illustration above has a very similar narrative as Google’s non-search business. Facebook’s owned and operated properties, primarily the Facebook app and Instagram, are only accessible to marketers through a proprietary bidding system called Facebook Ads. We estimate that these O&O properties (the dark blue region above) will capture $66.1B of ad spend in 2019. Facebook Ads additionally allows marketers to deploy campaigns to a select set of open internet websites and apps who participate in the Facebook Audience Network. We estimate that Facebook Audience Network (the light blue region above) will be a $4.8B business in 2019. Finally, Amazon has emerged as a credible rival to Google and Facebook for both high value owned-and-operated inventory and open internet distribution. We estimate that marketers will spend $10.2B promoting their products on the Amazon website and app in 2019 (the dark orange region above) and another $4.4B distributing these promotions to open internet media placements through the Amazon Advertising Platform (the light orange region above). We forecast that Google Display Network, Google DV360, Facebook Audience Network, and Amazon Advertising Platform will collectively power $33.6B of open internet advertising in 2019. That represents 57% of all open internet ad spend, up from 34% in 2017. © Jounce Media, LLC 2019 7 The corollary to the expanding spending power of Google, Facebook, and Amazon is that the total addressable market for DSPs is declining. In 2017, independent buying platforms (mostly DSPs, but also some legacy ad networks) collectively powered $18.7B of global ad spend. We estimate that this total market will decline to $16.2B in 2019. In our view, there are two key dynamics that will shape the next 12-24 months of DSP operations. First, DSP growth is largely supported by the decline in publisher reservations. As enterprise brands pull budgets away from their reserved publisher commitments, they redeploy those budgets through programmatic channels. But as reservations dip below a $10B category, there is simply less opportunity for DSPs to capture inflows. We have not yet developed a forecast beyond 2019, but we expect to see reservations begin to reach a bottom at approximately $5B of “must buy” inventory. DSPs will have to look for other sources of growth. The second source of DSP growth is industry consolidation. Operating a DSP is largely a fixed cost business, primarily driven by the cost of processing millions of auctions per second. When DSPs like The Trade Desk manage many billions of dollars of ad spend, the allocated overhead platform costs on a per-campaign basis are manageable. But sub-scale DSPs cannot absorb the fixed cost of processing the full “bid stream.” These sub-scale platforms must instead filter the impression opportunities that they evaluate to contain costs, and even with smart filtering, campaigns that operate on sub-scale DSPs experience a scale penalty. As marketers test new platforms and discover the benefits of scaled DSPs, they re-allocate their budgets, putting further pressure on the economics of © Jounce Media, LLC 2019 8 sub-scale platforms. While the overall DSP market contracts by about $1B in 2019, we expect to see the largest DSPs continue to grow by taking share from their sub-scale rivals. Those sub-scale rivals may (a) become value added resellers of scaled DSPs, (b) merge with scaled DSPs, or (c) discontinue operations. We do not see a viable path for a sub-$100M DSP to compete beyond 2019. Two Diametrically Opposed Sales Strategies The six walled gardens conduct programmatic auctions in a radically different way from the 10,000 publishers who participate in the open internet. When a YouTube impression becomes available, Google conducts a single unified auction in which all Google Ads campaigns compete. Critically, the YouTube auction does not accept bids from third party technologies like DSPs. And even more critically, Google only conducts a single auction for each impression. (This second point may seem obvious, but it is quite different from the way open internet publishers operate. More on this later.) Similar to Google, the other five walled gardens (Facebook, Amazon, Twitter, Pinterest, and Snapchat) all conduct closed auctions for their owned-and-operated inventory. Because of the scale, targetability, and perceived must-buy nature of the walled gardens, these companies can take a “come to us” programmatic sales strategy. The largest brands would greatly prefer to access this inventory through a single buying platform (likely a DSP), but the walled gardens have the negotiating leverage to dictate terms for how marketers must transact. The open internet, however, takes exactly the opposite sales strategy. Instead of a “come to us” approach, open internet publishers take a “we’ll come to you” approach. They allow marketers to choose any third party bidding system (a DSP) to participate in programmatic auctions for each available impression. Further, the advent of header bidding has enabled open internet publishers to conduct multiple simultaneous auctions for each available impression. The