In this article, we will focus our attention on an innovative content decision engine, which selects articles most likely to succeed as premium content. The summary here is a high-level review of cutting-edge analytics being adopted by leading publishers.

The results below are from a three-month pilot implemented at one of the largest U.S. media brands. The programme validates our hypothesis that publishers can achieve significant boosts in revenue and subscriptions with minimal lift.

Impact from premium content test

On average, articles selected by the model and set as premium generated nearly three times more conversions than comparable articles not set as premium. The impact from the model boosted the total expected 2021 starts volume by 20% and topline digital revenue by 8% (both figures forecasted for a 12-month period).

“Content is king”

Everyone agrees the adage is true. However, in practice, publishers undervalue their own content when executing premium content strategies.

North American publishers have experimented setting certain articles behind the paywall irrespective of the user or paywall rule. Most publishers in the United States target 10%-15% of content for premium. The effective percentage of articles set to premium is much lower for most publishers as newsrooms tend toward caution and managing pageview risk.

Many newsrooms are also strapped for resources to execute effectively. Systematic workflows and decision criteria are elusive. Decisions are often based on a gut feeling without thorough data review before or after the decision. Sometimes going viral or reaching the widest possible audience (measured by the ubiquitous “pageview”) is just too alluring to keep an article behind the paywall.

By design, this approach by most publishers is bound to leave money on the table. Flagging premium articles is an “opt-in” decision. Any marketer will attest that “opt-in” is guaranteed to perform worse than an “opt-out” approach.

Newsrooms and data

The creation of quality journalism is inherently a human activity. The newsroom is the core of the subscription value proposition. The role of data and the job of any embedded data analysts is that of “informer,” nudging journalists but nevertheless deferring decisions to editors. Being data-informed is indeed powerful and has led to success for many publishers.

Most newsrooms have accepted and even welcomed data into their daily operations. Knowing which articles generate the most subscriptions, pageviews, repeat visits, or subscriber views gives journalists tangible evidence of their efforts.

However, balancing quality journalism and data-driven decisions comes down to which decisions are informed versus automated. Some are better for humans while others can be unburdened onto algorithms.

Apply data science to repeated automated decisions to get the greatest ROI

Once the journalism is written, the decision of how to monetise does not need to remain solely within the newsroom. The “separation of church and state” has existed for decades for good reason. Marketing tactics, pricing, paywall rules, and setting premium content are tactical decisions ripe for predictive modeling and automation.

The diagram above shows the workflow of the premium content engine. Listener was integrated with the ARC content management system to capture articles before publication. The predictive model uses taxonomy, metadata, and natural language processing to predict how likely an article would succeed when set as subscriber-only. The model is initially trained on several months of history but continues to improve with new data and performance tracked every day. A recommendation is sent via Slack alert within minutes of the article being “ready to publish.”

The chart below shows last-click conversion performance of the articles tested over a three-month period. It notes the following types of conversions:

• Non-Premium: recommended non-premium articles set as non-premium.
• Premium Not Set: recommended premium articles set as non-premium.
• Premium: recommended premium articles set as premium.

Recommended non-premium set as premium were rare and excluded from the chart.

The key insight here is the middle bar, which indicates part of the lost opportunity. When implemented fully, setting the recommended articles behind the paywall in this market generated nearly six conversions per article. Though the recommended articles not set as premium still performed strongly (two conversions per article), similar articles set to premium generated nearly three times the conversion volume.

The net result to the bottom line was a measured 20% boost in subscription start volume and an 8% boost in net digital revenue (accounting for marginally fewer pageviews from anonymous users).

One limitation in the implementation of this workflow is the reliance on a Slack alert and follow-through by the receiver to implement the recommendation. Even with an automated alert, many recommended articles were still not set to premium, leaving revenue unrealised.

Looking ahead = back to the future

Mather is continuing to work with the publisher referenced in this blog (and others) to evolve and tighten the newsroom workflow. A next step is to directly set the flag within the CMS or paywall system to ensure full follow-through from the recommendations. Since the launch of the programme, Mather has also developed a post-publication model to augment the pre-publication decision engine, ensuring real-time article performance is accounted.

Inevitably, the model of the future will combine advanced audience and content analytics in tandem to optimise subscription value. Technology is catching up to the thought leadership and analytics.

Mather has been fortunate to support news media brands through significant digital transformation over the last decade. Years ago, Mather introduced the Intelligent Paywall, but paywall technology at the time lagged capabilities to personalise by user certain parameters, such as metre settings, offers, and creative. Over time, multiple paywall companies emerged to enable such functionality.

Like the rapid evolution of paywall technology, Mather anticipates a similar modernisation of content management systems, not just in the production of content (see headless CMS) but also in supporting analytics and enabling subscription optimisation. A “best of breed” solution will likely define the tech stack of the future.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .