Filter blogs by:
or

Fifty Shades of Recommendations: A Trilogy - Part 2 | Viaccess-Orca Blog

recommendation.Sefy-iStock_000015452084XSmall.jpg

 

A Casual Fling can come back to haunt you!! The Five Dangers that will Doom your Recommendations

In the last post, we examined the rush towards Recommendation solutions and the motivation behind it. I also hinted that due to a variety of limitations, many of the early implementations of Recommendations have been premeditated attempts at under-achievement.

I should point out that, regardless of the genius of the Recommendations algorithms themselves, the greatest challenge to widespread adoption of Recommendations is getting that first chance. According to Forrester Research, fully one third of online consumers actually trust a stranger‘s opinion on a public forum or blog, more than they do the information provided by brands. In other internal focus groups executed by some Pay TV operators, the level of distrust exhibited towards Recommendations from the provider is nearly 50%. This does not bode well. With such skepticism as a starting point, the goal of becoming a trusted advisor is that much more difficult to achieve.

It’s the Experience, Stupid!!

A major part of encouraging the adoption of Recommendations has to do with the user experience itself. Early attempts, such as, “Get your Recommendations here!,” were naively designed with the belief that the directive, “if you build it, they will come,” could be universally applied (for all you Millennials out there who didn’t get the reference, Google the film “Field of Dreams”). Turns out that users were not prepared to go out of their way to expose themselves to what they considered to be blatant attempts to ‘sell them something’. Much more sophistication is required to design a UX that infuses the Discovery, including Search, Recommendations and Explore, into a navigation path that viewers want to pursue or are used to pursuing. Even more important, since viewers are courted daily by new services vying to provide for all of their content needs, you need them to WANT to come to you. It’s called engagement (more on that in our next post). Once they are actively engaged in what you’re telling them, you can pour on the Recommendations and the conversions will flow.

Once you have crossed the chasm of converting yourself from a TV and video catalog to a veritable interactive ‘Disneyland’ of content, you will get a chance to pitch your Recommendations to the viewers. According to James Kane, an expert in the science of loyalty and one of the foremost authorities on establishing trust with consumers, “establishing a sense of trust is critical to a loyal relationship. So all you have to do is gain their trust and you will be squarely on the path to a long-standing, committed relationship.

Alas, as relationships often go, the breakup is looming. Read on to find why…

So, the Honeymoon is Over? Here are Four Reasons Why (and One that May Surprise You)

Interestingly enough, very little of the Recommendations hype is echoed by consumers. Sure, for years now, Netflix subscribers have been aware of the Recommendations that are offered to them, but when asked, many of them voice frustration with the results. So is that a win for Netflix?

Once the novelty factor wears off, viewers who continue to use the Recommendations will experience disillusionment and decreased satisfaction, unless the personalized Content Discovery solution addresses the following challenges:

1.    The Recommendation Cliff: It’s not a Myth

When Recommendations are rolled out, you can expect a spike in conversions, i.e. orders or purchases that are executed as a result of a Recommendation. If the operator has been savvy enough to couple the Recommendations with some sort of free usage promotion, the numbers should be very rosy indeed. However, when the honeymoon is over”, and my apologies to all you romantics…- we face a cold truth: a single Recommendations’ vector will only generate 5-7 relevant Recommendations from any given input. Once those movies or shows are consumed, the next Recommendations down the line are much less attractive. If we assume that the viewer eagerly follows all 5-7 Recommendations, he or she will surely enjoy the selections. But at the same time, they will have exhausted the supply of relevant Recommendations, so subsequent Recommendations will be much less appealing and the satisfaction curve will, you guessed it, fall off a cliff.

One might argue that, based on the additional evidence generated by those first 5-7 orders or purchases, the system may generate yet another batch of relevant Recommendations, however, this new evidence is merely a reinforcement of the initial vector, in essence, a sell-fulfilling prophecy. So the new evidence simply piles up on top of the old, and the resulting Recommendations, far from generating excitement and introducing new content experiences, very quickly become monochromatic and boring. To counter this and to create more user satisfaction, research proves that additional vectors or networks of Recommendations need to be layered on top of the first in order to provide more depth and perspective to the Recommendations system as a whole. For those who like to wade through academic research, you’ll find an exponentially more learned explanation in the research of Dr. Gal Oestreicher-Singer.

2.    Winter is Coming

The cold start challenge never really goes away. Many people have learned to be wary of the cold start, that early phase of the Recommendations rollout when there is little prior usage history to rely on when making Recommendations. While it is certainly true that this phase has to be addressed, what is less obvious is that throughout the lifetime of the service, there are always cold elements that need to be handled. Newly created accounts and new users will constantly be added to the Recommendations system and they deserve high-quality Recommendations right from the start. New movies and content catalogs are also loaded into the system, fresh off the DVD window or newly inked contracts, with little or no historic data to classify them in the Recommendations model. Therefore, the challenge of the cold start is an ongoing one, which if not handled gracefully, could prove to be a turn-off for your hard-earned existing users.

