FREE! Subscribe to News Fetch, THE daily wine industry briefing - Click Here


SPONSORED BY:
SCA-WII-April2021-160

Commerce7Click to learn more

Why solving the “Paradox of Choice” is a major reason Netflix’s recommender is worth $1 Bil/yr. And why wine fails at that.

NOTE: This is article #2 in our series on why wine needs a Netflix-like recommendation solution.

 

Article #1 is here: How A Netflix-Style Recommender Is Vital to Reversing Wine’s Market Marginalization

 


“[H]umans are surprisingly bad at choosing between many options, quickly getting overwhelmed and choosing ‘none of the above’ or making poor choices”Carlos Gomez-Uribe, Netflix VP of product innovation & Chief Product Officer Neil Hunt.

The “Paradox of Choice” is one of the two main reasons that Netflix values its recommender system at $1 billion or more. The second reason has everything to do with why Netflix discarded its use of ratings and reviews.  That  will be covered in the next article of this series.

 

Interestingly enough, Netflix first realized that  “Paradox of Choice” for movies has a lot in common with wine, chocolate, and upscale fruit jam.

 

The common thread among those products lies in a scholarly paper  (free PDF) published in 2015 by two of the key Netflix developers of the recommendation system: Carlos Gomez-Uribe, VP of product innovation and Chief Product Officer Neil Hunt.

 

In that paper, they credit a classic book by a Swarthmore professor with providing a key concept based on several consumer choice decision studies.

One study was set in a gourmet food store in an upscale community where, on weekends, the owners commonly set up sample tables of new items.

When researchers set up a display featuring a line of exotic, high-quality jams, customers who came by could taste samples, and they were given a coupon for a dollar off if they bought a jar.

In one condition of the study, 6 varieties of the jam were available for tasting.

In another, 24 varieties were available. In either case, the entire set of 24 varieties was available for purchase. The large array of jams attracted more people to the table than the small array, though in both cases people tasted about the same number of jams on average.When it came to buying, however, a huge difference became evident.

Thirty percent of the people exposed to the small array of jams actually bought a jar; only 3 percent of those exposed to the large array of jams did so.

That was one example in The Paradox of Choice – Why More Is Less, by American psychologist Barry Schwartz. At its core, the book describes how the time and thought-consuming task that bedevils consumers as they grapple with too many choices. And how that leads to anxiety and frustration.

 

And, as the result of the anxiety and frustration, people often give up and simply do nothing.

 

As the Schwartz book states:

A large array of options may discourage consumers because it forces an increase in the effort that goes into making a decision. So consumers decide not to decide, and don’t buy the product. Or if they do, the effort that the decision requires detracts from the enjoyment derived from the results.

Or, as Gomez-Uribe and Hunt put it:

[H]umans are surprisingly bad at choosing between many options, quickly getting overwhelmed and choosing “none of the above” or making poor choices (e.g., see Schwartz [2015]). [Emphasis added.]

Consumer research suggests that a typical Netflix member loses interest after perhaps 60 to 90 seconds of choosing, having reviewed 10 to 20 titles (perhaps 3 in detail) on one or two screens. The user either finds something of interest or the risk of the user abandoning our service increases substantially.

The recommender problem is to make sure that on those two screens each member in our diverse pool will find something compelling to view, and will understand why it might be of interest.

Better Recommendations = Reduced Stress = More Sales + Greater Customer Retention = More Profits

When creating the foundations of the current Netflix recommender system, developers Gomez-Uribe and Hunt recognized the psychological basis of why consumers get worse at making a decision as the number of choices increase.

 

They also recognized that helping Netflix users make better decisions without stress and confusion was vital for the company’s financial success.

Our subscription business model suggests a framework to find the answer. Because our revenue comes exclusively from the monthly subscription fee that our current members pay, and we make it very simple to cancel the subscription at any time, we think that maximizing revenue through product changes is fairly equivalent to maximizing the value that our members derive from our service.

Revenue is proportional to the number of members, and three processes directly affect this number: the acquisition rate of new members, member cancellation rates, and the rate at which former members rejoin.

If we create a more compelling service by offering better personalized recommendations, we induce members who were on the fence to stay longer, and improve retention. [Emphasis added]

In addition, all members with an improved experience (not just those on the fence) may be more enthusiastic when describing Netflix to their friends, strongly influencing new subscriber acquisition through word-of-mouth effects. Both recall of a better experience and stronger word-of-mouth may influence former members to rejoin more quickly.

The paradox of overwhelming choice looms huge for wine

Netflix: 197 million users, thousands of choices

  • U.S. Users – 197 million (60% of the population)
  • Product Inventory – 5,879 video streams
  • Customers Per Product: 33,509

Here’s how that adds up:

Netflix: U.S. Users – 197 million (60% 0f population)

Paid subscribers, US and Canada, Q3 2020 73 million (reported by the company). Historically, U.S. is estimated at about 90% of UCAN.

 

That would put US Q3 2020  (73 million x 0.90 ) at 65.7 million subscribers. Which is about 20 percent (65.7 million /328 million total population).

 

Netflix allows up to five personal profiles per paid subscriber. While estimates vary about the average number of profiles per subscriber account, three could be a reasonable estimate.

 

Screen Shot 2020-12-08 at 2.56.48 PM

That means that the number of users (as opposed to subscribers) is 3 x 65.7 or 197 million people which equals an estimated 60% of the U.S.population.

