by Bernard Marr
It isn't just simple coffees though – from a classic espresso to an iced caramel mochaccino there are actually around 87,0002 combinations available on a typical Starbucks menu!
Although the brand has long been a market leader in the United States where it was founded, in the mid-1990s it began expanding overseas into markets where it often faced strong competition from local chains. As of 2018 its biggest non-domestic market is China, where 12.4% of its branches are now located.3
The way coffee and tea – which make up the majority of Starbucks sales – are consumed varies across cultures. The coffee shop giant generates and burns through mountains of transactional and customer data in its attempt to offer personalized service to millions.
What Problems Is Artificial Intelligence Helping To Solve?
With so many stores offering so many products in so many parts of the world, precise calculations must be made to keep the thousands of outlets stocked and ready to serve their customers. Slight miscalculations can mean large overspends on logistics such as transport and storage. With businesses as large and broadly distributed as Starbucks, small inefficiencies very quickly add up to large ones.
Operating internationally, Starbucks is often competing with local chains and brands that are already synchronized with the lives of the customers they serve. Starbucks may offer US-style customer service to markets that have traditionally taken a more relaxed approach to serving hot drinks. But it has to adapt to local cultural norms and fit with local customers’ habits if it wants to become a part of their everyday lives.
How Is Artificial Intelligence Used In Practice?
Starbucks gathers data on its customers’ behavior by tracking them (with their permission) through its loyalty programs and mobile apps.
Customers can use the app to pay for food and drinks in advance before collecting them at the counter, and as of 2017 more than 17 million people were using it.4
This information is correlated with other internal and external data, including meteorological data, local data and company data such as inventory levels to help Starbucks understand what is driving sales.
This means that personalized promotions can be offered – offering deals on items that each customer is likely to be interested in directly to them.
The artificial intelligence (AI)-enabled system that it uses to do all of this is called the Digital Flywheel Program.5 Its job is to take into account every factor from locality, to time of day, to weather – to predict what customers will order when they walk through the door or load up the app on their phone.
AI will also play a key part in Starbucks’ move to branch out into offering delivery in China.
It has partnered with Alibaba, and will use technology developed by the retail tech giant's recent acquisition Ele.me,6 a food delivery service that is heavily built on smart technology. In 2017, Ele.me unveiled its autonomous food delivery robot,7 which uses machine learning to navigate as it distributes beverages and snacks. Designed to operate autonomously in large office buildings, perhaps it won't be long before we can have coffee delivered directly to our desks by robots?
Choosing to partner with Alibaba rather than develop its own delivery service in China could be a shrewd move. Alibaba (and subsidiary Ele.me) has a sophisticated delivery network in place and AI to drive efficiency. Developing these from scratch would be very expensive for Starbucks.
What Key Tools, Technology And Data Were Used?
Starbucks announced that AI would be built into its Digital Flywheel data analytics program in 2017.
Much of the data it feeds into its analytics comes from processing 90 million transactions every week through its stores and apps.8
This tells them everything they need to know about who is buying what, where and when. The data can then be correlated with individual customers’ data gathered through loyalty programs or app use.
The app itself features AI in the form of the virtual barista feature.9 Like other virtual assistants, this one uses natural language processing to understand the nuanced way humans talk. This one is specifically trained to adapt to the complex and evolving languages we use when ordering drinks at coffee chains.
What Were The Results?
Through better understanding its customers habits, Starbucks is able to build brand loyalty by offering the right products at the right time, and personalized promotional offers.
This means it can tailor its product range and marketing strategies towards individual markets, using localized datasets that are highly likely to be relevant.
Starbucks has said that by 2019, 80% of its stores around the world will have access to the Digital Flywheel.10
Key Challenges, Learning Points And Takeaways
Working worldwide across an enormous number of markets makes getting a thorough overview of your customer base challenging, but today's machine learning technology means it can be done.
Starbucks offers customers conveniences such as being able to order ahead and skip queues, in exchange for data it can use to improve services.
Just as in other areas of retail, food and drink outlets are transforming in an attempt to offer online convenience. This means interactivity between mobile phones and in-store systems, and in-app purchasing.
Partnerships are often a huge benefit when branching out in new directions. Today, partnering with tech specialists means you can share their data, analytics technology, or increasingly both, on an as-a-service basis.
