We regularly get questions from our followers and also members about the technology we use at Cigar Sense.
David Wells, vice-president, co-founder and technologist at Cigar Sense, explains how Cigar Sense works, why our technology is not AI based and gives us insights on many more connected aspects.
How does Cigar Sense work?
At its core, in the Cigar Sense system, we have essentially two databases.
- We have one database containing cigar profiles and it contains lots of attributes, over a hundred attributes on each cigar. Those are all derived by our panelists doing blind tests according to a rigorous method. It’s a tremendous database with detailed profiles of thousands of cigars, the aromas and tastes that they present and lots of other characteristics.
- The other database is essentially information about our members who have joined Cigar Sense and created their their personal preference profile.
The core of the service is to recommend to our members which cigars are going to best suit their personal preferences. And that’s those two databases really coming together. We find that best intersection. So, as a member of Cigar Sense, you can look up any cigar that’s in our database and see how well or how poorly it fits your personal preferences.
There are a couple of ways that users can tell us what their preferences are. One of our onboarding methods is basically connected to an inference engine. If you can tell us a few specific cigars that you know you like, then our engine will infer the characteristics of those cigars, create a composite, and that becomes your initial set of preferences.
The other way is manual: you can tell us specific aromas or specific tastes or other descriptors that you do or don’t like, and we’ll take that into account.
One thing that is really important is that we don’t assume that everyone has the same priorities. So, part of the onboarding process is in specifying how important are all these characteristics. So, this means what is the relative importance of flavors versus the nicotine strength, versus the type of wrapper, the ring-gauge, the price… We have many different preferences that are not strictly the olfactory and gustatory ones. Some of them might not matter to you. So, for someone who smokes any sizes, from small cigarillos to huge gordos, probably the ring-gauge doesn’t matter at all. If you tend to like a fairly narrow range of nicotine strength levels, and a fairly narrow range of gauges, your profile reflects both of those preferences and the fact that those characteristics are important to you.
Knowing how to name an aroma
We have observed over time how difficult it is for cigar lovers in general to name what they like and what they don’t like. You have to respect consumers for what they can tell you, and not expect from them some impossible answers that you would like to have. So, on one hand, our members can select each descriptor that they like and they don’t like, on the other hand we need to help them out.
In fact, we do see evidence that sometimes consumers are a little confused, especially early on, with our aromas taxonomy, for example. Our panelists go through training to normalize the language: if we use an aroma category like “animal”, the panelists understand exactly what falls within that category. Our aroma wheel also tells our members exactly what falls in that category. But it’s possible that that a consumer might not understand when they see a term like “animal”. Actually, they may have a wrong impression, as there’s often a natural aversion to the concept of “animal” as an aroma. There is an association that is made to something dirty. And it’s very interesting to see that leather is one of the most recurring aroma descriptors in a cigar. But sometimes people don’t know that they just don’t know how to name it. And when you tell them it’s part of the “animal” category, they don’t react very positively.
The solution: Profile Advisor
The input that you give us, it’s not just something that happens one time during the onboarding into the service, but it’s actually really important that over time you continue to give us feedback on specific cigars you like and you don’t like. In fact, that’s what actually triggers the Profile Advisor to essentially do a reassessment of your profile against the compendium of all the cigars you’ve liked.
So, each time you submit a tasting note, which is the way that you give us that feedback, Profile Advisor will look at all the cigars that you’ve liked and will again create a composite of the characteristics of those cigars and compare it to your profile. If there’s a substantial deviation between them, it’ll tell you. For instance, it tells David all the time that “you say you don’t like bitter, but in fact, many of the cigars that you like present bitter”. What that represents is really a difference in the way the consumers thinks about what they’re perceiving versus the trained panel descriptions. The panelists perceptions are normalized with a specific vocabulary. So a consumer might imagine that they don’t like bitter, but then again, in David’s case, he loves to drink a Manhattan. And it’s definitely not the same without the bitters in it. So, what Profile Advisor is trying to do is it’s trying to help you make sure that your profile is dialed in correctly.
Also, if your preference has actually changed over time, if you’re with us for a matter of years for example, it’s possible that your tastes evolve. Especially if you joined us relatively early in your career as a cigar smoker, and as your tastes evolve and you give us that ongoing input, Profile Advisor will help you tweak, if needed, your profile.
The objective is always to make sure that we have the best understanding we can have of what you truly like and dislike, so that we can give you the most accurate recommendations of cigars that you like.
Helping the industry with a standardized language
This matching of cigar profiles by the trained panel, with the detailed, personal preferences of our members allows us also to do something which is revealing to be very informative. It is indeed combining the preferences and the importance of the preferences of consumers with the standardized language, which makes it much easier for anybody in the industry to understand their consumers. Many manufacturers, for instance, have confirmed to us how difficult it is to actually make some sense of all the social media comments, or to actually capture accurate information from focus groups, in a way that is really actionable.
