by Bernard Marr
These characteristics can be matched against those of other people who have proven themselves to be successful in the specific roles being recruited.
Natural language processing also powers Unabot, which is built on Microsoft's Bot Framework.
Using company data from internal documents and company handbooks, it can process questions asked in natural human language and provide answers about the employee's role, company procedures, benefits such as pension plans and even timetables for shuttle buses to and from Unilever's campuses.
What Were The Results?
With 1.8 million applications for employment to process every year, Unilever's Chief HR Officer, Leena Nair, told me that the employee screening process had saved around 70,000 man hours of interviewing time.
As the system also generates automated feedback for applicants, even those who are unsuccessful benefit, she says.
“What I like about the process is that each and every person who applies to us gets some feedback.
“Normally when people send an application to a large company it can go into a ‘black hole’ – thank you very much for your CV, we'll get back to you – and you never hear from them again.
“All of our applicants get a couple of pages of feedback, how they did in the game, how they did in the video interviews, what characteristics they have that fit, and if they don't fit, the reason why they didn't, and what we think they should do to be successful in a future application.
“It's an example of artificial intelligence allowing us to be more human.”
Unabot is being incrementally rolled out across Unilever's global operations. Currently, it is active in 36 of the 190 countries where Unilever operates.
Nair tells me that 36% of staff have engaged with it so far, and 80% of those who do go on to become repeat users.
Users asked to rate how satisfied they are with the answers it provides currently, rate the system at 3.9 out of a maximum score of 5.
Key Challenges, Learning Points And Takeaways
Having the capability to assess applications from hundreds of thousands of people means more applicants can be considered for a role.
It also means that those who are likely to be successful are less likely to slip through the net than when the process depends on a human recruiter sifting through piles of CVs.
Human recruiters would not have the time to carry out this initial screening and analysis manually, but AI and machine learning mean that a shortlist can be quickly drawn up, no matter how many initial applications are made.
Chatbots provide hassle-free interfaces where employees and new hires can quickly get answers to common questions, and AI is used to understand which answers people are most likely to need.
Notes
1Unilever: https://www.unilever.com/about/who-we-are/about-Unilever/
2Hirevue, Unilever finds top talent faster with Hirevue assessments: https://www.hirevue.com/customers/global-talent-acquisition-unilever-case-study
3Huffington Post, High Turnover Costs Way More Than You Think: https://www.huffingtonpost.com/julie-kantor/high-turnover-costs-way-more-than-you-think_b_9197238.html
4Business Insider, Consumer-goods giant Unilever has been hiring employees using brain games and artificial intelligence – and it's a huge success: http://uk.businessinsider.com/unilever-artificial-intelligence- hiring-process-2017-6
20
Walmart: Using Artificial Intelligence To Keep Shelves Stacked And Customers Happy
With over 11,000 retail stores worldwide, Walmart is the world's biggest company by revenue1 as well as the largest private employer with around 2.3 million employees. Its online and offline retailing operations are deeply interlinked as part of the company's strategy – real-world stores double as warehouses for its e-commerce business,2 while artificial intelligence (AI) and big data initiatives initially developed for e-commerce are also put to work on the shop floor.
From pioneering customer data gathering through loyalty schemes to its latest robotic, artificially intelligent shelf-scanning robots, Walmart has ensured it has kept itself at the technological cutting edge for decades.
What Problem Is Artificial Intelligence Helping To Solve?
With so many stores, tracking inventory is a major challenge for enterprises of Walmart's scale. To remain competitive in a market where businesses sink or swim on price and customer convenience, it must consistently and accurately predict customer buying trends, where locality, weather patterns, customer demographics and economic conditions all have an impact.
Without very close to real-time sensing, accurately monitoring how products are shipped and sold is challenging. Often it will involve different inventory systems being used by different departments. Siloed data may not be available at the moment its needed, and the data itself is subject to human error due to being manually collected or updated.
A customer-centric example would be in-store apps that let shoppers locate particular items on a supermarket's shelves. These have existed for a while and, as anyone who's tried one will know, can be hit and miss. This is because many things can happen between items being stacked onto shelves and the next time it will be “sensed” – as it is beeped through the checkout – which could render the data uploaded to the app outdated.
How Is Artificial Intelligence Used In Practice?
The fine line that all big retailers tread is between keeping prices down and offering conveniences to their customers. The slightest wobble along this path can lead to losing market share to competitors.
In one particularly interesting initiative, autonomous, shelf-scanning robots have been deployed to bring real-time video analytics to the shop floor.
These robots, installed in a trial capacity initially in a small number of US stores, patrol the aisles capturing video footage of products on the shelves. This means that Walmart's data on how stock levels on the shop floor fluctuate hour to hour between replenishments is virtually real time.
As a result, more accurate models of customer behavior can be built with better predictions of what items will sell at different times of the day. This data is of course fed back to supply chain and inventory systems, which will more accurately forecast demand.
