Did you know …
Nearly 60% of adults subscribe to some sort of email newsletter, and when polled, their trust in newsletter content ranks significantly higher than the trust they have in traditional media outlets.
An aspect of this trust is built by delivering focused value to the reader through the careful curation of a newsletter content…only the most highly relevant and engaging content makes the cut. If you’ve been keeping up with the research being conducted on content and readership over you’ll have been made aware of this insight: finding quality content isn’t easy – with 2.4 billion newsletters emailed to worldwide readers every week, countless hours are spent scouring the internet for topics relevant to specific audiences.
At Turbine Labs, we are well aware of this insight, and the overwhelming amount of time associated with finding the right media to bolster your brand.
As we put together our customer newsletter content (Turbine sends company-specific newsletters to about 13,000 subscribers every week) we encounter a similar issue as our data analysts navigate through the firehose of Twitter. Among the millions of posts, authors and arguments, how does one know what actually matters to newsletter readers? From that pain point was born Turbine’s Twitter Impact and Engagement Rating., a proprietary algorithm that quickly identifies meaningful content. And from the pain point of scouring through the billions of potential content to include in a newsletter or report, we have uncovered another solution.
When crafting newsletters, news briefs and other emailed content for our customers, our data analysts consistently ran into two major obstacles.
The first obstacle we face when surfacing the best content for our newsletters is the number of syndications – especially press releases – that weigh-down our Artificial Intelligence data pulls focused on extracting the meaningful articles from the noise.
The second obstacle was finding relevant content that would otherwise be obscured and hidden by traditional media monitoring tools. Moz, Similar Web, and BuzzSumo, for example, offer valuable metrics like how many times an article was shared to social media or back-linked from other news sources, but our analysts found that while those metrics offered some direction in their prioritization and surfacing, these metrics didn’t necessarily equate what is important to our clients’ business.
For example, take this scenario into account: Say that Coca-Cola hired Turbine Labs to find information that supports their company’s initiative to reduce its carbon footprint. The company would probably be interested in which steps their competitors are taking to do the same, or maybe which trends are emerging in the larger beverage industry to protect the environment.
To make sure we are getting all of the information relevant to this initiative, previously a Turbine Labs analyst would expand their search to more general themes like “plastic pollution,” “carbon neutrality” or “climate pledges.” But…search for those keyword phrases and terms in a traditional media monitoring tool, and you’ll get 3,000-plus articles in the last week alone! If you were to then filter that information according to social shares, one of the top hits would be a story on how a wine bar in New York City is planting a forest in Peru.
Interesting, but not exactly relevant.
And it works both ways: stories that weren’t highly shared because they are too niche, but very valuable to Coca-Cola, would be missed entirely.
And then there is the mountain of duplicate content: Articles from “the wire” news sources like Associated Press or Reuters have been reposted to hundreds of local news sites. Some sites also post duplicate press releases in full or in part into their market reports and financial sites in an effort to get higher impressions, increased readership and therefore a greater number of pageviews. While pageviews are a key KPI for online news sites, this KPI doesn’t always equate to value for the reader.
But if even one syndicated story is unimportant, it swells the amount of irrelevant content you’re sifting through by hundreds of articles. And if one article is important, but another isn’t? Nearly impossible.
How to provide significant value within the newsletter ecosystem? Introducing… TurbineRecommender!
To combat the corrupted and noisy data we have built TurbineRecommender – a media monitoring tool that recommends and surfaces only the impactful and important media content while discarding all of the noise and bad content.
TurbineRecommender uses a custom-built, proprietary machine learning model that utilizes human labeled data to teach the AI context and business rules surrounding your company’s interests and needs. Through a simple workflow, analysts can mark topics, brands, and events of importance, and conversely mark content that is irrelevant or noisy. TurbineRecommender then uses this training data to recommend and show relevant and impactful content, while keeping junk out. AI, FTW (for the win)!
So back to Coca-Cola, which is decidedly not interested in a New York City wine bar. Using TurbineRecommender , our analysts can give that kind of content a “thumbs down,” and over time, TurbineRecommender incorporates those nuances into its own self-teaching labeling system. Irrelevant articles, like the forest in Peru, are filtered out.
You Label. TurbineRecommender Learns.
Take a look at the graphic below. There are two workflows that train TurbineRecommender : an on-the-fly labeling system to quickly mark an article with a thumbs up symbol as relevant and a thumbs down as not relevant, and a more in-depth labeling workflow that allows the user to initiate a labeling session in a separate module. Both support the same training concept and provide TurbineRecommender the same context for learning.
TurbineRecommender Finds What Matters, Removes Noise.
After the user starts to label and mark the topics of importance, TurbineRecommender’s magic begins. Depending on the amount of human labeled content, TurbineRecommender recommends topics of interest or importance. Below, you can see TurbineRecommender’s labels in action.
Users can use a quick and easy filtering system, depicted below, to zoom in on the content TurbineRecommender has recommended and surfaced. Users can also view the content TurbineRecommender marked as irrelevant and unimportant – allowing full control to understand how accurately TurbineRecommender is recommending content, and the labels applied.
Users Can See When TurbineRecommender Needs Help.
TurbineRecommender is smart enough to nudge the user when it needs more training data, and can recognize when it is below a certain accuracy level. In the video below, TurbineRecommender alerts the user that it needs more training data, and can automatically initiate a labeling session for the user.
How it works Why does it matter
Here at Turbine Labs, TurbineRecommender is now a household term. It is used daily by our data analyst team and fuels nearly all of our enterprise work. We have experienced first-hand the solutions TurbineRecommender can provide by cutting out repetitive content, restoring time for meaningful work, and relying on TurbineRecommender to do the heavy-lifting.
Cut Out Repetitive Content.
TurbineRecommender automatically removes duplicate and repetitive content to show unique and validated articles. With TurbineRecommender, Coca-Cola would not have to sift through 300 reposted Associated Press articles. Instead, the company could use that time to read one unique Associated Press article — the one that had the greatest impact and use the time to take action on that instead of still searching.
TurbineRecommender has removed nearly 36% of duplicate coverage across Turbine Labs’ ingested content. That is the equivalent of more than 100,000 articles. Imagine having to sift through that content and how many wasted hours would be in that effort!
Restore time for meaningful work.
Because TurbineRecommender removes hundreds of thousands of articles from our workflow, Turbine Labs’ enterprise-focused work has gotten smarter and more profitable.
“TurbineRecommender cuts nearly 68 hours of work time monthly. Time that would have been spent sifting through noise.”
TurbineRecommender cuts out 25% of machine and human production time on our Digest products, slashing nearly 68 hours of work time monthly. Instead of plodding through meaningless content, our team has time to identify impactful news and provide it to our customers more quickly, and we’re confident these results can happen for any organization we work with.
Let TurbineRecommender do the Heavy-Lifting.
In the last 60 days, TurbineRecommender has filtered out more than 84% of our ingested content based on human labeled data – 142,000 articles our analysts will never have to read because of their lack of relevance or accuracy
Instead, TurbineRecommender flags 16 articles out of every 100 as ‘worth a look’. And as TurbineRecommender gets smarter for our customers and our analyst teams, it will continue to fine-tune those recommendations – leaving your businesses and our own team with 100% more time to make important decisions and discover actionable insights.
TurbineRecommender does the work to find content that matters – and is easily now your new best media monitoring tool.
Want to learn more?
If you’re ready to discover a better, more efficient way of media monitoring, schedule a consultation with one of our specialists and find out how Turbine Labs can help.