schematic below represents a very common (and simplified) view of how open internet publishers sell programmatically. In this example, the publisher uses two different header bidding wrappers, each of which initiates auctions with multiple ad exchanges (commonly called SSPs). The result is that a marketer’s bidding system is asked to participate in 5 auctions for a single impression: 2 from exchange A, 2 from exchange B, and 1 from exchange C. © Jounce Media, LLC 2019 9 This auction duplication is an intentional publisher strategy that drives up yield by maximizing the probability that the publisher captures a premium bid price from a DSP bidder. But it also introduces cost and complexity for programmatic buyers. The DSP (labeled “bidder” in the schematic above) is burdened with the infrastructure cost of processing duplicate auctions for a single impression. And the marketer is burdened with determining whether to participate in all auctions, a single auction, or some subset of auctions. Solutions for this multi-auction challenge are broadly called “supply path optimization,” and we will discuss the various techniques for supply path optimization later in this report. As part of our ongoing research, we study the selling practices of thousands of open internet publishers, and we find that the average publisher partners with 15 different ad exchanges to power programmatic auctions. In total, our supply path benchmarking data identifies over 50 different ad exchanges, many of whom suffer the same scale challenges as DSPs. But unlike the DSP category, where we expect to see rapid consolidation, we do not see any near term signs of ad exchange consolidation. Marketers are rationally incentivized to execute all of their bidding through a single platform. This approach enables global frequency control, supports unified bid price optimization, and prevents bid duplication for a single impression. Publishers, by contrast, are rationally incentivized to partner with as many exchanges as possible to support auction duplication. While we think it is likely that sub-scale exchanges will suffer the same economic challenges as sub-scale DSPs, publishers will keep sub-scale exchanges afloat for much longer than marketers will support sub-scale DSPs. In particular, we think publishers will accept non-transparent take rates and will tolerate questionable auction practices from their exchange partners in order to continue to harvest the value of auction duplication. © Jounce Media, LLC 2019 10 It is also worth noting that there are some media companies who have one foot in the walled garden category and the other foot in the open internet category. LinkedIn is perhaps the best example of this phenomenon. LinkedIn’s in-feed ads are available exclusively through a closed bidding platform, but its banner ad inventory is transacted through open programmatic technologies. Other notable companies that walk the line between walled garden and the open internet include Pandora, Spotify, Quora, and TripAdvisor. Our highly speculative view is that this hybrid selling strategy may represent the most attractive future for premium publishers whose reservation businesses are in decline. Where the business’s scale supports technology investments, a closed bidding system may enable the publisher to capture 100% margin (similar to reservations) while supporting the marketer’s desire to transact in a non-guaranteed biddable environment. Platform Fragmentation & Steps Toward Orchestrated Media Buying The walled garden “come to us” philosophy imposes a significant operational tax and performance penalty for marketers. The largest digital marketers who need to reach their target customers across the entirety of the open internet must operate campaigns through at least 7 bidding platforms: one for each walled garden plus a DSP to access the open internet. © Jounce Media, LLC 2019 11 For the marketer, this necessitates 7 logins, 7 places to modify creative messages or targeting strategies, and 7 sets of cost reports to reconcile. There is a category of companies (most commonly called “preferred marketing developers”) who provide workflow automation to overcome this challenge, but most of those companies focus on value-added solutions for a single walled garden rather than a holistic solution to walled garden fragmentation. There are additionally companies called meta-DSPs who provide workflow automation for marketers who choose to use multiple DSPs. But as savvy marketers consolidate to a single DSP, we see waning relevance of the meta-DSP category. At its core, the workflow issue is simply an operational tax. This tax can potentially be solved through workflow automation technology, or it can be solved by staffing low cost agency teams. The more strategic challenge of platform fragmentation is the performance penalty of siloed bidding logic. As an acute example of the problem, a marketer who wants to achieve high reach at low frequency does not currently have a mechanism for enforcing global frequency control across its 7 bidding platforms. If a user is reached with an Instagram story, there is no mechanism to