In the example given above of the customer in-store app, real-time data relayed from the robot's sensors can tell app users exactly where items are on the shelves – rather than where they should be, according to a separate database.
Crucially, at a time when the long-term impacts of automation on human workforces is starting to be considered, Walmart robots are not intended to replace humans.3 Rather, they are designed specifically to assist with very routine (boring) manual tasks. This will free up shop floor staff to spend time assisting customers.
What Technology, Tools And Data Were Used
Walmart's robots are designed by California, US-based Bossa Nova Robotics.4 They stand around two feet in height and are equipped with extendable cameras and sensors for scanning higher shelves.
Bossa Nova recently announced that the capabilities of its robots – including those used by Walmart – are set to be enhanced thanks to the acquisition of Hawxeye, which specializes in computer vision technology. Hawxeye's technology is notable because it carries out machine learning directly on a device – such as a camera – rather than first having to send data to the cloud.5 This increases speed and reduces the amount of data of no value, which has to be stored and processed by server-side systems.
The robots operate in the same manner as autonomous vehicles, and are able to recognize obstacles (such as people) in their path in real time and avoid colliding with them. They do this using cameras, which monitor their immediate surroundings.
Walmart's robots are designed to operate while stores are open by applying machine learning algorithms to safely go about their business in busy, public environments.6
In 2017, Walmart said it was working on the construction of the world's biggest private cloud, capable of processing 2.5 petabytes of
data every hour.7 It contains e-commerce and in-store transactional data, CRM records, customer feedback, social media and bought-in third party data. Data collected by the in-store robots will undoubtedly feed into this too, to help make stocking decisions around the world.
Their analytics platform is built on largely open source technologies, which gives data science teams the flexibility of picking between industry standard software solutions without having to code them themselves or buy expensive, closed proprietary solutions from third parties.
For its inventory and supply chain processes it uses tools, including Apache Spark, Cassandra and Kafka.8 These data tools are geared towards enabling real-time analysis of very large, fast-changing datasets. Insights from the data are visualized in Tableau, meaning they can be quickly understood by the humans who need to act on them.
What Were The Results?
Initial tests of the shelf-scanning robots were successful enough to warrant rolling the pilot out to a total of 50 US stores.
Customers benefit from the added convenience that there is a higher likelihood the product will be stocked and on the right shelf for them to find when they need it. Walmart benefits by reducing wasted expenditure and shelf space on items that aren't going to sell.
Bossa Nova's chief business officer, Martin Hitch, told Forbes: “We already know with an incredibly high degree of accuracy what gets shipped into the store and what gets sold through the register. Now, for the first time, we also know that we're selling so many products because we have the right number of them on display at certain points in the day.”9
Key Challenges, Learning Points And Takeaways
When you're Walmart and your competitors are the likes of Amazon and Alibaba, AI isn't a choice today – it's a basic necessity for survival.
Big retailers tread the path between minimizing costs and maximizing customer convenience. If AI and data initiatives are well planned and executed, they can sometimes do both.
In businesses of Walmart's scale, dependent on a highly complex network of moving parts, many small efficiencies can add up to major gains.
Walmart says its robots are not designed to replace human workers, and it may very well be making that statement in good faith today. But the long-term implications of automating large numbers of functions typically carried out by humans are still largely unknown.
Notes
1Walmart, Walmart 2018 Annual Report: http://s2.q4cdn.com/056532 643/files/doc_financials/2018/annual/WMT-2018_Annual-Report.pdf
2Fortune, Five Moves Walmart is Making to Compete with Amazon and Target: http://fortune.com/2017/09/27/5-moves-walmart-is-making-to-compete-with-amazon-and-target/
3Business Insider, Walmart reveals why it has robots roaming the aisles in 50 of its stores: http://uk.businessinsider.com/walmart-robots-in-50-stores-2018-3
4The Verge, Walmart is using shelf-scanning robots to audit its stores: https://www.theverge.com/2017/10/27/16556864/walmart-introduces-shelf-scanning-robots
5Venturebeat, Bossa Nova Robotics acquires Hawxeye to improve inventory object detection: https://venturebeat.com/2018/07/18/bossa-nova-robotics-acquires-hawxeye-to-improve-inventory-object-detection/
6Crunchbase: https://www.crunchbase.com/organization/bossa-nova- robotics-inc#section-overview
7Forbes, Really Big Data At Walmart: Real-Time Insights From Their 40+ Petabyte Data Cloud: https://www.forbes.com/sites/bernardmarr/2017/ 01/23/really-big-data-at-walmart-real-time-insights-from-their-40- petabyte-data-cloud/#2a7bee6b6c10
8WalmartLabs, How we build a robust analytics platform using Spark, Kafka and Cassandra: https://medium.com/walmartlabs/how-we-build-a-robust-analytics-platform-using-spark-kafka-and-cassandra-lambda-architecture-70c2d1bc8981
9Forbes, This Shelf-Scanning Robot Could Be Coming To A Store Near You: https://www.forbes.com/sites/jenniferjohnson/2018/06/29/this-shelf -scanning-robot-could-be-coming-to-a-store-near-you/#b0a32c73fb1c
Part 3
Media, Entertainment and Telecom Companies
21
The Walt Disney Company: Using Artificial Intelligence To Make Magical Memories
Disney's Magic Kingdom theme park sells itself as “the most magical place on Earth”. Since the first park opened in California in 1955, technology has played a part in bringing that magic to life.
