The recommendation engine is one of the biggest martech innovations of the last few years, and it’s shaping our entire digital experience. For example, let’s say I visit the Harvard Business Review website and read about manufacturing marketing tactics (as one does). Soon, I’m getting retargeted marketing emails recommending the latest manufacturing supply chain management articles and other related content, as well as customized recommendations the next time I visit the site. When I fire up Netflix, I know that my experience is going to be managed in part because Netflix knows I’ve been binging on the show Haven and it’s going to suggest other supernatural crime fighting series.
The development of recommendation engine technology has brought digital personalization to a whole new level. For brands developing content marketing and martech strategies, it’s critical to pay attention to this emerging technology and understand where trends are headed when shaping the user experience.
Image attribution: Moon M.
Understanding The Recommendation Engine
“Recommender systems help users discover new movies to watch, books to read, items to purchase, music to listen to, and news to read. They in effect do the job of ‘word-of-mouth’ recommendations that we consult friends for, but at a much larger scale,” says Mustafa Bilgic, associate professor of computer science and director of the Machine Learning Laboratory at Illinois Institute of Technology. Bilgic’s work on interactive and transparent recommendation systems is fascinating; for a quick primer, watch his TEDx talk here.
The algorithms that power recommendation engines can be based on a variety of factors, from simple methodology where content is associated through tagging, to more complex techniques such as looking at the behavioral data of people who consume content and mapping complex relationships. Some platforms also pull in multiple data sources and look at outside factors such as the behavior of your social networks.
Whatever back-end architecture is used to construct a recommendation engine, these technologies provide several advantages, especially from a martech and user experience perspective:
- Users have a personalized digital experience that better targets their needs and interest.
- Over time, as new content becomes available or users interact within a technology ecosystem, the personalization continues to refine and evolve.
- Marketers are able to deliver deeper personalization at scale.
- Content that’s created can be surfaced to interested audiences by topic or by buying stage.
- Some recommendation engines can help surface gaps and opportunities in the content landscape.
As marketers try to imagine and create digital experiences that resonate with audiences, it’s important to keep in mind how recommendation engine technologies shape those experiences. Brands like Netflix, Amazon, Spotify, and Pandora are helping to create a digital world where programs and platforms understand user needs and help guide them toward content they’ll love. Does your brand’s content experience offer that same level of connected personalization?
Image attribution: Alexander Popov
The Potential Pitfalls of Recommendation Technology
Navigating the emerging field of recommendation engines also involves being aware of the inherent pitfalls, such as bias. “A recommender system is your ‘virtual’ friend that has an opinion about that particular movie that your ‘physical’ friend does not know about. It is tremendously useful. But it can also be pretty biased and we might not know it, unless it is a white box that can explain its reasoning to us,” says Bilgic.
In other words, while it’s useful to have things recommended to us that are similar to what we’ve consumed and enjoyed in the past, we run the risk of operating in a small and focused digital bubble. Bilgic notes that while recommendation engines are useful, it’s important to be aware that, “like many tools in life, they can also have unexpected consequences. For example, unintended algorithmic biases can bias what we watch, hear, and read and we might not be even aware of it. We might be presented similar items to what we already like and not presented the choice to explore new tastes. Certain types of news might get popular or unpopular in our social media circle and we might miss on the variety of opinions. And so on.”
For marketers, there’s the risk of continuing to recommend a static set of content that’s focused on the same issues—and failing to advance that content as customers become more sophisticated in their knowledge or move further along the buying cycle. Marketers who are aware of this potential can foresee a handful of potential challenges:
- Recommendation engines have to move beyond making all suggestions simply based on “similarity.”
- Integrating a wider range of sources, such as social media, can help provide more dynamic and useful recommendations.
- Transparency in recommendations—that is, explaining why you’re making the recommendation and how you arrived at the conclusion that someone should see specific content—keeps your recommendations more useful and educates users on how the digital ecosystem around them has developed. Consider logging into your Netflix account and seeing specific recommendations under the heading of “Because you watched X”: It’s immediately clear how and why the system decided you would enjoy certain movies.
Incorporating Recommendation Engines into Content Marketing
The impact of the recommendation engine on martech is wide reaching; it’s not just about how we consume movies or music. It’s also playing a role in the way that buyers consume information during purchase decisions, and also how brands conceptualize their end-to-end content experiences. For example, Skyword recently developed its own personalized recommendation engine.
“Skyword Personalized Recommendations (SPR) is an add-on to our core content marketing platform,” says John Mihalik, Skyword’s chief technology officer. “SPR is designed to give users a personalized experience by connecting them with content created through Skyword’s content marketing platform.” SPR applies a set of machine learning technologies to Skyword’s database of online browsing profiles, collected from years of tracking user interactions with content produced through Skyword, to create personalized content recommendations for online browsing experiences as well as email channel communications.
In many ways, SPR is echoing the trend of brands looking for better ways to apply their data and technology to a more engaging reading experience. Research is happening on multiple fronts, including the best ways to use interactivity, transparency, and bigger data sets to help inform content marketing recommendations.