3.    Highly Social and Utterly Friendless

There are few terms that are as widely used as the term “Social,” especially when trying to demonstrate that we are hip and in tune with the younger audiences. While Facebook is fire-breathing proof of the power of the social phenomenon, to leverage the power of social effectively, we need to break it down and address all aspects of it. There is social, as in what vocal people on social networks are saying, i.e. Tweet Buzz. There is social, as in what active people who are on social networks are doing via check-ins, etc. And there is social, as in what my friends are actively saying and doing. What some service providers fail to recognize is that it is not enough to offer the brand of social that is most convenient to them. Imagine searching through your social Recommendations only to find they have nothing to do with what your friends think and everything to do with the opinions of your teenage daughter’s Twitter- friends. The social functionality must be an organic part of the overall experience, and not a hastily slapped on add-on. Moreover, since the social juggernaut is anything but consistent, social Recommendations by themselves do not adequately cover the lineup or catalog, leaving many assets unaddressed (the directive, “Be the first to review this,” is the digital equivalent of roaming a deserted hallway and hearing the echo of your own footsteps). Therefore, social Recommendations are at their most effective when they are layered on top of other Recommendations sources, supporting those Recommendations with a valued second opinion.

4.    Getting Personal with Everyone can sometimes be a Bad Thing

Our high school nurse taught us that getting personal with too many people at once could result in some painful afflictions. The reason you can’t flag this last sentence for inappropriateness is the same reason that “Personal Recommendations have acquired a bad reputation. There are simply too many possible interpretations where the term social is merely misused; the term ”personal” is downright abused. Many systems boast of personal Recommendations that are either generated ‘en masse’ for all viewers, or use very broad criteria to group viewers into clusters and while there are a variety of algorithms that can help predict what might be interesting to a specific viewer, based on previous viewing, their favorites or people with similar tastes, ultimately, people are more complex than their viewing history. Viewers are less than accurate when indicating their explicit preferences and they are absolutely difficult to pigeonhole into a specific type. To get personal, you need a personal touch. This means applying the algorithm that best suits the viewer in a givencontext by casting a wide net that uses different algorithms simultaneously. For example, some people will be comfortable with you invading their personal space and mining all of their personal data, while others will run for the hills at the any hint of Big Brother. Moreover, offering meaningful personal Recommendations ultimately means providing non-obvious Recommendations, those assets that a viewer would otherwise not have found. These are the Recommendations that individual viewers will appreciate, not one-size fits all Recommendations that happen to be theatrical mega-hits. To do this, a Recommendations system needs to dig into the catalog beyond the blockbusters or the highest rated and popular titles, to retrieve relevant movies, even if they are from a different era or a different country. Many of today’s Recommendations systems are focused on milking the top 20% of a catalog to pick the easy wins, but these are the Recommendations that are easiest to forget and do not leave any lasting impression on the viewer.

5.    Surprise: Even Good Data can Go Bad

It turns out that even well preserved data has an expiration date. Have you ever tried buying toys for a child you used to know, maybe a niece you now meet once a year? Their preferences have the life span of an adult mayfly, so good luck with that!! The fact is, with all the focus on collecting every last shred of consumer behavior for big data wizardry, companies often lose track of a simple truth: many people don’t appreciate being reminded that much of their past. Who among us haven’t done things we are ashamed of: the odd late night movie-of-the-week we watched in a moment of weakness; the stray chick-flick that the girlfriend ordered when we weren’t looking or the mind-numbing flop of a movie in which good actors got caught up in a bad script? We thought that by not talking about it the next day it would be left between us and the remote control. Yet, to our horror, here we are getting “because you watched this” Recommendations. The fact is, research shows that even movies we once thought were perfectly enjoyable have, after a while, absolutely no connection to our current tastes. This means knowing which evidence to discard is almost as important as knowing which evidence to collect. The result can be that an unsuspecting algorithm is now serving up Recommendations based on years of accumulated data that is more misleading than helpful. Not paying close attention can be the downfall of any relationship and it is guaranteed to result in poor Recommendations as well.

We have now pointed out some of the hidden icebergs that can sink your Recommendations system (and your credibility along with it). Before your relationship with your viewers hits the rocks, allow us to give you some free couples therapy.

Tune in to our next post to hear how to listen and be listened to so your viewers and you can enjoy a long-lasting relationship.

Interested in learning more about content discovery? Click here to download our latest white paper!

 

  • Home
  • Blog
  • Fifty shades of recommendations a trilogy part 2