Technology White Paper
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology Outlined: Ad Price Prediction Technology Benefits: • Improve monetization of every single publisher ad impression • Solves ad network defaulting problem for publishers • Two new technologies improves algorithm accuracy • Works with ad networks, ad exchanges, and non-guaranteed insertion orders Technical Level of White Paper:
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
Table of Contents Executive Summary.......................................................................................................... 3
Price Prediction: 2nd Generation The Need for Real-Time AdAd Revenue Optimization.......................................................... 4 Ad Revenue Optimization for Publishers
Three Levels of Ad Optimization....................................................................................... 6 Ad Price Prediction: How It Works..................................................................................... 8 Ad Price Prediction Publisher Case Studies.................................................................... 12 Conclusion....................................................................................................................... 12
Technology Outlined: Ad Price Prediction Technology Benefits: • Improve monetization of every single publisher ad impression • Solves ad network defaulting problem for publishers • Two new technologies improves algorithm accuracy • Works with ad networks, ad exchanges, and non-guaranteed insertion orders Technical Level of White Paper:
Published by PubMatic, 2009
2
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
Executive Summary The non-guaranteed segment of online advertising is the highest-growth online advertising category and will reach $11 billion by 2013, according to a recent in-depth report by ThinkEquity. Ad Price Prediction: 2ndadGeneration Rapid innovation by companies within the ecosystem, particularly networks and ad exchanges, is enabling publishers to significantly increase the ad revenue made their non-guaranAd Revenue Optimization forfrom Publishers teed inventory. Nonetheless, challenges exist to maintain sustainable growth. Along with the growth of non-guaranteed ad inventory, the rise in the number of ad networks and ad exchanges over the past few years has created a need for them to diversify themselves by focusing on different audiences and targeting capabilities. For example ad network “X” might be better suited than ad network “Y” to monetize a specific ad impression while ad network “Y” might be better able to monetize another impression than ad network “X.” Publishers can benefit from this diversification, and maximize the value of each ad impression, if they have the ability to match the optimal ad network or ad exchange to each impression in real-time. Over the last three years with a team of over 40 engineers and statisticians, PubMatic has developed optimization technology and invented a whole new category of service provider that enables publishers to do this. Technology Outlined:
PubMatic developed the first and only real-time optimization solution in 2006. Since then, PubAd Price Prediction Matic has collected hundreds of billions of data points through a machine learning approach Benefits: and has introduced two new Technology technologies in 2009 that have enabled the technology to enter • Improve monetization of every single Ad publisher into a new, more precise phase of real-time optimization: Price Prediction. ad impression • Solves ad network defaulting problem for publisherswith every ad impression, Ad Price Prediction matches the optimal ad network or ad exchange • Two new technologies improves algorithm accuracy on behalf of the publisher, in •real-time. It enables the publisher to significantly grow their ad Works with ad networks, ad exchanges, and revenue as they also manage increasing of non-guaranteed inventory. non-guaranteedamounts insertion orders
For large online publishers with millions Technical Levelofofads Whiteshown Paper: per day, this means that millions more of their advertisements are optimized than otherwise would be if they used a manual ad operations approach to working with ad networks. This white paper describes in detail why PubMatic developed the sophisticated algorithms that power the Ad Price Prediction technology, how it increases publisher ad revenue, and several publisher case studies.