With more than 56,000 guests arriving on the average day, each expecting a magical experience, park planners and entertainers (called “cast members”) have the task of making sure everyone goes home with only good memories.
Queues, congestion and overbooked attractions certainly aren't magical. So, The Walt Disney Company has turned towards advanced data analytics and smart technology to remove the “friction” from their theme parks.
What Problem Is Artificial Intelligence Used To Solve?
With tens of thousands of visitors each wanting to experience as many of the hundreds of attractions in each park as they possibly can on any given day, coordinating the flow of humans is a highly complex operation.
The headline attractions – the newest and hottest rides, as well as the legendary favorites – create bottlenecks. It's frustrating for the visitors if they feel they are standing around all day, and when they're in line, they aren't spending money at the ice cream stands and gift shops. Which is frustrating for Walt Disney.
On top of that, those ice cream stands, as well as hundreds of other food and drink outlets, have to be kept stocked with refreshments and optimally located to catch passing trade at the right moment.
How Is Artificial Intelligence Used In Practice?
In 2013, Disney introduced its MagicBand wristbands, which are issued to every visitor and let them book rides and attractions, access their hotel rooms, order meals at the park's restaurants and pay for purchases in gift shops.
They also give Disney detailed information about what each visitor is doing at every point in the day. This lets them offer personalized experiences – for example, allowing restaurant staff to greet them by their names when they arrive. It also gives park planners detailed aggregated datasets about overall visitor movements.
This means that planners can repurpose spots that fail to attract footfall to ease congestion around the top attractions and hotspots that cause bottlenecks throughout the day.
Because data is analyzed in real time, response can be real time too – for example, staging an impromptu character parade to draw crowds from a heavily congested area to a quieter one.1
The bands are part of a wider Disney initiative called MyMagic+, which involves removing “friction” – unnecessary stress – from visitors at every point of their Disney journey, starting from when they book their tickets online.
The system lets visitors use the My Disney Experience app to plan what attractions they want to visit, where they want to eat and which characters they want to meet. It then presents them with an automated schedule designed to minimize crowds and waiting, while doing everything on the visitor's list.
Hotel visitors don't even need to check in – the wrist band automatically tells the hotel staff that the guests have arrived, and families can head straight to their room and use the band to unlock the door.
It's also worth noting that the wider Walt Disney Company is active in artificial intelligence (AI) and machine learning research through its Disney Research initiative.2
The facilities allow the company to partner with universities on initiatives involving AI, visual computing and robotics. Their academic links give them the opportunity to develop technology solutions that could be used across the Disney empire. Innovations originating in the labs have applications in Disney theme parks, films, video games and television shows.
One recent breakthrough has been the development of a system for speeding up the rendering of graphics in its Pixar computer-generated movies. It involves using deep learning convolutional neural networks to eliminate “noise” generated during the rendering of the 3D graphics. This means that each frame can be rendered
at a lower fidelity and still result in a production-quality image.3
Disney is also working on a tool to measure real-time audience reactions in its motion theatres.4 Here, it uses deep learning algorithms that have been trained to watch an audience of hundreds of people in a darkened theatre to analyze their facial expressions to determine if they are, for example, happy, sad, bored, etc. This gives Disney real-time insights into audience engagement and could potentially be a move towards responsive experiences that change according to what people enjoy or how they react.
What Technology, Tools And Data Are Used?
Like any good magician, Disney is notoriously secretive about the technology at work behind the scenes, which is used to create the magic!
However, it is known that Disney's MagicBand wristbands use radio-frequency identification technology to communicate with thousands of sensors positioned around its parks. They also contain a radio device similar to those used in mobile phones for longer range communication.5
Rolling out the initiative cost around $1 billion and required Disney to integrate every facet of the park that visitors interact with – turnstiles, hotel room doors, rides and point-of-purchase terminals – into its data capture systems.6
This means that data points are created whenever guests use the wristband to check into rides, restaurants or shows, order food or make purchases at gift shops.
While it was being designed, the team responsible built a large-scale “demo” of the system on a disused soundstage at the theme park. Individual rooms were built and decorated to represent each stage of the customer journey, such as a living room to represent the point where the customers first book their holiday online, and a mini replica of the Haunted Mansion attraction, to represent a park attraction.7