Case Study: Anodot Finds ‘All the Anomalies Fit to Print’ for Media Giant PMC

About the Company

Penske Media Corporation (PMC) is a constellation of global media brands fueled by remarkable content across its digital, video, print, & event properties. The company boasts 179 Million Monthly Active Users according to comScore August 2016 and 101 Million Video Views per month (as of August 2016). Among its 22 brands are Variety, Deadline, Hollywood Life and WWD.

Tired of Setting Alerts

With millions of users across dozens of household-name and professional publications, PMC’s data science team needed a better way to track business incidents than Google Analytics’ alerting function.

“You have to know what you’re looking for in order to set alerts in Google Analytics, which is really time consuming and you tend to miss things,” said Andrew Maguire, Head Data Scientist at PMC. “It’s sort of old school.”

The company was experiencing significant delays in discovering important incidents in their active, online business.

Brought in Anodot to Solve One Problem, Gained Solution for Dozens of Issues

The initial use case for PMC was to start using Anodot to track its Google Analytics activity. For example, to identify anomalous behavior in impressions or click through rates for advertising units. Anodot learns the normal behavior of each metric, with its seasonality or other patterns, and alerts automatically when a metric behaves differently than expected. So an alert about a change in impressions or click through rate could indicate that a Demand Side Platform (DSP) connection is down.

While many important issues were uncovered lurking within the Google Analytics data, an important one was the fact that a new trend had appeared whereby a portion of traffic to one of PMC’s media properties was coming from a bad actor. This is known as “referral spam,” and it artificially inflates visitor statistics. Spotting the issue would have required PMC’s analytics team to know what they were looking for in advance. By discovering this through Anodot, and then blocking the spam traffic, PMC was able to free up resources for legitimate visitors, as well as accurately track the traffic that matters most, enabling executives to make better informed decisions and plans.

PMC was also able to use Anodot to identify software bugs early, before small issues turned into major crises. For example, in one case, Anodot flagged higher than expected bounce rate in a publication’s gallery pages, and correlated this alert with an increase in the corresponding image sizes after a back end code push. The PMC team was able to determine that some new gallery functionality had inadvertently resulted in bigger then normal image sizes on mobile which caused a spike in page latency, harming user experience with visible effects on traffic trends. This also had an effect on revenue due to the impact of user trends on monetization opportunities. PMC quickly rolled out a fix to the affected pages solving the issue at its root. Without Anodot, it may have been days before the issue was noticed, which would have resulted in further frustration for users and declines in revenue.

“The results were so immediate when we started using Anodot to better understand our Google Analytics that we decided to try additional data streams, to solve other issues within the organization,” Andrew explained.

For example, for search engine optimization (SEO), PMC uses Anodot to analyze its Google search console and Google webmaster data. As a result, sudden changes in search engine ranking for the company’s relevant keywords will automatically trigger an alert on Anodot. This could indicate a change in Google’s ranking algorithm, or poor availability of some of PMC’s pages, either of which would require immediate action from PMC.

“When Anodot notifies us of an anomaly in our streaming data, the first thing I try to figure out is if it is due to something we did ourselves (for example a product upgrade), or if it’s due to the world around us (for example an issue with one of our advertising partners). The information I get from Anodot helps me identify the root cause quickly, since it correlates related alerts without our having to configure anything,” Andrew said.

Not all Incidents Are Bad, Some Are Positive Indicators

In order to better understand user behavior, PMC closely tracks how much readers are engaging with the site, using metrics such as scroll depth (how far they scroll on a page), whether they share the content, and whether they engage with additional content such as video. By analyzing these metrics within Anodot, PMC is able to identify positive business insights, that is, which types of content are gaining the most traffic and engagement from its core audience, and use it to make editorial decisions.

“If a story in one of our publications goes wild, we see spikes in our data which Anodot catches. Our editorial team can use the information about successful content themes derived from Anodot’s analytics to make educated plans for new types of content to create that are designed to generate interest,” Andrew said.

Looking Ahead: More Nuanced Custom Tracking

PMC planned on using Anodot and Google Analytics data as an initial internal proof of concept for an intelligent alerting system driven by machine learning with minimal direction required from business users in terms of complex rules or triggers.

Now that we have proven what we set out to prove,” Andrew explained, “our plan is to incorporate Anodot into more core sources of data across the business and to implement even more nuanced custom tracking on top of Google Analytics so that we can even more flexibly track key metrics that matter in a highly customized way.”

“If a story in one of our publications goes wild, we see spikes in our data which Anodot catches. Our editorial team can use the information about successful content themes derived from Anodot’s analytics”

– Andrew Maguire, PMC Head Data Scientist