Published by PubMatic, 2009
3
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
I: The Need for Real-Time Ad Revenue Optimization Publishers that seek to increase their ad revenue from non-guaranteed inventory face significant challenges. While limited ad operations could be performed on a daily or weekly basis to Price based Prediction: 2nd Generation improve revenue, a real-time,Ad technology solution that not only helps manage ad networks, but also ensures that the publisher is getting the most revenue possible for every single Ad Revenue Optimization for Publishers impression, is needed. Five reasons why a real-time solution is needed to ensure publishers get maximum ad revenue 1. Static Ad Network Daisy Chains Are Ineffective: Ad pricing from ad networks changes constantly throughout the day and with a manual solution the publisher often doesn’t have the highest paying ad network at the top of their static daisy chain. 2. More Ad Networks Need to Compete for Every Single Ad Impression: With a manual solution, managing multiple ad networks is challenging. The result is that publishers often don’t have enough relationships with ad networks that specialize in Technology monetizing different segments of theirOutlined: audience. Therefore, ads are often delivered from ad Ad Price Prediction networks that value the ad impression less than a different ad network would that is trying to reach that specific audience. Technology Benefits: • Improve monetization of every single publisher ad impression 3. Ad Networks Default Often, And It’s A Big Problem: • Solves ad network defaulting problem for publishers PubMatic has found that ad• Two networks defaultimproves 56% ofalgorithm the time on average and as much as new technologies accuracy 87% of the time according to a study in April of • Works with adconducted networks, ad exchanges, and2008. As ad targeting becomes non-guaranteed insertion orders increasingly dependent on both the user and the ad context, defaulting will only increase. Technical Level of White Paper:
4. Low Quality Ads Have Low Click-Through Rates: Managing multiple ad networks with a completely manual solution is challenging. Fewer ad networks mean the options for ads that can be served are limited, and that can result in served ads that are unattractive to the user which negatively impacts the click through rate. 5. One Solution Is Needed for Non-Guaranteed I.O.s, Ad Networks, and Exchanges: Publishers increasingly need to manage all of their non-guaranteed demand, whether it’s from ad networks, ad exchanges, or direct advertiser insertion orders, from the same bucket of non-guaranteed inventory.
Published by PubMatic, 2009
4
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
PubMatic identified the challenges that publishers face at the onset of building out its ad revenue optimization technology, and continues to advance it based on the needs of the publisher and market growth.
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers Real-Time Technology Provides a Long-Term Monetization Strategy for Non-Guaranteed Inventory PubMatic’s Ad Revenue Optimization Advances With the Growth of the Market PubMatic Real-Time Technology Releases
Non-Guaranteed Segment Growth In Online Advertising $11 B in publisher ad revenue, 34% of total publisher display ad revenue by 2013**
30% non-guaranteed ad revenue growth for 2011**
2013 Ad Price Prediction
2nd Generation Ad Optimization for Publishers
15% non-guaranteed ad revenue growth - 2009** Technology Outlined: $4.1 B in non-guaranteed publisher Ad Price Prediction ad revenue in 2008**
Bidding API for ad networks - 2009 Frequency optimization - 2008
Default optimization - 2007 Technology 30% of publisher inventory sold Benefits: through ad networks• -Improve 2007* monetization of every single publisher ad impression PubMatic pioneers ad optimization • Solves ad 2006 network defaulting problem for publisherslaunches real-time category, • Two new technologies improves algorithm accuracy solution for publishers - 2006 • Works with ad networks, ad exchanges, and non-guaranteed insertion orders
*Bain/IAB Digital Pricing Researcg, August 2008 **ThinkEquity - The Opportunity in Non-Premium Display Advertising, May 2009 Technical Level of White Paper:
PubMatic’s Machine Learning Approach Machine Learning is based on algorithms that improve automatically through experience. It includes data-mining that processes more than 100,000 data transactions per second. • PubMatic has over 6,000 publishers using the optimization platform, which continually provide rich data that contributes to the machine learning. • The longer machine learning is working, the more precise and accurate it becomes. • The data collected through machine learning is used for predictive modeling and is the basis of PubMatic’s Ad Price Prediction technology
Published by PubMatic, 2009
5
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
II: Three Levels of Optimization No other company offers optimization in real-time
$
Manual In-House Ad Operations
Technology Outlined: Ad Price Prediction
Real Time
$$
Weekly or Monthly
Publisher Revenue
$$$
Daily or Weekly
