Wine recommenders, and the tyranny of choice

Half a century ago, a man named Alvin Toffler published a book called Future Shock (1970). This was a big thing in my youth, and it set the scene (and the tone) for much of what has happened in human societies since then. You see, the book was about the way in which the world had dramatically changed since the end of World War II, and what the effects of this might be on cultures worldwide.

One of the fundamental tenets of the book was about the increasing number of choices that we all face. Back in the Old Days, most people had relatively few choices in their lifetime, because they lived in a geographically and economically restricted world. The more restricted your life, then the fewer choices you actually have — most things happen to you, rather than you choosing them.

Well, the various cultural revolutions that started in the second half of last century put paid to all of that, except in the poorest communities. We now face a myriad of choices every day; and Toffler could see the possibility of this becoming overwhelming, at a personal level. After all, we have no inbuilt mechanisms for making choices rationally, and no cultural background (at that time) that could aid us. This had never happened before in history; and it has become know as The Paradox of Choice (freedom of choice is Good, but enforced choices can be Bad).

This led to the issue that Toffler focussed on in his book: "information overload" — when change happens fast, it results in social confusion and the break-down of prior decision-making processes. How do we make choices in a world that keeps creating more and more choices for us? These choices range from life-changing decisions, all the way down to daily preferences. But the "shock" of the future is that the number of choices to be made will not decrease, simply because our societies are changing at a rapidly increasing rate.

The wine industry

So, what has all of this to do with wine? Well, wine provides one really practical example of a (presumably less than life-changing) set of choices. In my lifetime, the choices for wine consumption have fulfilled Toffler's vision perfectly. I don't, as my forebears did, drink local wine, beer or spirits, depending on what plants could be cultivated in the local climate. I can get wine from anywhere on the planet, due to revolutions in transportation, and I can afford an awful lot of it, due to economic and social revolutions.

How do I make my choices? This is actually a pretty important question, not just for me personally, but for the wine industry as a whole. How are customers going to make decisions, when faced with potentially over-whelming choice, especially at the retail-shop Great Wall of Wine? What are the best ways for them to do so, and how do we help them do it? It would be a serious failure of the industry not to address these issues, sooner rather than later.

This is not an issue unique to the wine industry, of course. All industries currently face it, from vehicle manufacture, to an increasing number of media types (music, movies, TV), to holiday destinations. However, the biggest difference is the sheer number of products involved in the wine industry, compared to any other industry; and we make it worse by insisting that each single vintage is different (Vivino reports 13 million distinct wines in its database!).

To me, the ultimate outcome of failed recommendations is not so much that customers might make a wrong choice, but instead they make an arbitrary choice, which may then (for them) turn out to be right or wrong in some unpredictable way. The only alternative is for them to stick to the same old choice that they made last time, thus effectively ignoring all other choices. Moreover, this choice will be relatively inexpensive, so that the financial consequences of the choice do not particularly matter. This is no way to run a business.

Recommendation systems

The Information Revolution is supposed to be able to help us with this situation. The key concept is to create what is called an "information filtering system". Choices are addressed by eliminating all of the unsuitable ones, for any given customer. The customer can then make a much easier choice from a much smaller subset of possibilities, with some degree of confidence that this will not result in a wrong choice.

So, various industries have tried to tackle the issue in a computer-based manner, with varying degrees of success. I don't intend going into detail here, as there are plenty of articles on the web for you to read (start with Recommender system, at Wikipedia). However, all of the recommendation systems employ one or more of the following generic types of approaches:

• aggregated reviews from previous customers (eg. Amazon, TripAdvisor, Yelp, CellarTracker, Vivino)
• aggregated reviews from alleged experts (eg. Rotten Tomatoes, Wine-Searcher)
• club membership with standardized products (eg. Winc, Tasting Room)
• content-based filtering, which matches customer preferences to product characteristics (eg. Pandora, Wine Genius, Sippd, Bright Cellars, Vivino)
• collaborative filtering, which matches customer preferences to the preferences of other customers (eg. Netflix, Vivino)
• recommendations from one or more commentators whom we personally trust to have a taste something like our own (not really a computer-based system, but these days all of the commentators have social media presences).

When comparing these options, I personally like the last one best, when it works, as it represents the Old School way of doing things. The problem, however, is that there are too many such commentators, and most of them write about wines that I do not have access to (and some of the critics rely too much on ratings rather than reviews). So, this antiquated approach can be very effective in practice, but the customer has to put in a lot of time and effort. It is unlikely that young people are going to become wine customers if this is all we have to offer.