suppress further bidding for this user in Snapchat or through the marketer’s DSP. There are no technology blockers to solving this orchestration problem. Instead, the business practices of the 6 walled gardens are the blocker. Navigating Shifting Auction Dynamics There are two emerging disciplines that are reshaping the way buyers and sellers transact on the open internet: supply path optimization and bid shading. Supply path optimization (SPO) refers to a wide range of strategies that buyers take to access publisher inventory through the best available path to supply. “Best” means different things to different buyers. Some care most about participating in fair and transparent auctions. Others care most about procuring inventory at the lowest possible cost. And others care most about maximizing access to high value inventory. Each brand’s SPO implementation will be different, but the common theme is making intentional choices about the paths through which the buyer and seller transact. We see three flavors of supply path optimization, each of which will significantly change the way that programmatic buyers and sellers transact in 2019: Cost-containment SPO: shutting off duplicate bid requests with the goal of reducing DSP infrastructure costs Auction duplication creates runaway infrastructure costs for DSPs. This issue also indirectly harms advertisers in the form of higher DSP fees (i.e., the DSPs could offer more attractive pricing if not for this infrastructure cost issue). As discussed earlier, the largest DSPs can overcome the fixed cost of processing the full bidstream (upwards of 10 million auctions per second), and they are able to offer their clients unfiltered bidstream access at a competitive platform fee. The smallest DSPs, however, © Jounce Media, LLC 2019 12 don’t have enough revenue to overcome bidstream costs. These companies are under intense pressure to operate profitably, and one key lever for containing operating costs is turning off duplicate auctions. They quite literally tell some exchanges “please stop sending me bid requests for publishers X, Y, and Z.” This is the most primitive version of SPO. Principles-based SPO: consolidating bidding into a short list of trusted ad exchanges with the goal of protecting against auction manipulation Auction manipulation has been an escalating issue that more directly affects marketers. The ad exchanges who power duplicate auctions are in head-to-head competition with each other to produce the maximum auction clearing price, net of fees, to the publisher. We initially expected that this dynamic would put ad exchanges into a fierce price war with each other, but we see mounting evidence that fees are a comparatively small contributor to auction clearing prices. Instead, we see auction manipulation as the primary mechanism for exchanges to achieve above-market clearing prices. We covered bid caching in detail in the summer of 2018, and we continue to see evidence of multiple types of auction manipulation. Having observed ongoing press coverage of auction manipulation, marketers and their agencies have lost trust in the integrity of programmatic auctions, and the largest programmatic buyers are implementing strategic reviews of their ad exchange partners to better understand each company’s auction dynamics. (Some are even tracing supply paths further upstream to gain transparency into the wrappers and publisher ad servers.) Most programmatic buyers currently distribute their bidding activity across over 50 different ad exchanges. By consolidating down to a short list of trusted exchanges, the marketer protects itself from auction manipulation and potentially also secures preferred treatment in the auction. Data-driven SPO: biasing bidding into the most performant supply paths with the goal of achieving superior auction economics When a publisher makes an impression available through multiple auctions, a buyer’s ability to win through each auction often varies widely. Our research indicates that duplicate auctions for the same impression create very different economics for the buyer and the seller. We primarily monitor the buyer’s ability to win impressions at various auction clearing prices, and we conduct always-on testing across dozens of publishers to benchmark this “ability to win” KPI for all available supply paths. The average publisher in our data set has one supply path that achieves at least a 3x better than average ability to win impressions. This means that a $1.00 bid through the most performant supply path often beats a $3.00 bid through a sub-optimal path. Buyers with this information advantage can develop custom DSP bidding algorithms to secure scaled inventory access with radically better economics. We call this approach “data-driven SPO” and it is currently an experimental technique being pursued by the most savvy in-house marketers in close partnership with their DSPs. © Jounce Media, LLC 2019 13 Bid Shading Driven by pressure to produce the highest possible clearing price, the industry has gradually shifted from a pure second price auction model (winning bidder pays the price offered by the next