Notes
1Starbucks, How many Starbucks stores are out there?: https://www. loxcel.com/sbux-faq.html
2Favrify, 18 Exotic Starbucks Drinks That You Didn't Know Existed…: https://www.favrify.com/starbucks-drinks/
3Starbucks, How many Starbucks stores are out there?: https://www. loxcel.com/sbux-faq.html
4Cio, Starbucks’ CTO brews personalized experiences: https://www.cio. com/article/3050920/analytics/starbucks-cto-brews-personalized- experiences.html
5Zacks, Starbucks’ Digital Flywheel Program Will Use Artificial Intelligence: https://www.zacks.com/stock/news/270022/starbucks-digital-flywheel-program-will-use-artificial-intelligence
6The Star, Starbucks partners with Alibaba, as it tries to keep its coffee throne in China: https://www.thestar.com/business/2018/08/02/ starbucks-partners-with-alibaba-as-it-tries-to-keep-its-coffee-throne- in-china.html
7Pandaily, Ele.me Delivery Robot Completed Takeout Delivery for the First Time: https://pandaily.com/ele-me-delivery-robot-completed-takeout-delivery-for-the-first-time/
8Cio, Starbucks’ CTO brews personalized experiences: https://www.cio. com/article/3050920/analytics/starbucks-cto-brews-personalized- experiences.html
9Starbucks, Starbucks debuts voice ordering: https://news.starbucks. com/press-releases/starbucks-debuts-voice-ordering
10Zdnet, Starbucks to step up rollout of “digital flywheel” strategy: https://www.zdnet.com/article/starbucks-to-step-up-rollout-of-digital-flywheel-strategy/
18
Stitch Fix: Combining The Power Of Artificial Intelligence And Humans To Disrupt Fashion Retail
Stitch Fix, founded in 2011 and based in California, USA, aims to revolutionize fashion retail by acting as a personal stylist, automatically shipping items that it thinks the customer will want to wear.
It does this by asking customers to fill in a survey stating their style preferences, budget and optionally giving the company's stylists access to their social media accounts.
The work of the stylists is augmented by data scientists and artificial intelligence (AI), which aim to provide customers with clothes they will want to wear by analyzing their preferences and comparing them with thousands of other customers who fit their profile.
What Problems Is Artificial Intelligence Helping To Solve?
The proportion of our shopping that we do online is continuing to soar – in the United Kingdom it climbed from 11.6% of total (non-food) retail spending in 2013 to 24.4 in 2017.1
Fashion retailers are uniquely
challenged by the relatively high rate of customer returns. They are also obliged to offer both free delivery and returns to compete. This can often add up to a large expense for businesses if their customers order large quantities of clothing to assess at home and end up returning most items.
As well as the cost of shipping and processing returns, this situation can make it difficult for retailers to manage their inventories, and often requires them to be overstocked to be able to fulfil customer demand. In the fashion industry, it is normal for large quantities of clothing to be sold at greatly reduced prices, or even destroyed due to inefficient forecasting of demand.2
This is all hugely wasteful and, of course, eats into profits. Providing clothes to customers that will fit and meet their quality expectations, therefore minimizing returns, is a key challenge for all online fashion retailers.
How Is Artificial Intelligence Used In Practice?
Stitch Fix uses AI to understand its customers’ body measurements as well as their tastes and style preferences.
These algorithms all work to augment the work of human stylists. As well as retailing, Stitch Fix designs clothes, and the concepts for new items are informed by insights into what is popular, derived from the AI analysis.
Their chief algorithm officer, Eric Colson, said: “Our business is getting relevant things into the hands of our customers. This is the one thing in the world we're going to be best at. We couldn't do this with machines alone. We couldn't do with humans alone. We're just trying to get them to combine their powers.”
What Technology, Tools And Data Were Used?
A team of around 85 data scientists work with the Stitch Fix AI platform to select items for which there is a high probability that customers will want to keep the items.3
Colson brought machine learning with him from Netflix, where he was previously employed as vice president of data science and engineering. This technology allowed the business to greatly increase the efficiency of its algorithms that were already being used to filter out items when they would be unsuitable for a particular customer.
When signing up for an account, prospective customers are asked for their measurements, weight, style preferences (such as slim or baggy fit), colour preferences, budget, personality questions such as how adventurous their selection should be, and for specific details such as whether they often find shirts or jeans too tight or too loose.
If you give it permission, it will also take into account what it can learn about your style and preferences through social media. It will also take data it collects from every point of feedback, such as when customers fill in forms to give details about why they are returning items.
As well as matching products to clients, Stitch Fix has specific algorithms for assigning personal stylists to clients, making inventory decisions, analyzing images posted to social media (Pinterest) by clients and assessing how happy they are with the service they are receiving.4
What Were The Results?
The money we spend on online retail is quickly catching up with what we spend in bricks-and-mortar stores. Online retail provides unique customer service challenges, and AI has the potential to offer a plethora of solutions.
Understanding customer requirements and preferences has meant that Stitch Fix is able to automatically despatch items that, according to its data, its customers are more likely to love. This allows it to avoid wasted warehouse space, shipping costs, return expenses and end-of-season overstock.