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology Benefits: • Improve monetization of every single publisher ad impression • Solves ad network defaulting problem for publishers • Two new technologies improves algorithm accuracy • Works with ad networks, ad exchanges, and non-guaranteed insertion orders
Manual Outsourced Ad Operations Technical Level of White Paper:
Automated Algorithms
+
Operations Support (Real-Time Optimization)
Published by PubMatic, 2009
6
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
In order for large publishers to truly maximize their ad revenue made from ad networks and exchanges, optimization is needed. Publishers can optimize in three principal ways with varying degrees of success:
Ad Price Prediction: 2nd Generation 1. Manual In-House Ad Operations (Weekly or Monthly Optimization): Most large publishers have an Revenue ad operationsOptimization team that works directly with ad networks to Ad for Publishers manually optimize them. They do this by logging into the ad networks and finding historical pricing and then setting up their “daisy chains” accordingly. The frequency of the optimization usually ranges from weekly to monthly, depending on human resources. This type of optimization does provide revenue lift in the vast majority of cases, but the inherent problem is that publishers are using historical data and are limited to a very small number of data points by which they can make optimization decisions. As the number of ad network relationships increases, this approach requires correspondingly more human resources to optimize. 2. Manual Outsourced Ad Operations (Daily or Weekly Optimization): In an effort to escape the resource trap of manual in-house ad operations, some publishers outsource the management of ad network relationships to third party vendors. There are often resource and expertise benefits, an outsourced service provider has typically identified Technologyas Outlined: Ad Price Prediction best practices, has ongoing relationships with key ad networks, and can often provide human resources at a cheaper cost. Technology Benefits: • Improve monetization of every single publisher ad impression However, despite the efficiencies gained in resource cost,for there is typically only marginal im• Solves ad network defaulting problem publishers • Two new technologies accuracy provement in revenue that is generated from theimproves use of algorithm third party vendors. These vendors rely • Works with ad networks, ad exchanges, and on the same historical data andnon-guaranteed limited number of data points to make ad serving decisions, and insertion orders
as a result cannot significantly increase publisher revenue. Technical Level of White Paper:
3. Automated Algorithms + Operations Support (Real-Time Optimization): This solution is the only solution that can best monetize every single ad impression. Having ad operations support is critical to a publisher in order to simplify ad network management and ensure that unwanted ads do not appear on their site, but only real-time algorithms can predict which ad network will pay the most for any given impression, 24 hours a day, 7 days a week. Real-Time algorithms can use significantly more data points to make ad serving decision, such as geography, frequency, context, demographic information and more. In addition, these algorithms can make a unique decision in real-time for each and every ad impression. More data and real-time decisions yield significantly higher publisher revenue.
Published by PubMatic, 2009
7
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
III: Ad Price Prediction: How It Works PubMatic’s Ad Price Prediction technology decides, in real-time, which ad network, ad exchange, or non-guaranteed insertion order is best able to monetize an ad impression for a Ad Price Prediction: 2nd Generation publisher.
Ad Revenue Optimization for Publishers
Page Request and Ad Impression Analyzed
Ad
ork
tw Ne
Ad N
e
Ad Exchang
etwor
k Ad
Filter
N
Ad
Ad networks and ad exchanges filtered based on geo, ad size, creative and more
2
ork
etw
Price
Technology Outlined: Ad Price Prediction
Inside the technology:
Determine Pricing
Prediction
Flat CPM campaigns sold through
Technology Benefits:
a publisher’s Algorithms determine• pricing Improve from monetization of every single publisher sales force can also compete for the ad impression ad impression ad networks and ad exchanges
• Solves ad network defaulting problem for publishers The algorithms consider frequency • Two new technologies improves algorithm accuracy as each time an impression is seen, Bidding• API Works with ad networks, ad exchanges, and the value drops non-guaranteed insertion orders
+
Machine Learning
Technical Level of White Paper:
3
Select Highest paying ad network or ad exchange is selected
$$
$
Should the selected ad network or ad exchange default, the steps are repeated and the next highest paying one is selected
Ad Network
$$$
Ad Delivered from Highest Paying Ad Network
Published by PubMatic, 2009
8
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
Detailed Flow Description:
The user on a publisher’s website makes a page request. The impression is then analyzed for dozens of different data points including context, frequenAd Price Prediction: 2nd Generation cy, geography, day part, browser, user demographics, and more.