highest bidder) to a pure first price model (winning bidder pays its full bid amount). Distorting the outcome of a second price auction to artificially inflate the clearing price was the earliest form of auction manipulation, and it helped some exchanges gain an early lead in header bidding monetization. By early 2018, many exchanges had publicly shifted to a first price auction, and in early 2019, Google announced its plans to move to a first price auction as well. While there are still some exchanges, particularly video exchanges, that operate on a second price or modified second price model, we expect the future of the open internet to be entirely first price. The move to first price auctions introduces new complexity for buyers. DSPs have traditionally been tasked with estimating the value of each impression for each marketer. In a simplistic example, DSP bidding arithmetic might look like this: If the marketer’s goal is to achieve clicks for $0.50, and the DSP predicts that there is a 1% chance that a particular impression will produce a click, then the value of that impression to the marketer is $0.005 (1% chance of generating $0.50 of business value). In a second price auction, the DSP should bid this full value with the expectation that the clearing price of the auction will be below the bid price. But in a first price auction, the DSP needs to “shade” its bid based on an estimate of the minimum price required to win the impression. Instead of sending a $5.00 CPM bid (the value to the marketer), a sophisticated DSP might submit a $4.00 CPM bid (the predicted minimum price required to beat all competitors). In the earliest days of first price auctions, DSPs weren’t very good at bid shading, and publishers experienced a temporary pop in revenue. Over the course of 2018, most scaled DSPs invested in bid shading algorithms improving their overall performance. Additionally, and somewhat counterintuitively, some exchanges also invested in bid shading algorithms. Rubicon Project in particular offers a feature called EMR (estimated market rate), in which they will reduce bid prices at the DSP’s request. While sell-side bid shading is not in the short term interest of the exchange or the publisher, the strategic bet © Jounce Media, LLC 2019 14 is that supply paths that support bid shading will achieve better marketing KPIs and will therefore capture outsized share of wallet. There is now evidence that CPMs are correcting downward as a result of bid shading. In Rubicon Project’s February 2019 earnings call, CEO Michael Barrett noted a decline in auction clearing prices and indicated that the change was “mainly attributed to buyer pricing tools in response to first price auctions.” As Google moves its auction to a first price model, there is speculation that CPMs will once again spike, though we do not subscribe to this logic. DSP spending power is quickly consolidating to a short list of highly sophisticated platforms, and all of these platforms have spent the past 12 months developing effective bid shading algorithms. Additionally, Google’s exchange currently passes a “price to beat” field in RTB bid requests, which can guide less sophisticated DSPs to properly shade their bids. Pricing Models & The Ad Tech Tax Take Rate Pricing Both buy-side and sell-side programmatic platforms most commonly price on the basis of a take rate. For each auction, the platform takes a percentage of the auction clearing price as its fee. This model is under pressure partially due to the crowded and competitive nature of the ad tech category, but mostly because of the economic burden of auction duplication. When a publisher conducts duplicate auctions for each available impression, most of the bidding activity flows to auctions that do not produce any revenue for any participants. Earlier in this report we gave an example of a publisher who conducted 5 auctions for a single impression; 4 of those 5 auctions will not produce an impression. Under the traditional take rate pricing model, both the DSPs and ad exchanges who participate in these 4 unproductive auctions are burdened with infrastructure costs but generate no revenue. The publisher carries the yield upside of auction duplication but none of the cost burden. This misalignment of incentives causes publishers to flood the market with duplicate auctions and that flooding of duplicate auction pushes many DSPs and ad exchanges to negative margins. SaaS Pricing One potential solve for this cost issue is to restructure pricing models to better align incentives of customer and platform. On the buy side, both Beeswax and Adelphic (Viant’s DSP) are charging customers subscription fees that are tiered in proportion to the number of auctions processed by the platform. Buyers who need full bidstream access are burdened with the cost of auction duplication, and buyers who can successfully operate with a filtered set of auction opportunities benefit from the associated platform cost savings. On the sell side, Pubmatic is establishing similar arrangements with publishers. © Jounce Media, LLC 2019 15 Usage-Based Upgrade Pricing A less elegant, but much more