Snowy Cheese Flavored Latte: Starbucks and Machine Learning in China

Starbucks has been an innovator in big data and machine learning through its highly successful mobile order and pay application. How can that translate to China, its next leg of growth?

Starbucks is a leading roaster, marketer, and retailer of coffee in the US and globally. The company’s more than 27,000 owned and licensed stores sell a variety of coffee, food, and other beverages (e.g., tea) to consumers around the world [1].

Starbucks launched its mobile ordering and payments application in 2014, raising the bar for using technology in the restaurant and retail space [2]. 5 years later, the Starbucks App has become a competitive advantage for Starbucks, allowing it to tap into the increasingly important area of machine learning to drive sales and customer loyalty in the US and around the world. As the company looks to China as the core of its growth strategy for the foreseeable future [3], it will become only more important to leverage machine learning technology to stay ahead of the competition in many core functions such as customer engagement, product innovation, real estate strategy, and human resource management.

Machine Learning at Starbucks

Starbucks’s machine learning strategy today is enabled by the vast amounts of customer purchasing data collected through the mobile ordering and payment application over the past 5 years. Today, more than 50% of customer purchases are either ordered or paid for through the mobile application [4]. Each time a customer makes a purchase using the mobile application, Starbucks records valuable data about the consumer’s food and beverage preferences, average order size, purchase location, and time of day. By stitching together this data with other data such as weather data and store-level product availability, Starbucks is able to offer each customer personalized offers and recommendations, as well as “games” to receive bonus rewards (e.g., buy 2 Frappuccinos over the next 3 days) [2].

Starbucks’s current approach to machine learning appears to be successful in part because it meets the three requirements of an effective machine learning implementation [5]:

  • Right problem – Starbucks is using the data and algorithms to predict which offers will delight customers, with no view towards trying to draw causal inferences from the relationships
  • Right data – due to the success of the mobile Order and Pay application, the company has access to a wealth of relevant, proprietary data about past consumer behavior which can be used to predict preferences and future behavior
  • High tolerance for error – unlike in other industries (e.g., medical, oil & gas) where the cost of error is very high, the cost associated with the algorithm making an incorrect or irrelevant recommendation is low; the consumer may just ignore that particular offer

The future is Chinese

As Starbucks turns its focus to China as the number 1 driver of growth in the foreseeable future, the company can continue to leverage machine learning to gain a competitive edge. However, the company has to be very careful in how it implements its US machine learning strategy in a new country with totally different preferences, behaviors, and culture. Mike Yeomans, a post-doctoral fellow at Harvard University, cautions about using existing machine learning algorithms for situations that are “out-of-context” [6]. When translating the algorithm from the US to China, Starbucks may need to consider not only regularization (i.e., the relative weight of different variables) but also reevaluate the feature extraction phase of algorithm design (i.e., which variables, or “features,” the model will use in the first place).

In addition to using machine learning for customer engagement, the company can consider using the technology in other areas as it expands its presence in China:

  • Product innovation – use the company’s rich transactions data to predict and test the success of new food and beverage products. For example, if the algorithm finds Starbucks consumers like cheese, the algorithm can present various innovative cheese-related food and drink products to the new product development team to design and test in certain markets.
  • Store location – use data and analytics to predict the success of potential store locations, as the company continues to build out its retail footprint. For instance, if the average wait time for customers in a cluster of stores in Beijing is high, the company could consider building an additional Starbucks store close by.
  • Hiring partners – use algorithms to predict success of potential candidates in the hiring process, as the company attempts to massively scale its workforce in China. The company could choose to develop the capability in-house or partner with third party providers such as Fountain [7] or OnboardIQ [8].

Please weigh in on the following questions as you consider Starbucks’s use of machine learning:

  • What will be the most significant barrier to Starbucks’s China strategy, and how can they leverage machine learning to address it?
  • What other ways can Starbucks use machine learning to impact parts of its business?

