Solving The Remnant Inventory Problem: AdMonsters 2009 Presentation
14 likes2,386 views
The document discusses the remnant inventory problem faced by premium publishers and outlines how discretionary optimization can enhance revenues, lower costs, and protect brand reputation. It emphasizes the importance of managing discrepancies, frequency, and fill rates in ad optimization while introducing real-time bidding (RTB) as a solution for maximizing revenue. The findings illustrate significant revenue lifts through dynamic strategies and the importance of data in the ad industry.
Solving The Remnant Inventory Problem: AdMonsters 2009 Presentation
- 1. Solving the Remnant Inventory Problem Ben Barokas, Co-Founder and CRO August 18th 2009
- 2. About Us Founded in October 2007 Select AdMeld Customers Focus on premium publishers 80+ customers Manages more than 300 million ad impressions daily Raised $15M in venture funding from Spark Capital and Foundry Group 1 © 2009, AdMeld Inc. All Rights Reserved.
- 3. Introduction How does discretionary optimization work? – Create your ideal network portfolio – Calculate the true value of every impression – Deliver it with scalability and quality of experience What does it do for you? – Boost your revenues – Save you time, lower your costs – Help protect your brand Looking forward – RTB and Data 2 © 2009, AdMeld Inc. All Rights Reserved.
- 4. Creating Your Ideal Network Portfolio Analyze your site Optimize your Understand portfolio network inventory Integrate and Find the right mix prioritize 3
- 5. Diversification is Key High Fill Low High CPM CPM Low Fill 4 © 2009, AdMeld Inc. All Rights Reserved.
- 6. Optimizing A Single Impression Network A Rev Share $1.50 Network B Rev Share $1.20 Network C Real Time Bid $1.10 Network D Fixed 3x24 $1.00 Network E Rev Share $0.50 © 2009, AdMeld Inc. All Rights Reserved. 5
- 7. Getting to True Value Discrepancy Frequency Fill 6 © 2009, AdMeld Inc. All Rights Reserved.
- 8. Discrepancy Many sources: internet latency, ad server latency, user moving away from page to quickly Without discrepancy management, optimization is ineffective Achieved 20% revenue lift at IAC through discrepancy management alone 7 © 2009, AdMeld Inc. All Rights Reserved.
- 9. Factoring in Discrepancy Start Discrepancy eCPM $1.50 40% $0.90 Network A Rev Share Network B $1.20 10% $1.08 Rev Share Network C $1.10 0% $1.10 Real Time Bid Network D $1.00 15% $0.85 Fixed 3x24 Network E $0.50 10% $0.45 Rev Share 8 © 2009, AdMeld Inc. All Rights Reserved.
- 10. Frequency Early views worth most CPM is an average across multiple views Many networks shift to CPC or CPA at higher frequencies Previously was done with multiple tags from networks which carries a lot of overhead for premium publishers 9 © 2009, AdMeld Inc. All Rights Reserved.
- 11. Factoring in Frequency Discrepancy Frequency eCPM Network A $0.90 60% $0.54 Rev Share Network B $1.08 120% $1.30 Rev Share Network C $1.10 100% $1.10 Real Time Bid Network D $0.85 100% $0.85 Fixed 3x24 Network E $0.45 100% $0.45 Rev Share 10 © 2009, AdMeld Inc. All Rights Reserved.
- 12. Fill Rates and Pass backs Highest paying tags usually have low fill Managing fill is essential to calculating revenue Daisy chains ensure an ad is shown What used to be done manually once a week, now done dynamically for every impression 11 © 2009, AdMeld Inc. All Rights Reserved.
- 13. Calculating Dynamic Daisy Chains 12 © 2009, AdMeld Inc. All Rights Reserved.
- 14. Factoring in Fill 13 © 2009, AdMeld Inc. All Rights Reserved.
- 15. True Value Privileged & Confidential 14 © 2009, AdMeld Inc. All Rights Reserved.
- 16. The Results Network B Network D Optimized Choice Rev Share Fixed 3x24 $1.16 $1.20 $1.00 Network A Network E “Common” Choice Rev Share Rev Share $0.47 $1.50 $0.50 150% Revenue Lift Over 100,000,000 impressions, an additional $70,000 15 © 2009, AdMeld Inc. All Rights Reserved.
- 17. Reality Check Doing this for large, premium publishers means: – Calculating 5000 chain combinations per impression, in real time, millions of times a day – Accounting for geo, frequency caps and network latency – Maximizing revenue during traffic spikes – Backing it up with consultative services and expertise – Executing against publisher business rules 16 © 2009, AdMeld Inc. All Rights Reserved.
- 18. Managing Business Rules Complete visibility into each ad, without leaving your website. • See the network that served the ad • Report or disable problem ads • View pricing, fill, targeting info, etc. 17 © 2009, AdMeld Inc. All Rights Reserved.
- 19. Real Time Bidding A Shorter Road to True Value With RTB, buyers bid dynamically for each impression instead of setting blind rates (futures) beforehand Less Risk, Less Friction With less risk, buyers confidently spend more at higher rates, and pubs will have more access to demand sources RTB To Ramp Up in 2010 As adoption grows, so will efficiency and performance A Big Win for Premium Publishers The most valuable inventory lies at the nexus of content, context and audience. Premium publishers have all three. Privileged & Confidential 18 © 2009, AdMeld Inc. All Rights Reserved.
- 20. It’s All About Data Privileged & Confidential 19 © 2009, AdMeld Inc. All Rights Reserved.
- 21. Thank You July 16th 2009
Editor's Notes
- #7: Let’s use a simple example:Network A – Acerno, AudienceScience, typical behavioral network. High CPM but low fill. Network B – Typical network player like ContextWeb, Dotomi who has high CPMs and decent fill.Network C – RTB, done through APIs, not tags.Network D – Fixed deal on fixed frequency. Typical of interclick, specific media, intercept, etc.Network E – “Catch-all” no geotargeting restrictions, will take as much traffic as you give it, typical of AdSense.
- #8: 1) Discrepancy= difference between what a network should be paying for and what they’re reporting they should be paying for.2) Frequency= the number of times a user has been shown an ad from a given network.3) Fill=the % of impressions we send to the network that they accept.
- #9: Latency primarily comes from latency of latency of the web in general, latency of ad networks, and ad servers. Often, legacy servers weren’t built to handle the high volumes and complexities (such as multiple passbacks), and have a hard time keeping up.
- #10: 1) No discrepancy for RTB tag because RTB networks pay off AdMeld’s numbers.
- #11: 1) Each network has its own optimization engine that chooses to serve the highest revenue ad first. On the second impression, a lower payout ad. As you move down the chain towards CPC and CPA deals, the payouts get even lower.
- #12: Tag A – In this example, our system applies a 60% frequency factor because this network has seen this user a couple of times already.Tag B – This network hasn’t seen this user yet, but they’re saying $1.30 is the average across a whole user session. Therefore, our system assigns a 120% frequency factor to valuate the first impression.Tag C – RTB, so this is what they’re actually paying, no frequency factor is applied on our end.Tag D – Fixed deal with 3x24 freq cap. They calculated the curve on their end so no need for us to apply a frequency factor to it.Tag E – This dbn
- #14: 1) All the viable possible chains for this example
- #15: 1) Pairing them reveals fill rates for each chain
- #16: 1) Combining those fill rates with our valuations after calculating the frequency factors reveals the true expected value of each chain. AdMeld picks the one with the highest expected value.
- #17: The “winning” chain has an expected eCPM of $1.16, which is ~150% higher than the “common” choice, which is to pair the highest priced tag with the highest fill tag.