widely adopted solution is to expose opt-in features that improve marketing performance and unlock incremental high margin platform revenue. The Trade Desk has been particularly successful at expanding its margins through value-added services. Customers of The Trade Desk pay a market competitive take rate plus usage-based fees for the following opt-in features: ● Third party audience targeting ● Brand safety, viewability, and contextual targeting ● Automated bid price optimization ● Bid shading ● Access to log-level data for custom analytics In aggregate, usage of these services lifts The Trade Desk’s effective take rate from the 9-12% found in most of their contracts to the 20% reported in their SEC filings. We see similar usage-based upgrade pricing strategies for most tier 1 DSPs including MediaMath, AppNexus, Adobe, and Amobee. Among the leading DSPs, only Google DV360 does not impose these upcharges. Performance Pricing The final solution, which in our view is a primary driver of the success of Google Display Network, Facebook Audience Network, and Amazon Advertising Platform, is to price on the basis of marketing performance. Each of these platforms offer a range of pricing models, including performance-based models. Marketers can choose to pay for impressions (the classic DSP model), but can also choose to pay for clicks, website visits, app installs, and a variety of other outcomes. Publishers, of course, always expect payment for each impression, and this disconnect in the buyer and seller transaction mechanism creates arbitrage opportunities for the enabling technology platforms. Earlier in this report we gave an example of a marketer who valued clicks at $0.50, and we worked through the arithmetic for how that marketer’s performance goal would translate to a CPM bid. But imagine the marketer chose to execute its campaign via Facebook Audience Network instead of a classic DSP. In this case, the marketer would have the option to transact on a CPC (cost per click) basis. Facebook would carry the risk of purchasing impressions from a publisher, but if Facebook can optimize its bidding strategy to produce a high volume of clicks with a small cash outlay to publishers, its margins are uncapped. It is worth noting, Criteo also offers a performance based pricing model (typically a cost per post-click sale), and yet its business is demonstrating year-over-year declines. There are many factors that explain Criteo’s challenges, but in our view, a major driver is a lack of product market fit for its performance pricing model. Specifically, we see mounting skepticism from scaled enterprise brands of performance pricing. These marketers have concluded that (a) they can achieve the same outcomes with lower effective fees through self-service transparent DSPs and (b) the KPI against which Criteo © Jounce Media, LLC 2019 16 measures its success is flawed and creates room to claim credit for sales that would have happened anyway. The unique success of Google Display Network and Facebook Audience Network (and to a smaller degree Amazon Advertising Platform) is driven by their ability to combine performance based pricing with a service model that supports millions of small to medium size marketers. Facebook in particular has been successful in winning platform adoption from the long tail of small marketers. The chart below reflects the number of global advertisers who are actively buying on a select set of bidding platforms (each dot represents a press release). Facebook most recently announced 7 million active advertisers. Google last disclosed its active advertiser count in 2014, and we estimate 4-5 million advertisers currently use Google Ads. By comparison, at its peak in 2017, Criteo disclosed that their platform was used by 18,000 advertisers. Relative to the DSP competitive set, that was an impressive number, but relative to Google and Facebook, it is a rounding error. The classic DSPs (including Criteo) cater to scaled enterprise brands with multi-million dollar budgets. These marketers demand financial transparency and robust attribution (more on this later). Google and Facebook (and increasingly Amazon) also cater to millions of long tail advertisers who are in need of an “easy button” and are very happy to de-risk their media buys through performance based pricing. Performance pricing is also a strategy that we see increasingly adopted by sell side technology companies. Multiple ad exchanges now lead with a flexible take rate model in which they take a variable and non-disclosed margin on each auction, subject to a maximum rolling average. Most of Google’s contracts are structured in this fashion, and we see growing appetite among the independent exchanges to move to this pricing model. Under a flexible take rate, the exchange’s role shifts from a yield optimization platform to a source of demand. By integrating multiple demand sources (duplicate exchanges plus directly-integrated ad networks and DSPs), the publisher creates market pressure to surface the highest possible bid prices, reducing the requirement for a low transparent fee. © Jounce Media, LLC 2019 17 We view