Taistr - In Search of Your Next Best Sip Locally

August 31, 2022

Humans are complicated creatures. It is not just other humans that are mysterious and unpredictable, sometimes we don’t even know why we are the way we are. We all have driven or walked alone every now and then and contemplated how who we are shapes the choices we make in life. We may never have the answer but the kaleidoscope of human diversity and availability of consumer choices in the marketplace are what make this world exciting and worthy of waking up to.  

Wine is one of those life preference mysteries. Some like white, some like red. Some like it fruity, some like acidic. Some like it lean and dry, others like it rich and sweet. But there are many who just don’t know what they want and choose a bottle because it has a shiny award label, like Barefoot for $20.99  or so. It is not only consumers who suffer from information overload. With the diversity of consumer preferences, wine retailers find themselves with no alternative but to stock some of everything. Often retailers rely on advice from procurement agents who make recommendations aligning with their sales commissions. This can result in inventory not selling, bringing carrying costs for storage, handling, shrinkage, and insurance. Adding to the problem of overstocking to compensate for the variety of customers’ needs, store clerks typically do not have enough knowledge of their customers’ preferences, and not being trained sommeliers, the wines carried by the store to make appropriate recommendations. Additionally, these recommendations can sometimes be biased.

Red Wine Choices

White, Sparkling and Rosé Wine Choices

Calgary-based startup Taistr is looking to address these problems experienced by retailers and oenophiles using modern data science and their proprietary recommendation engine. The goal is essentially to get past the fallback choice of selecting the best label on the consumer end and allow retailers to use customer preference analytics to make targeted procurement decisions based on statistics.

Taistr identified Alberta as its key market and for good reason, Alberta is the only province in Canada where liquor retailing is entirely privatized, with the warehouse management still owned by the Alberta government. This is a legacy of the Prohibition Era, as documented in the HBO show Boardwalk Empire, when governments across much of North America stepped in to protect consumers and families in the absence of public health measures. With privatization came Adam Smith’s invisible hand and the explosion of liquor stores with a catch – the provincially run Connect Logistics Services offers a levelled playing field between big retailers and independent retailers and offers no volume discounts so retailers big and small pay the same price for liquor.

HBO's Boardwalk Empire on The Prohibition Era (Behance)

Seeing the opportunity of a hugely fragmented market with many retailers lacking the technological and data science know-how to solve this economic inefficiency, Ontario squash playing buddies Iain Crozier (serial entrepreneur whose last venture developed safety surfacing for playgrounds) and Paul Gartenberg (patent award-winning engineering executive), along with Iain’s wife Ashleigh (also serial entrepreneur) founded Taistr in Calgary. With 52 retailers (39 in Calgary and 13 in Edmonton) in Alberta already signed on to the platform, the three founders are excited to be part of the Alberta tech ecosystem. They hope to further grow their presence in Alberta with the goal of adding 30 additional stores and 20,000 users in the coming months and to launch in Toronto in Q4 2022.

Taistr started with a focus on customer-related recommendations, but more recently they’ve shifted to providing vendors with data insights that were previously non-existent. With the help of Taistr’s product and capabilities, these businesses are beginning to make decisions supported by data analytics, which far surpasses previous strategies, such as using historical sales and relying on the internet to help predict trends (ie. Birdbox-style blindfold wine procurement). Taistr is now committed to making B2B a key component of their business and continuing to innovate through data science.

In a similar way that Netflix suggests movies to us, Taistr takes your personal preferences in terms of wine selection, compares you with others in a grouping that matches on these traits, and returns a curated list of wines for your taste. This approach aims to solve the problem of information overload for both consumers and retailers with the benefits that consumers can have recommendations that match their preference from similar wine fans, and retailers benefit from the data insight thats help them to procure the wine that consumers will actually want to buy.

In general, recommendation engine is a subclass of machine learning and consists of two main types:

  1. Collaborative filtering
  2. Content-based filtering

While collaborative filtering makes recommendations based on user preferences similar to yours, content-based filtering makes recommendations based on the attributes of a product. For example, in collaborative filtering, if John and Amy like the same wine, and Amy highly rated a bottle of wine that John never tried, the system would recommend the wine Amy rated to John. In the case of content-based filtering, if John likes a red wine from Spain, the recommendation engine would recommend other red wines from Spain that share similar attributes, such as its tasting notes, to John. Content-based filtering suffers from an item belonging to multiple categories. For example, a book on car repairs can belong to the book, car, and car repair categories, so the recommendation system would struggle to recommend a product: is the consumer after the book, a car or a car repair shop? On the other hand, collaborative filtering suffers from the “cold start” problem, where initial user data is needed for the engine to make recommendations. This is why Netflix or other customer portals ask users to click through preference questionnaires. To address these individual shortcomings, a hybrid system is generally employed.

Collaborative Filtering vs Content-based Filtering

This hybrid approach is what Taistr uses, combining content-based and collaborative filtering. When the app onboards a new user, it first asks s/he/they to take a preference test to glean their wine tastes, which allows the engine to utilize this data to make a recommendation, based on similar users’ profiles. As the user continues their wine journey through the app, the data captured by the engine learns more about the user’s preferences and continuously optimize the engine to make accurate recommendations and find the wines that best meet the user’s tastes.

While a wine recommendation engines are not new, with Vivino being the biggest name in this category, a different issue is that it is very easy to find highly rated products online and only to find out that the product doesn’t ship to your country. Despite the seemingly permanent pandemic-related shift to online shopping, and the associated decimation of brick-and-mortar retailers, the same cannot be said for liquor retailers. Over 65% of the market share in wine retailing is still through brick-and-mortar. We all heard alcohol use skyrocketed during the pandemic, and liquor stores were raking in Christmas-leadup sales levels because of the immediate liquor consumer behaviour of “needing it now’.  

Taistr fills that immediacy void by building close relationships with local retailers and recommendations are based not only on the users’ preferences, but also the local availability of products. As users browse through the referrals, once they find what they want, with only a few clicks, the wine will be conveniently available for pickup or delivery, thereby completing a user’s wine journey from discovery to enjoyment, usually on the same day.

While some recommenders have a bad rep for giving unfair visibility to paid sponsors like Amazon’s “sponsored” products or Google promoting their own products in their search engine, studies show that it only takes one or two lousy recommendation experiences for a customer to lose trust in its recommendation engine. It stands steadfast in its value of unbiased recommendation. This contrasts with other competing business models of recommending what’s popular to churn sales to myopically profit in the short-term while potentially sacrificing customer trust in the long-term.

With both the consumers benefiting personalized recommendations, and retailers saving money on their inventory costs, all wine lovers can cheers to that!

May be an image of 2 people and indoor
Founders Iain Crozier and Ashleigh Crozier

More from the YYC Data Blog