Stitch Fix says that its adoption of machine learning has allowed it to increase revenue and customer satisfaction, while decreasing overall costs.5
Key Challenges, Learning Points And Takeaways
The better understanding that AI can give you of your customer, the less chance there is that you will disappoint them with your products and services.
AI poses a real risk to human jobs, just as all other industrial revolutions before it did. Designing intelligent systems that augment the capabilities of human workers rather than make them redundant is a key challenge across all industries. Stitch Fix's algorithms inform the work of human stylists and analysts, who have the final say. Which is probably just as well – remembering it was the weavers who were among the first to take up arms during the first industrial revolution!
Notes
1Financial Times, Online retail sales continue to soar: https://www. ft.com/content/a8f5c780-f46d-11e7-a4c9-bbdefa4f210b
2Fashion United, The fashion industry at a dead end: new products worth millions destroyed: https://fashionunited.uk/news/business/the-fashion- industry-at-a-dead-end-new-products-worth-millions-destroyed/ 2018071930847
3ZD Net, How Stitch Fix uses machine learning to master the science of styling: https://www.zdnet.com/article/how-stitch-fix-uses-machine-learning-to-master-the-science-of-styling/
4ComputerWorld, At Stitch Fix, data scientists and A.I. become personal stylists: https://www.computerworld.com/article/3067264/artificial- intelligence/at-stitch-fix-data-scientists-and-ai-become-personal-stylists .html
5ZD Net, How Stitch Fix uses machine learning to master the science of styling: https://www.zdnet.com/article/how-stitch-fix-uses-machine-learning-to-master-the-science-of-styling/
19
Unilever: Using Artificial Intelligence To Streamline Recruiting And Onboarding
International consumer goods manufacturer Unilever sells over 400 branded products in 190 countries. Worldwide, it has over 160,0001 people, making it one of the world's largest employers.
With any company, people are the most valuable resource. To make sure they are enticing the right talent, Unilever deploys artificial intelligence (AI) solutions aimed at attracting, analyzing and ultimately selecting the best people to fit the thousands of roles it needs to fill each year.
What Problem Is Artificial Intelligence Helping To Solve?
Any recruiting process involves risk. Advertising for talent, screening applicants and onboarding new hires is an expensive and time-consuming process. It has to be done properly though, as hiring the wrong people can have expensive consequences and a damaging impact on business.
Recruiters have limited time available to them to search for the right candidates and once they've come up with a shortlist there is a narrow window of opportunity to make a decision on whether they are a good fit for the role.
In Unilever's case, when recruiting for their Future Leadership program, the company knew it had between four and six months to reduce a pool of 250,000 applicants from all around the word to fill 800 available positions.2
The costs don't end once the right person has been found for the job – according to the Society for Human Resource Management, the cost of training a new hire averages out at between six and nine months’ wages for the post in question.3
How Is Artificial Intelligence Used In Practice?
Unilever partnered with AI recruitment specialists to roll out a global initiative aimed at efficiently matching applicants to posts.
It involved developing a multistage process, which starts with asking applicants from anywhere in the world to submit an online CV or LinkedIn profile.
From there, applicants are asked to take part in 12 different online games. Developed by Pymetrics, the games are designed to test aptitude in a number of different areas relevant to the roles they are applying for.4
The games are not necessarily designed to be “won” or “lost” but rather to provide a measure of a candidate's characteristics, and ideal outcomes may be different depending on the role the candidate is applying for.
For example, one game involves inflating balloons to assess a candidate's appetite for risk, using a “stick” or “twist” gameplay mechanic similar to blackjack. Candidates are awarded points for pumping more air into virtual balloons, and must try to stop pumping before the balloon bursts.
The next stage of the process involves submitting a video inter-view.
As with the games, this can be completed in the candidates’ own time using just
their smartphones or a computer equipped with a webcam.
Here, AI algorithms analyze the language, facial expressions and body language to determine whether they are likely to fit the profile of someone who will be successful in the role.
From this, a final shortlist of 3,500 applicants was pulled together, who were all invited to assessment centers where they met Unilever recruiters in person for the first time, and the final selection of 800 was made.
Once hired, Unilever's new recruits have access to Unabot, an AI-powered chatbot designed to speed up the onboarding process by answering questions through a natural language chat interface.
What Technology, Tools And Data Were Used?
Pymetric's games allow detailed aptitude profiles to be built up, evaluating candidates’ strengths and weaknesses in a more quantitative way than a traditional face-to-face job interview process.
These profiles can then be measured against values that the machine learning algorithms pick out as being likely to signify suitable applicants.
Unilever then uses facial image analytics technology developed by HireVue to interpret the data collected through its pre-recorded video interview process.
Computer vision and natural language processing technology are used to analyze the videos to capture data points that can be autonomously labelled to give readings indicative of characteristics such as “sense of purpose”, “systemic thinking”, “resilience” or “business acumen”.