Ad Revenue Optimization for Publishers
Ad serving entities, including ad networks, ad exchanges, and non-guaranteed
insertion orders are then filtered.
The filtering process takes into consideration the analyzed impression and user data as well as the creative policy of the publisher. For example, ad networks that serve suggestive or alcohol ads will not make it through the filter if the publisher’s business rules require that suggestive or alcohol ads not be shown on their website.
Algorithms determine pricing from Ad Networks, Ad Exchanges, and Non-Guaranteed Entities. rom the eligible ad serving entities that passed through the first filter, PubMatic’s algorithms F process over 100,000 of data Technology points perOutlined: second to decide which ad serving option is best able to monetize the impression. Data is collected Ad Price Prediction from PubMatic’s machine learning algorithms and decisions are made in real-time based on learned pricing behaviors and dynamic pricing data delivered from ad networks viaTechnology the real Benefits: time bidding API (application programming interface).
• Improve monetization of every single publisher ad impression • Solves ad network defaulting problem for publishers A key advantage for publishers in this process is PubMatic’s ability to determine ad pricing • Two new technologies improves algorithm accuracy based on how many times the• Works user with hasadseen a particular ad. Ad networks generally value the networks, ad exchanges, and non-guaranteed orders first impression a user sees more than theinsertion second impression, and so forth. Ad networks then
allocate the highest paying campaigns to the first user impression, followed by the next highest Technical Level of White Paper: paying ad campaign and so on. The algorithms take this frequency pricing into consideration and will choose not to show an ad if the user has seen it enough times that the value is too low. Because frequency capping is a part of most campaigns today, this technology is incredibly valuable and provides significant and long-term revenue lift.
Published by PubMatic, 2009
9
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
Frequency eCPM Curve $3.60
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers Area reflects publisher revenue lift
Predicted eCPM
$3.00
due to optimal decision-making
$2.40
Ad Network 1
$1.80
Ad Network 2 Ad Network 3
$1.20
Ad Network 4 Optimal Ad Network Allocation (based on frequency eCPM curves)
$0.60
>25
21-25
16-20
11-15
10
9
8
6
5
4
3
2
1
The Highest Paying Ad the Ad to the User.
7
Technology Outlined: Ad Price Prediction
$0.00
Technology Benefits: • Improve monetization of every single publisher ad impression • Solves ad network defaulting problem for publishers • Two new technologies improves algorithm accuracy Serving Entity is Selected • Works with ad networks, ad exchanges,and and Then non-guaranteed insertion orders
Delivers
Once PubMatic selects the entity that Level is best able Paper: to monetize the ad impression, an ad request Technical of White is sent to that entity. The ad is then served directly from that entity, whether it is from an ad network, ad exchange, or from an insertion order.
*If the selected ad entity were to default, the previous steps would be repeat and the next highest paying ad serving entity is selected.
Published by PubMatic, 2009
10
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
Dynamic Default Optimization Ad PriceImpression Prediction: 2nd Generation 2 Impression 3 Ad Revenue Optimization for Publishers
eCPM
Impression 1
A B C D E
Technology Outlined: Ad Price Prediction
D C A B E
C B E D A
Technology Benefits: • Improve monetization of every single publisher Ad Networks Ad Networks ad impression Ad Networks • Solves ad network defaulting problem for publishers • Two new technologies improves algorithm accuracy • Works with ad networks, ad exchanges, and insertion orders therefore adjusting static Ad network pricingnon-guaranteed changes constantly,
daisy chains weekly, or even daily, isn’t enough to maximize yield. Technical Level of White Paper: Dynamic Default Optimization updates the daisy chain in real time, for every impression, which ensures that the impression goes to the highest paying ad network.