Of the other alternatives, I personally find customer ratings/reviews mostly worthless. A 5-star review tells me nothing more than that the person had a good experience, and a 1- or 2-star review says the opposite — I usually get no idea from the reviews why either person reacted the way they did. I therefore have to wade through the 3- and 4-star reviews, looking for some useful information. Amazon has stuck to this idea, but Netflix abandoned it.

Moving to the next option, one essential problem with content-based filtering is the nature of the content. That is, what characteristics do we measure about the product, and how do we describe them to the customer, in order to link the two? There have been some very ambitious attempts at this, such as the Music Genome Project for the Pandora music recommender. Sadly, such systems are often “data-rich but information-poor”, which is a well-known aphorism in the data industry (see my blog post: The Music Genome Project is no such thing).

Another issue is how the computer should actually link the products and customers. Artificial intelligence (AI) is becoming a common technological option (Old world, new tech: how the wine industry is taking on AI). The ideas behind AI, as currently applied to wine, have recently been discussed by the Wine Curmudgeon (“Curated” wine clubs, and why they don’t seem to be the answer). He has also previously had a go at standardized products from wine clubs (Blue Apron wine: disappointing and depressing) (see also: Are wine clubs worth joining?, and the recent Life under lockdown prompts wine club revival).

That leaves one final option, for which AI could also be used. In an ongoing series of articles, Lewis Purdue is currently discussing collaborative filtering, very positively (and critiquing most of the alternative methods):

1. How a Netflix-style recommender is vital to reversing wine’s market marginalization
2. Why solving the “Paradox of Choice” is a major reason Netflix’s recommender is worth $1 Bil/yr.; and why wine fails at that
3. Reviews and 5-Star are so useless for recommendations that Netflix tossed its prized $1-million algorithm; they’re even worse for wine
4. Wine wreck-commendations: Genes determine that no two people taste the same wine the same way
5. Netflix amped up recommendations with its own Big Data. What that means for wine
   
6. A “Netflix of wine” is impossible with current recommendation methods because Sight & Sound dominate Smell & Taste
7. Capturing individual perceptions of wine is the path to Netflix-quality recommendations

An example

This is not to say that people in the wine industry are not trying to create useful recommendation systems. However, I am not convinced that anyone has succeeded yet. We can look at the example provided by Vivino, which you will notice was listed several times above.

This system started in 2010, when a couple of Danes, who admitted to knowing very little about wine, decided to create a visual wine information system, on the basis that most people process information much better pictorially, rather than via text.

So, they took the basic model used by CellarTracker, which was originally almost entirely text-based, and then allowed people to access the wine data via pictures of bottle labels (using the Vuforia image-recognition engine). This worked, so they created an app, because in the modern world people access information using a mobile device — this is apparently now the most downloaded mobile wine app (see also: Wine: the trouble with wine apps).

They then decided to monetize the system, by allowing people to purchase wine through the app, via retail partnerships (Selling through Vivino - how's it working for you?). To do this, they basically adopted a content-based filtering procedure, in which customer preferences (based on their use of the app) are matched to similar products. That is, if you show an interest in a certain type of wine, then the app will direct you to allegedly similar wines.

More recently, they have moved towards a more collaborative filtering system, which matches customer preferences to other customers with apparently similar preferences. So, if some other customer is allegedly similar to yourself (based on their use of the app), and they like a particular wine that you have not yet tried, then the app will suggest that you might like it, too. Netflix has had great success with this approach to recommending movies.

We shall see how Vivino progresses (ie., is their current model going to work?); but you can see that they are at least trying.

The bottom line

The purpose of this (long, non-data) post is to emphasize the importance of the basic point. The contemporary wine industry confronts potential customers with a scale of choice almost unparalleled in the commercial world; and it currently does so with very little in the way of assistance for those customers. This is not a viable business strategy. The wine industry currently focuses on producing wine and on selling wine; but it does not (yet) fully address customer service in the Information Age.

In this sense, the wine industry is a hang-over from the Good Old Days. Back then, customer service was all about a person walking into a shop and being served individually by another human being, who may or may not succeed in helping that customer. This was inefficient, even back then, because it relies on the customer making the first (big) move.

Well, those days are rapidly going; and the current pandemic has accelerated the departure. I'm an old guy, and I miss being able to browse in shops; but it is too risky for me to do it just now, precisely because I'm an old guy. This issue is actually encapsulated in another idea from Alvin Toffler, noting that the personal touch is often the key, rather than a focus on information. After all, we are humans, not just information processors. Toffler said:

Society needs all kinds of skills that are not just cognitive; they're emotional, they're affectional. You can't run the society on data and computers alone.

The wine industry has always relied on this personal approach. However, it cannot work well in the face of information overload, particularly not the sort currently being perpetrated in the wine industry. We thus face our own choice regarding potential customers: start providing effective help with making choices, or drastically reduce the number of choices. Less really is more.