Is this AI?
One question that we get very frequently is “is this artificial intelligence?”
Everybody’s getting more and more exposure to artificial intelligence. The current state-of-the-art of AI in most cases is also sometimes referred to as machine learning.
And that’s actually not the method that we use.
The way machine learning works is essentially you train an algorithm. By having it examine an enormous set of data and process it thanks to an algorithm, you can look for statistical correlations between different aspects in the data, and that’s not what we’re doing.
And the reason we don’t work that way is because we actually find it less reliable, less accurate than our method. Our method is more of what would be traditionally called an expert system. The difference is: we don’t just look at the fact that you liked A and B and you know that 90% of the people who liked A and B also liked C. That’s the basic method that most recommenders use.
On Amazon or on eBay, or even at the big online cigar retailers, they’re watching what you’re looking at and what you’re buying. They’re giving you a recommendation. But most people have probably seen, there’s a pretty high error rate in those recommendations.
The thing that is missing is an understanding of causality. Why do you like A and B? And that’s exactly what our inference engines do, in our Profile Advisor and during onboarding. They do not only look at just a coincidence that you liked A and B, but they look at the characteristics that you liked. They form that composite profile for you that they then use to either create your initial profile or to check, to see if your profile needs to be adjusted in the case of Profile Advisor.
It’s that understanding of the causality, of the characteristics you like, that helps us give such reliable recommendations. This has absolutely nothing to do with what other people have done or what other people have liked. It’s all about you as an individual who has the right to make your own choices among thousands of cigars in our database, as opposed to imitating other people’s taste.
Anyone who’s familiar with the music service Pandora, that’s absolutely how it works. They created what they call the musical genome and they try to figure out what are the characteristics you like to suggest other music.
One other thing that comes up pretty often, is that data privacy is something that’s really growing importance across the world. And this is both regulations and ethics and respect for consumers.
The GDPR, as an example, regulates data privacy. We designed for this from the very beginning, when it was not mandatory. We never reveal to any other party the personally identifiable information of our members. So when you register your information never goes anywhere else but in our database to serve you better. We do sometimes share anonymized statistical data with cigar companies. This means that we will never tell a cigar manufacturer your email address and which cigars you liked. We might tell a cigar company, for a given geography, what specific characteristics consumers prefer or do not want. That’s all anonymized and contains no personally identifiable information.
We don’t share email addresses with anybody ever, in spite of many businesses monetizing this, because we believe in the importance of data privacy and that’s very important to us. So your data is absolutely privately secured and we’ve got very good security mechanisms.
Why isn’t there an app?
There’s another question that we asked pretty often.
That is one point that we’ve examined and then re-examined on multiple occasions. We don’t have an app. We have a mobile responsive website, which means that you can use on your browser, on your phone. Images reconfigure themselves so that the service is usable and looks good on your iPhone, your Android phone, your tablet, your PC or Mac.
The primary reason for which we haven’t actually invested in a native app, either for IOS or for Android is really the policies that the platform companies have around publishing apps that are related to tobacco. Anti-tobacco campaigns are very aggressive. It’s our understanding that very few of any cigar related apps are being approved by Apple app store and likewise at Google for the Android Google play store.
So it’s a business decision, more than anything else. We’ve not been willing to make the investment and hope that the app stores would approve our apps. We are hearing about lots of other people who are trying to get cigar related apps approved by the app store and they’re having trouble with that. So we try to stay away from that and really focus on delivering good features that people can use on any device.
On correlation vs. causation: https://www.cigarsense.com/correlation-or-causality/
On accuracy in recommender systems: https://www.cigarsense.com/accuracy-in-cigar-recommender-systems/
On personalization in recommender systems: https://www.cigarsense.com/personalization-in-cigar-recommender-systems/
On transparency in recommender systems: https://www.cigarsense.com/transparency-in-cigar-recommender-systems/
On user control in recommender systems: https://www.cigarsense.com/user-control-in-cigar-recommender-systems/
HELP US SPREAD THE WORD!
Please share Cigar Sense Podcast with your Twitter followers:Just listened to #CigarSensePodcast's awesome #podcast about the Cigar Sense #technology with @cigarsense! It's worth a listen. Check it out Click To Tweet
If you like this episode, help us keep the podcast alive. Head over to Apple Podcast, or your favorite player app and kindly leave us a review and follow the show!
Feedback and promotion on the show
We have no external investors, nor cigars or ads sale revenue. The people we accept/invite on our podcast are not paying us for the exposure we offer them. Just like for the Cigar Sense membership and our educational resources, we strive to provide independent information.