Published by PubMatic, 2009
11
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
IV: Ad Price Prediction Publisher Case Studies Publishers using Real Time Optimization consistently see higher ad pricing as compared to manual ad operations solutions, whether in-house or outsourced. The following case studies based on three large publishers highlight the increased pricing (eCPM) generated from Ad Price Prediction: 2nd Generation PubMatic’s Real Time Optimization solution using automated algorithms..
Ad Revenue Optimization for Publishers
Real-Time Optimization vs. Manual Outsourced Ad Operations
88% Lift
eCPMs
Publisher Case Study 1: 88% Ad Revenue Lift
Site Type: News Site Reach: 10MM+ Global Users
$1.09
$0.58
Other Yield Optimizer
Technology Outlined:
81% Lift
$1.21
eCPMs
Publisher Case StudyAd2:Price Prediction 81% Ad Revenue Lift
PubMatic
$0.67 Real-Time Optimization vs. • Improve monetization of every single publisher Manual Outsourced Ad Operations ad impression Technology Benefits:
• Solves ad network defaulting problem for publishers
Site Type: Women’s Interest• Two new technologies improves algorithm accuracy Works with ad networks, ad exchanges, and Site Reach: 30MM+ Global •Users non-guaranteed insertion orders
Other Yield Optimizer
PubMatic
Publisher Case Study 3: 92% Ad Revenue Lift Real-Time Optimization vs. Manual Outsourced Ad Operations
Site Type: Social Network Site Reach: 15MM+ Global Users
Published by PubMatic, 2009
eCPMs
Technical Level of White Paper:
92% Lift
$0.50
$0.26 Other Yield Optimizer
PubMatic
12
Ad Price Prediction: 2nd Generation Ad Revenue Optimization for Publishers
Technology White Paper
V: Conclusion As ad inventory continues to grow, new methods of ad revenue optimization must be adopted by large publishers if they are to protect and improve the value of their advertising space. Ad Price Prediction: 2nd Manual ad operations, either in-house or with outsourced help, doesGeneration improve revenue ad revenue lift in most cases, but those options are not feasible for long-term growth. Because Ad Revenue Optimization revenue for Publishers the non-guaranteed segment of online advertising is expected to grow to $11 billion by 2013, publishers must adopt a specific strategy for this segment of inventory. Due to the volatility of online ad pricing, and because no single ad network can guarantee the highest price for a publisher’s ad space all of the time, only a real-time optimization solution can ensure that every impression is monetized by the highest paying source. PubMatic offers publishers the most advanced method of ad revenue optimization available: Ad Price Prediction technology (automated algoritms) in addition to full service team support for the publisher’s ad operations team. Publishers that have been using PubMatic’s solution regularly see ad revenue lift ranging from 30-300%.
About PubMatic
Technology Outlined: Ad Price Prediction
PubMatic is a global Ad Revenue Optimization Technology Benefits: company that provides online publishers with a full service solution to manage andmonetization monetizeofnon-guaranteed • Improve every single publisherad inventory. PubMatic’s realad impression time ad price prediction technology ensures that online publishers get the most money from • Solves ad network defaulting problem for publishers their advertising space by deciding intechnologies real-time improves which ad network or exchange can best mon• Two new algorithm accuracy etize each impression. • Works with ad networks, ad exchanges, and non-guaranteed insertion orders
There are currently over 6,000 large and medium publishers working with PubMatic. PubMatic Technical Level of White Paper: is venture backed by Draper Fisher Jurvetson, Nexus India Capital, and Helion Ventures.
Contact PubMatic If you are a large online publisher interested in learning more, please contact Josh Wetzel, VP Publisher Solutions.
[email protected] If you are an ad network interested in learning more, please contact Jeanne Houweling, VP Business Development and Advertising Sales.
[email protected]
Published by PubMatic, 2009
13