 

Netflix: Product Inventory – 5,879 video streams

Netflix had at least 13,941 titles across all its international libraries as of April 2020. The US has the most Netflix titles, with a total of 5,879.

Netflix: Customers Per Product – 33,509

The Netflix recommender system must accurately help each of them decide (in 60 to 90 seconds or less) which of the 5,879 video streams in its U.S. title library will make them happy.

 

As a result, Netflix retains 92% of its customers.

Wine: Product Inventory – Probably 160,000+

No one actually knows.

 

However, in the United States, a label approval is necessary for a new wine.

 

If one assumes that half of last year’s wines will still be for sale in the present year and NONE of the wines from previous vintages are, then 160,000 would seem to be a reasonable guesstimate.

 

That’s definitely low because minor changes in a label (such as the vintage date) do not need approval. This is significant for recommendations because vintages often differ in taste and consumer preference which makes them a separate product.

116,000+ new wine products approved by TTB in past 12 months: bw166

Wine has approximately 28 times more products but only about 1/8 the number of people to work with.

  • U.S. Customers – 37.3 million (11% of the population)
  • Wine: U.S, Product Inventory: 160,000+
  • Wine, U.S. Customers Per Product: 249

Because current wine and other recommendation systems (Amazon, etc) attempt to find similarities among people and their product choices, the more data a system has, the greater the potential for an accurate recommendation.

 

Conversely, the fewer people in a product population, the harder it is to tease a valid recommendation out of the data. This is greatly aggravated by the phenomenally larger number of wine choices as opposed to video.

 

In addition to 160,000+ new wines registered by the U.S. Trade and Tax Bureau every year, many wines from previous years are still available in retail channels.

 

In reality, there could be 300,000 or more to choose from. But like the sorry, impoverished state of the wine industry as a whole, “nobody knows.”

 

Further, for the sake of accurate recommendations, the taste of a Chateau La Plonk 2021 wine may differ substantially from a 2019 or a 2020 due to grape growing, blending, and winemaking variations. Thus a new vintage is a new product for the sake of recommendations.

Global data on the paradox of choice is even more mind- and palate-boggling for wine and Americans

The international choices are even more mind- and palate-boggling and — ultimately — unknowable.

 

According to Julian Perry, CEO of  wine-searcher — the world’s largest, most accurate and most used search engine, price comparison engine, and database for wines, spirits, and beers — the number of wines for sale at any one time around the world at any one time is estimated at greater than 1 million.

 

This is significant for the U.S. which is the world’s largest consumer of wine.

 

Screen Shot 2021-01-17 at 9.52.40 AM

Source: OIV (International Organisation of Vine and Wine) via Statista (A hectoliter is equal to 100 liters, approximately 26.4 U.S. gallons)


 

How did Netflix solve the Paradox of Choice (And boot ratings and reviews)?

“Consumer research suggests that a typical Netflix member loses interest after perhaps 60 to 90 seconds of choosing, having reviewed 10 to 20 titles (perhaps 3 in detail) on one or two screens. The user either finds something of interest or the risk of the user abandoning our service increases substantially. The recommender problem is to make sure that on those two screens each member in our diverse pool will find something compelling to view, and will understand why it might be of interest.” —  Paper  by two of the key Netflix developers of the recommendation system: Carlos Gomez-Uribe, VP of product innovation and Chief Product Officer Neil Hunt.

Netflix found that ratings and reviews failed miserably in solving the Paradox of Choice  for millions of people with thousands of possible choices within a minute or minute and a half.

 

However, for now it is vital to understand that the Netflix recommender is not An algorithm, but a system of multiple algorithms, a number of which personalize the recommendation screens for each Netflix profile.

 

Personalized Video Ranker

From the Gomez-Uribe, Hunt study:

Now, our recommender system consists of a variety of algorithms that collectively define the Netflix experience, most of which come together on the Netflix homepage. This is the first page that a Netflix member sees upon logging onto one’s Netflix profile on any device (TV, tablet, phone, or browser)—it is the main presentation of recommendations, where 2 of every 3 hours streamed on Netflix are discovered. An example of our current TV homepage is shown in Figure 1.

It has a matrix like layout. Each entry in the matrix is a recommended video, and each row of videos contains recommendations with a similar “theme.” Rows are labeled according to their theme to make the theme transparent and (we think) more intuitive to our members.

Personalized Video Ranker: PVR:There are typically about 40 rows on each homepage (depending on the capabilities of the device), and up to 75 videos per row; these numbers vary somewhat across devices because of hardware and user experience considerations. The videos in a given row typically come from a single algorithm. Genre rows such as Suspenseful Movies, shown on the left of Figure 1, are driven by  our personalized video ranker (PVR) algorithm.

Screen Shot 2021-01-29 at 9.14.34 AM

As its name suggests, this algorithm orders the entire catalog of videos (or subsets selected by genre or other filtering) for each member profile in a personalized way.


The resulting ordering is used to select the order of the videos in genre and other rows, and is the reason why the same genre row shown to different members often has completely different videos. Because we use PVR so widely, it must be good at general purpose relative rankings throughout the entire catalog; this limits how personalized it can actually be.

More detail (and diagrams) on the system of algorithms and how they work to solve the Paradox of Choice at the Netflix scholarly paper and in our next article, along with a discussion of how data blew away ratings and reviews.


Next article in this series: Why Netflix bailed on ratings and reviews. And why a recommender for wine has got to do that too if it wants to sell more wine to more people.