Beyond that, suggests Mihalik, next-generation martech is likely to make it easier to guide users through the buying journey. “I think in the next year or two one of the trends in recommendation engines will be toward story-form-aware recommendations and buying stage progression. Providing stand-alone ‘similar content’ recommendations is valuable, but more and more the need is to progress users through your brand’s story arc or buying stages. That’s the next generation of recommendation engines. The mathematics of one-off recommendations of ‘users who saw this also saw this’ will give way to the broader contextual story.”
Image attribution: Luke Porter
Using Recommendation Engine Technology for Your Own Brand
If you’re ready to take a deeper look at using recommendation engine technology in your own brand experience, here are seven things to keep in mind:
Move beyond the consumption myths
While it’s easy to focus on the way that recommendation engines shape content consumption, Mihalik points out that there’s an important application for content creation as well. In part, this shaped the direction of SPR. “The biggest key impact is in recommendations for content creation, not just content consumption. So many vendors and recommendation engines can only influence content consumption. That’s actually really bad and only fills half the true need. You have to also fill in the gaps of new content that is needed. Without sustained new content creation, recommendations with even the best AI will fail as the scope of what they recommend is static and inadequate for engagement. Knowing what those content gaps are is the other half of the puzzle,” says Mihalik.
Ensure your recommendation engine is intelligent
Today’s engines must go beyond basic associative mapping. It’s important that the recommendations are tied to stages in the buyer journey, for example. Not only will this help suggest the right content, but it will help suggest that content at the right time and help move prospects toward conversion. “Recommendation engines must be episodically aware of your brand’s story and understand the user’s personal progression path through that story. Combining these gives a much higher engagement potential. It’s like watching a movie, but every next act you see is a recommendation. It keeps you hooked until the end,” Mihalik notes.
Rethink your customer-focused data strategy
As the Harvard Business Review eloquently states, developing a recommendation engine that’s focused on the customer experience is the next best use of your company’s data: “Perhaps the single most important algorithmic distinction between ‘born digital’ enterprises and legacy companies is not their people, data sets, or computational resources, but a clear real-time commitment to delivering accurate, actionable customer recommendations. Recommendation engines (or recommenders) force organizations to fundamentally rethink how to get greater value from their data while creating greater value for their customers. In other words, they’re a terrific medium and mechanism for transitioning traditional managements to platform perspectives.”
Don’t just promote content—improve the user experience
One of the benefits of recommendation engine technology is promoting content. The way that you develop your recommendation engines has to take the user experience and—most importantly—user goals into account. “When a company recommends a product, the company might face a choice between promoting a product (promotion) versus helping the consumer choose the right product and keep the consumer happy (satisfaction),” notes Bilgic.
Be transparent about how you’re arriving at your recommendations
Increase the value of your recommendations by showcasing how you’re arriving at your recommendations. “Earlier recommender systems were quite simple systems whose recommendations were easy to explain to the user. Recent techniques are much more accurate in their recommendations but they are also much more complex and opaque and thus it is hard to explain their recommendations to the users. Explanations are often useful for the users in that they can make more informed choices. For example, imagine knowing why a movie is recommended to you, whether because of its plot, a certain actor, or another factor. In that case, you’ll know better whether you are in the mood to watch that movie right now, later, or never. Explanations are even more crucial when something goes wrong so that a fix can be applied,” says Bilgic.
Strike the balance between exploitation and exploration
When properly leveraged, recommendation engines can also help test whether users benefit from specific content that might not be immediately obvious. But it’s a delicate balance, Bilgic explains: “Recommender systems often face a trade-off between presenting the items that you are likely to love (by exploiting what they already know) versus presenting you an item that the recommender system has no clue whether you’ll love, hate, or don’t care (exploration). Too much exploitation risks boring you (imagine eating the same favorite dish every day), and too much exploration risks being frustrating or useless.” Marketers have to be cognizant of the risks, and strike the right balance.
Be aware of algorithmic bias
The algorithm itself can amplify a specific piece of content; it’s important to assess whether content is really useful to your audiences. “A major issue in recommender systems is the fact that popular items, obviously, tend to be liked by several users, and hence it is safe to recommend popular items. However, this approach has several pitfalls. Arguably the most important of all is that the algorithm might make something popular itself by presenting it to several users, who click on these items, which the algorithm might interpret for popularity, and recommend it to even more users, and so on. The downside is that other items are rarely recommended and risk being completely ignored. This is one kind of algorithmic bias that can affect real-world users,” says Bilgic.
Recommendation engines are helping shape our digital experiences. As they continue to become more intelligent, incorporate a wider range of sources, and develop sophisticated capabilities such as identifying gaps in the content creation cycle, they’ll become more useful to marketers and users alike. By focusing on best practices such as transparent communication, assessing whether content is truly connecting with audiences, and avoiding biases, marketers can leverage this new generation of technology to deliver personalization at scale.
For more about Skyword Personalized Recommendations, visit our resources center.
Featured image attribution: Jamison McAndie
The post How the Recommendation Engine Explosion Is Changing Martech and Content Consumption appeared first on The Content Standard by Skyword.