Published 23 November 2012
“Data is power” was the catchphrase of Bumblebee from Transformers in the 1980s
Bumblebee’s “Data is power” catchphrase sums up late 20th-Century thinking but, in the 21st-Century, I believe that data is only powerful once paired with categorisation.
Imagine I gave you a list of 1,000 names and email addresses. Great, you can mail them about your latest product. Your click-through rate will probably be 4-6%. But if my list told you which categories each address fitted into for age, gender, location, spending power, etc, then it becomes truly useful: you can tailor your marketing to specific groups and reap much better click-through.
Our modern world relies on categorisations: everything from Amazon to your favourite online wine store uses them. At its most basic you’ll be narrowing down your search to red or white wines, maybe choosing a price band. But can we do something better for wine shoppers? Most customers don’t know that if they like left-bank red Bordeaux then they might like Coonawarra reds, or that their beloved Sauvignon Blanc from NZ uses the same variety as Sancerre. Can we build a system that doesn’t rely on consumer knowledge, but still offers up intelligent recommendations to them? Can we build a system that uses facts about the wine they like to return recommendations, rather than the fashionable crowd-sourced ratings? Do you really trust the masses to recommend wines to you when taste is so subjective? Blossom Hill anyone? No thanks.
The super-database
My previous post asked what the main quantitative factors are for a wine. I am pretty happy that sub-region, vintage, colour, grape varieties, alcohol level and residual sugar have it pretty wrapped up. By taking a bottle of wine that a consumer likes, we can easily obtain this objective information: any simple database could hold this information. But what can we do with it?
Let’s take an example. I like the wine I am drinking right now. The objective data:
- Region: Carcassonne, France
- Grapes: Syrah (100%)
- Colour: red
- Vintage: 2011
- Alcohol level: 14%
- Residual sugar: 1.3g/l
Don’t think that tells us much? Wrong. With our database holding similar data for the other wine in our stockroom, it is possible to build an algorithm to find wines similar to the starting point (i.e. the bottle the consumer likes). Or slightly different if the consumer chooses to deviate. For example, knowing that I like Syrah it could take me away from Carcassonne and show me wines from Côte Rôtie, Saint-Joseph and Hermitage. Or it could return wildly different suggestions, if that’s what the consumer wants. You could even feed in a bottle you didn’t like and ask for something diametrically opposed. It empowers the consumer to choose how much they want to deviate from their starting point.
Does this remove all the fun and adventure? 99% of consumers aren’t looking for an adventure; they just want wines they like. But, our database even caters for those that want a journey. Give it a starting point and it could take you on an adventure by selecting wines that are different in some factors but not others.
How about using it to sell tailored cases? As a retailer, you pick 12 bottles that you think represent (for example) reds from around the world, the customer selects how much they want to pay, and our super-database returns 12 bottles that are as close to your recommendations that their selected price point will allow.
Conclusion
“The goal is to transform data into information, and information into insight”
~ Carly Fiorina (President of Hewlett-Packard in 1999)
Most consumers don’t know a lot about wine but nearly all are able to show you a bottle they like. By taking this, and feeding it through our super-database, we could allow consumers to select wines as similar or different as they want. It would empower the consumer without asking them to have any knowledge of “why” they like it.
Vintage is the most complex to deal with, but one of the most exciting. In our example, we know that the 2011 Languedoc season started well, with vines ahead of schedule and holding healthy grapes, but that conditions then changed drastically towards harvest. Languedoc wines from 2011 are especially varied, depending on how winemakers responded to the difficult conditions (in terms of yield, microclimates and harvest strategy). As a result, our algorithm would give the vintage for this region less weight, because we know that there is likely to be variation between producers.
Fortunately, we know this information is obtainable because Tom Stevenson [http://en.wikipedia.org/wiki/Tom_Stevenson] used to produce this – in far more detail than we would require – in his annual Wine Report. Although our database would not be able to store every nuance of every vintage, it would be able to get the main aspects, apply an overall quality value and weight of consistency, and use in our algorithm for recommendations.
David, the technologist in me is always fascinated by recommendation engines. Although they usually end up being dramatically complex, iterative processes if they are to deliver any value.
However the #winelover and customer of many independent wine merchants in me recoils in horror. The issue is that the “if you like this then you might like this too” approach is surely bound to result in more, rather than less homogenisation. You saythat it would be possible to build in “adventure” but i doubt this is the option most would choose.
Human interraction is so much richer, and more likely to foster loyal customers (evenin a supermarket).
Perhaps a more interesting use of a resource like this would be as a “train the trainer” tool – for budding wine retailer to refer to.
Hi David – I just ran across your site, as a new wine, food and travel blogger. On the one hand, I, of course, love the idea of being able to “press a button” and get the spot on wine that hits my palate just so (at an attractive price as well, why not?) But I fear, like Simon Woolf above, that there are so many intangibles. Haven’t we all had the blockbuster from one vineyard and the cooking wine from a few steps down the road? Don’t viticulture and “handling” come into play? I know there’s a lot less futzing with wine in Europe (or at least they won’t admit it as readily as my fellow Californians.) But it seems like there are a whole lot of variables between the lines on this one. I’d love to hear if you tested out your theory…In fact I’d love to sign up as a lab rat
Cheers! Sonomaist