the performance pricing model on both the buy side and the sell side as the most attractive option for publishers and for small to medium size marketers. Enterprise brands, however, are increasingly focused on untangling supply chain economics to ensure they are not exposed to above-market take rates. Either through custom arrangements with their DSPs and ad exchange partners or through emerging auditing platforms like Amino Payments and Lucidity, we expect the largest programmatic marketers to focus their bidding through supply paths that have fully transparent margins. Audience Targeting Headwinds In late 2017, Winterberry Group released a report in collaboration with the IAB and the DMA called “The State of Data.” The report provided detailed estimates on the total size of audience targeted media buying, and it is the most comprehensive public data available for what is otherwise a poorly tracked digital advertising sub-segment. Based on self-declared financial data from companies who operate in the audience data space, Winterberry estimated that in 2017, marketers in the US spent $20B on audience-targeted digital advertising. In our view, the Winterberry report greatly over-states the size of the market. When we apply reasonable assumptions to our 2017 *global* market sizing numbers, we reach only $11.4B. In our view, it is simply not possible to believe that audience-targeted digital advertising in the US was a $10B category in 2017. © Jounce Media, LLC 2019 18 More critically, we think there are three factors that have caused (and will continue to cause) the overall size of the audience targeting market to decline since 2017: legislation, browser policies, and walled garden policies. Legislation On May 25, 2018, the General Data Protection Regulation (GDPR) took effect in the EU, shifting the consent framework from opt-out (consumers are tracked and targeted by default) to opt-in (consumers are not tracked and targeted by default). Data from the Reuters Institute of the Study of Journalism indicates that the presence of third party tracking technologies on European websites declined by 22% between April 2018 and June 2018. We’ve also receive anecdotal insights from buy-side and sell-side technology companies who have indicated that consent rates (i.e., the percentage of impressions that permit audience targeting) hover around 70%. We estimate that 20-30% of European open programmatic spend no longer supports audience targeting. As the California Consumer Privacy Act (CCPA) and other GDPR-like legislative measures take hold in markets beyond Europe, the addressable market for audience targeting faces mounting challenges. Browser Policies In 2017, Apple began introducing technology in its Safari browser that limited the ability for advertising technology companies to track user behavior. This technology is now called Intelligence Tracking Prevention (ITP), and it has become increasingly sophisticated and aggressive in blocking the ability for data brokers to collect and distribute audience data. Although Safari has negligible desktop web market share, over 25% of mobile web browsing happens on a Safari browser. In mid-2018, Mozilla’s Firefox browser released its equivalent to ITP, and the company has continued to advance its anti-tracking technologies. Additionally, in early 2019 Google began to publicly signal that it is considering deploying similar anti-tracking technologies in its Chrome browser. We estimate that the current state of anti-tracking technologies (Safari + Firefox) handcuffs audience targeting for approximately 20-25% of global internet traffic. If Chrome were to adopt anti-tracking technologies, we estimate that number would climb to nearly 80%. There will of course be workarounds to anti-tracking technologies but the history of Safari ITP’s cat-and-mouse game with ad tech vendors suggests to us that browsers have the upper hand in controlling user data. Walled Garden Policies The most significant factor influencing our view that third party audience data is in decline are the moves seen by walled gardens to restrict third party data usage. For as long as we have tracked the market, Google Ads has severely limited the ability of advertisers to utilize third party audience data. Specifically, Google only permits advertisers to use third party data to segment their customer lists (e.g., use Experian data to identify current customers with high © Jounce Media, LLC 2019 19 household income). Advertisers cannot use third party data to target ads to new prospective customers. Similarly, Facebook has now imposed significant limitations on the ways in which advertisers can activate third party data in Facebook Ad Manager. As of late 2018, Facebook no longer exposes a catalog of third party audiences (called “Partner Categories”) in their user interface. Marketers can continue to use third party data for targeting, but the process now requires the marketer to (a) enter a contractual agreement with the data provider, (b) assume legal liability for privacy compliant data collection and use, (c) manage the technical process of loading data into Facebook Ad