How PubMatic Is Reducing Video Creative Errors

Post on June 26, 2018 by Pankaj Jha

Pankaj Jha Senior Principal Software Engineer

Video ads can fail to serve for a variety of reasons ranging from creative to player issues. The subsequent video creative errors result in a poor experience for publishers, buyers and the user.

The PubMatic team is actively working to better understand the cause of critical video errors, discover error trends and determine why not all publishers are impacted. While our original processes filtered out error-prone creatives, a reactive strategy, we also wanted to pre-emptively stop errors before they happen.

Our engineering team recently updated our reactive strategy, reducing the manual effort required, and developed a preventative strategy as well. Take a look at what we did and how it has improved our efficiency and client experience.

Updating Our Reactive Strategy

Our original solution for blocking creatives with a high-error count was operational, however, the process was manual, time-consuming and incomplete. To improve, we recently automated the process with positive results including:

  • Improved identification of creatives with a high-error rate: We implemented an updated workflow of scheduling, downloading (FTP/HTTP) and processing big data reports which capture publisher video errors. Data is aggregated and updated rules are applied to review and block creatives with high-error rates for affected publishers.
  • Updated custom rule engine: We added support for blocking rules with control at the publisher and DSP levels. This allows for skipping or blocking creatives, as necessary, or allowlisting a particular creative, publisher, or DSP. Additionally, blocking criteria was added to a database so they can be updated, at will.
  • Daily reporting: Our operations team now uses the available reports to reach out to DSPs and publishers.

Updating Our Preventative Strategy

After analyzing data and video-error logs, we discovered two ways to pre-emptively reduce the percentage of video creatives erroring out on our platform. We filter creatives based on:

  1. Error codes: The IAB has standardized definitions for video creative failure, as error codes, to improve ad tech’s understanding and subsequent troubleshooting methods. Among the standard set of VAST error codes, there is a collection referred to as “fatal errors.”

Fatal errors stop the ad request and consequently result in lost revenue. To address this, we implemented checks for fatal error codes in our scanning workflow for each video creative. Creative is now blocked or filtered before the auction. This process allows for another video creative, without a fatal error, to be served successfully—thus demand is not impacted. To address the problem at the source, we also reach out to the DSPs providing creatives with error codes, signifying creative issues. In the future, this type of contact will also be automated.

We specifically look for the following fatal error codes:

# Fatal Error Code Definition
1 100
  • Invalid XML Document
2 301
  • Demand partner HTTP (status: 404) file not found error.
  • Demand Partner (VAST URI) HTTP request failure.
3 303
  • No ads VAST response after one or more wrappers. This also includes the number of empty VAST responses from fallback.
  • No ads VAST response after one or more Wrappers from demand partner ad server.
4 397
  • Demand partner redirect URI (<VastAdTagURI> element) not found in VAST XML.
5 401
  • Media file (<MediaFile> element) is missing in Linear Ad.
  1. Attribute matching: Each video creative is scanned for attributes like mime types, bit rate, API (VPAID) version, etc. This information is then used to match each ad request or impression data with video creative attributes. Creatives are then shortlisted for participating in an auction. Un-matched creatives are excluded from an auction, thus improving the probability of getting successfully served and played.

Key Learnings and Results

Updating our processes was beneficial but not without challenges. We quickly realized the enormity of handling multiple big data reports, in addition to performing the necessary data aggregations and creating rules for the data set.

As a result of our updated solutions and processes, we (and our clients) have experienced multiple benefits.

  1. Decrease in platform error rate: After the updates, our daily platform error rate saw decreases of up to 50 percent. With further analysis and implementing new rules, we expect the error rate to continue to decrease.
  2. Reduction in turnaround time: The original reactive process now takes 3-4 minutes rather than two hours.
  3. Quicker responses: DSP and publisher contacts have become much easier, and more efficient, due to the newly available reports.

What’s Next?

PubMatic is constantly working to improve our processes so that we can more efficiently detect and address video creative errors. To learn more about our video innovations, check out our recent content or read more about our solutions.