 

(781 words)

[1] Starbucks Corporation. “Starbucks Fiscal 2017 Annual Report,” 2017. [Online]. https://s22.q4cdn.com/869488222/files/doc_financials/annual/2017/01/FY17-Starbucks-Form-10-K.pdf. [Accessed: 12-Nov-2018]

[2] Martin-Flickinger. “Starbucks Corporation 2017 Annual Meeting of Shareholders,” 2017. [Online]. https://seekingalpha.com/article/4057287-starbucks-corporations-sbux-ceo-howard-schultz-hosts-2017-annual-meeting-shareholders. [Accessed: 12-Nov-2018]

[3] Johnson. “FY17 Annual Letter to Shareholders,” 2017. [Online]. https://s22.q4cdn.com/869488222/files/doc_financials/annual/2017/01/FY17-Annual-Letter-to-Shareholders.pdf. [Accessed: 12-Nov-2018]

[4] Brewer. “Starbucks (SBUX) Q4 2018 Results – Earnings Call Transcript,” 2018. [Online]. https://seekingalpha.com/article/4217413-starbucks-sbux-q4-2018-results-earnings-call-transcript. [Accessed: 12-Nov-2018]

[5] Fedyk. “How to tell if machine learning can solve your business problem,” 2016. Harvard Business Review Digital Articles

[6] Yeomans. “What every manager should know about machine learning,” 2015. Harvard Business Review Digital Articles

[7] PRWeb. “Fountain Breaks into Franchise Market with Free Hiring and Onboarding Service,” 2018. PRWeb. [Online]. https://www.prweb.com/releases/fountain_breaks_into_franchise_market_with_free_hiring_and_onboarding_service/prweb15903234.htm. [Accessed: 12-Nov-2018]

[8] Lawler. “OnboardIQ raises $9.1 million to automate hiring for hourly workers,” 2017. Techcrunch. [Online]. https://techcrunch.com/2017/07/06/onboardiq-9-1m/. [Accessed: 12-Nov-2018]

[Featured Image] Starbucks Corporation. “kv-1,” 2018. [Online]. https://www.starbucks.com.cn/en/menu/. [Accessed: 12-Nov-2018]

Previous:

Detroit and the Future of the Automobile: Will Additive Manufacturing Help GM Beat Silicon Valley at its Own Game?

Next:

Machine Learning – In Theatres Now

5 thoughts on “Snowy Cheese Flavored Latte: Starbucks and Machine Learning in China

  1. Great article Mutian! China is a huge market with tons of opportunity to grab and analyze “big data”. One danger of using big data in China and then generalizing it to the whole population is that the market is so diverse. It’s possible that one product could be extremely popular in one region, but not sell in another region. Some things to consider would be differences in tastes for local cuisine etc.

  2. It’s a very articulate article! I believe Starbuck China can benefit a lot from the future machine learning technology application mentioned in the article.
    I would like to raise two concerns about the future implementation:
    1. There’s only a handful of Starbucks beverages, and most of the customers will stick to one type of their “favorite” beverage. Will the recommendation of beverages be useless in this case?
    2. The store location prediction is very useful. Nevertheless, because the Chinese city develops extremely rapid, most of the locations don’t have any previous data to analyze. If you only select location based on the data you previously had, you may lose a lot of chances to open a new store in a future CBD.

  3. Great article! I agree that applying machine learning to F&B industry has a big potential. I would like to reflect the following points.
    – Behavior patterns of Starbucks customer will be of good quality, because of the broad network of Starbucks.
    – Can Starbucks leverage customers’ online behavior through the WIFI network that it offers in store?
    – In China, how much can Starbucks freely acquire customer data and use it for global operation, from the government’s regulation perspective?

    I shall order Snowy Cheese Latte with a strawberry on top next time at Starbucks in China.

  4. Really interesting article Mutian!
    I think the largest challenge that Starbucks faces in China is understanding the different consumer behaviors. Consumers in China are fundamentally different than consumers any place else. ML will provide starbucks the necessary data to get a deep sense of consumer preferences. Then Starbucks can adjust their products and experience to satisfy Chinese behaviors better. Starbucks should also consider using machine learning in their supply chain. I would hypothesize that expansion in China comes with many challenges with distribution. ML may help them be more efficient in keep materials fresh and traveling over far distances.

  5. Great read Mutian, thank you so much! Similar to my article, one of the biggest issues to tackle with machine learning are the more personal stuff such as taste. In order to identify what a customer would like as taste, you would need to identify some attributes (ideally many) to somewhat indicate the flavors. These personal attributes (other than basic demographic ones) might be difficult for starbucks to access and analyze.

Leave a comment