Manager, and (d) manage payment to the data vendor. As Google and Facebook gain market share, these policies have an outsized impact on the total addressable market for third party audience data. Google Ads and Facebook Ad Manager platforms now control more than 85% of non-search digital advertising ($106B of a $156B category), and while third party audience data is compatible with these two platforms, we think usage is well below 5% of the ad spend flowing through them. There are too many unknowns for us to determine a precise estimate of the total size of the third party data category, but we estimate that less than $10B of global advertising investment will be powered by third party data in 2019. The decline in audience targeting will have three significant knock-on effects in 2019. First, programmatic marketers will become increasingly dependent on walled gardens, whose closed systems are much more insulated from audience targeting headwinds than the open internet. Second, programmatic advertising on the open internet will become increasingly dependent on alternative targeting strategies including geo-fencing and contextual targeting. And third, open internet publishers will further increase ad load in order to offset the sharply lower CPMs associated with inventory that does not support audience targeting. The Squeeze On Third Party Attribution The three factors that put pressure on third party audience targeting also put pressure on third party attribution. Any force that limits the ability for third party technology to track and target users also limits the ability for marketers to measure the future behavior of users who are exposed to advertising. While much of the market-facing messaging from attribution companies has focused on sophisticated data science, the much more practical point of differentiation is access to marketing data. And access to marketing data is under pressure due to legislative changes, browser privacy initiatives, and walled © Jounce Media, LLC 2019 20 garden data sharing policies. Imagine a user whose path to purchase includes the following six marketing touchpoints: If this were a non-consented user in the EU, it would be a violation of GDPR to track these six ad exposures and the resulting sale with a common ID. The marketer could count how many impressions its campaigns serve in total, and it could count how many sales it generated in total, but it could not link exposure and sales data at the user level, even using an anonymous ID. Even if this user were fully consented, browser policies might prevent third party attribution. If, for example, this consumer used a Safari browser, each touchpoint would be tracked with a new cookie ID. Instead of a single user with six marketing touchpoints followed by a sale, the marketer’s data would appear to record seven different users: six who each saw a single impression and one who made a purchase. And finally, if this example were for a consented user on a tracking-friendly browser, the policies of Google and Facebook would prevent the marketer from capturing a complete understanding of user behavior. Google’s attribution system (Google Campaign Manager, formerly DoubleClick Campaign Manager) does not have the business permissions to track Facebook impressions. Similarly, Facebook’s attribution system (Facebook Advanced Measurement, formerly Atlas) does not have the business permissions to track YouTube impressions. While we do not see a solve for legislative and a browser headwinds, we do see the emergence of data cleanrooms as a solve for walled garden data sharing issues. A short list of companies (most notably Oracle and Neustar) have negotiated limited rights with Google, Facebook, and the other walled gardens to perform unified analytics that quantify the combined impact of multiple ad exposures on future consumer behavior. In our view, attribution winners and losers will be determined primarily by their ability to negotiate trusted data sharing relationships with the walled gardens. © Jounce Media, LLC 2019 21 Research Schedule We update this research on a quarterly basis to align with the schedule of public disclosures from publicly traded media and technology companies. Subscribers to our research can expect updates on the following dates: Date Market Sizing Model Output Report Output March 1, 2019 Initial 2019 forecast + actualized 2018 data Comprehensive report detailing all Jounce market sizing estimates and key industry trends June 1, 2019 Updates to all 2019 assumptions based on Q1 earnings reports and public disclosures Memo describing key changes to original model and trends analysis September 1, 2019 Updates to all 2019 assumptions based on Q2 earnings reports and public disclosures Memo describing key changes to original model and trends analysis December 1, 2019 Updates to all 2019 assumptions based on Q3 earnings reports and public disclosures Memo describing key changes to original model and trends analysis March 1, 2020 Initial 2020 forecast + actualized 2019 data Comprehensive report detailing all Jounce market sizing estimates and key industry trends © Jounce Media, LLC 2019 22