However, what if we look at the other side and see how machine learning (ML) affects more familiar industries, such as retail. The progress there is also moving relentlessly, you know. With machine learning, retailers can better analyze and predict buying behavior. After all, today's customers need maximum shopping convenience and the ability to shop through different channels. Consumer preferences are constantly changing, but AI and machine learning in retail help to anticipate and respond to these changes in a timely manner. As soon as it becomes a must-have, we offer you to get acquainted first with the impact that this technology has on the industry, and then with 5 interesting examples of machine learning implementation in retail.
Why does retail need machine learning?
The ability to predict is the basis for retailers' survival. With AI and automated machine learning tools, you can predict how many goods will be needed on a given day in response to retail customers’ demand, saving money and time. Machine learning and AI can also be of great help in planning for optimizing purchases, inventory, and sales. This way you’ll know whether the current assortment is correct regarding retail customers’ demand, whether prices are set correctly, which stores need to organize the supply of certain goods, taking into account regional and other characteristics.
According to Juniper Research, retailers' annual spending on artificial intelligence will increase by 230% - from $3.6 billion in 2020 to $12 billion in 2023. The main direction of machine learning implementation in retail will be the use of machine learning tools for demand forecasting. Revenues of products and services providers in this industry will reach $3 billion by the end of the "reporting period" (an increase of 290% compared to $760 million in 2019).
In retail customer service, Juniper sees great potential in chatbots that work with customers in the salesrooms. By 2023, bots helping shoppers navigate the store will yield annual savings of $439 million for retailers ($7 million in 2019). They will make 22 billion contacts with visitors and will help make $112 billion worth of purchases. Sounds pretty impressive, isn’t it?
To sum it up, what can machine learning offer to retailers?
Retailers should consider different aspects of individual department management in geographically spread areas to ensure a constant supply flow with lower costs and minimized losses. The challenges faced by the machine learning platform are as follows:
- Inventory and supply chain management with assortment planning;
- Customer model analysis showing invalid/scam requests;
- Customer interaction analysis using virtual assistants and chatbots;
- Execution of retail analytics on a scale to understand annual growth;
- Personal recommendations utilizing collaborative filtration, content filtration, hybrid filtering, etc;
- Detection of item shortfalls in stores;
- Interpretation of text and images from invoices, packing lists, bills, etc.
Here come 5 great examples of machine learning in retail
In early 2018, the world's e-commerce giant Amazon opened its first staff-less store Amazon Go for public use (previously, for two years, the store without cashiers and staff was available only to company employees). It is not surprising that Amazon decided to try out the latest AI developments not only on the automated payment system.
All sales departments of Amazon Go are equipped with high-tech cameras with automatic object identification RFID (Radio Frequency Identification). Usually, such a system is used in unmanned electric cars to track the behavior of passengers in the cabin and automatic processing of visual information by a computer. But Amazon Go cameras went further and used to monitor retail customers’ behavior from the moment they enter the store until they pay for their purchases. May sound a little bit creepy, but effective nonetheless.
There are square-shaped cameras all over the ceiling of the store which keep track of customers. The main purpose of the cameras is to determine which items are on the highest demand, which products are most often returned to the shelves by the retail customers, etc.
The cameras are used to monitor the buyer behavior from the moment they enter the store up to the moment of payment. Additionally, Amazon Go cameras recognize faces and determine the height, weight, skin color and other physical characteristics of retail customers. Subsequently, the AI connected to the video system, based on all the data obtained, not only determines the most popular products for specific retail customer groups, but also offers options for changing pricing policy. All this work is automated by a computer without any human intervention.
The cameras are also connected to the store's automated warehouse system and shelves equipped with ‘Sensor fusion’ sensors. In case it is impossible to detect the goods taken by the customer, the camera finds it in the warehouse system and coordinates with the weight and movement sensors, which are located on each shelf.
Let's say the buyer took the milk carton and began to read the composition of the product, but suddenly saw a familiar brand nearby and returned the package to its place. Even during these few seconds of making the choice in Amazon Go will be tracked by the camera, sensor and inventory system, and the machine learning algorithms will draw the appropriate conclusions.
With AWS' impressive revenue growth of 37 percent in Q2 2019 to $8.38 billion, the company clearly intends to consolidate its rapidly growing machine learning capabilities.
And here is some inspiring promo from Amazon themselves.
Cosmetics retailer Sephora is pushing the envelope for innovation in e-commerce by creating a machine learning application that helps retail customers identify certain shades by simply uploading a photo. The platform, which is the product of a partnership with ModiFace, a face analysis and visualization technology company, is likely to have profound implications beyond the beauty industry. This technology will be an organic complement to Sephora's online shopping experience and will stimulate transactions, allowing the customer to visualize the benefits of the product after the transaction.
"We have been working with Sephora and other cosmetics brands for almost a decade now," said Parham Aarabi, CEO of ModiFace. "The problem of unveiling the product was a major challenge that can be partly addressed with augmented reality testing. Yet using artificial intelligence to pick up shades and suggest products before trying them is a key step. We have been working on this for almost 5 years, but we felt that the technology had finally become accurate enough for a large-scale deployment."
The app utilizes ModiFace face recognition and visualization technology, which enables Sephora clients to post photos to Facebook Messenger while chatting with Sephora Visual Artist bot. The technology then detects the most compatible shade automatically and suggests current products in Sephora's stock using the Artist Intelligence mechanism. The system renders a photograph of a user wearing a Sephora product visualizing the make-up, giving them an idea of what they will look like without having to depend on the consumer's imagination.
"We found that people are using this technology for intelligence," Aarabi added. "For example, they have a dress that they want to match the exact same shade of lipstick to, or they find a product in the store that they want to match to see similar shades from other brands. It's quite versatile in usefulness, and the amount of use and level of involvement definitely indicate this." The results are already here, in the first quarter of 2017, the Sephora digital conversion enabled LVMH's parent company to increase revenue by 11 percent. Even during 2020’s financial crisis Sephora is thriving, as recently it announced it would open 100 stores.
In 2018, the Swedish H&M division began using machine learning to select a range of stores. This way the company hopes to return retail chain customers to compensate for the worst sales in its history: in the 71-year history of H&M sales have been falling for 10 consecutive quarters.
Before that, the assortment of H&M stores was formed by designers, writes The Wall Street Journal. H&M is the last clothing retailer to embrace technology to win the clients. For example, Inditex (Zara, Bershka, Massimo Dutti, etc.) already use robots to make it easier for retail customers to select online orders in stores, and Gap turns to Google Analytics data to monitor consumer preferences.
Analysts are skeptical about H&M strategy. Yet, the story of one H&M store in a wealthy district of Stockholm reveals that machine learning can really help. In this store, attention was paid to products for the whole family because the managers were oriented towards local residents. However, the analysis showed that the majority of customers were women and fashionable goods, like flower skirts, sold surprisingly well along with expensive goods. H&M reviewed the store's assortment and sales grew, although so far, the company refused to disclose how much. The algorithms operate 24/7 and are adjusted to continuously shifting client preferences and expectations.
.4 Apotek Hjärtat
Apotek Hjärtat is Sweden's leading chain of private pharmacies with about 390 drugstores and over 3000 people working in it. They own pharmacies all over Sweden - in big cities, in rural areas as well as in sparsely populated areas. Throughout the AI usage, the company has improved its strategy by providing more accurate pricing compared to competitors in both online and offline stores.
Apotek Hjärtat have chosen Revionics Competitive Insights AI development company for their collaboration and developing a strategy to optimize prices based on machine learning in early 2017. They cleverly used these opportunities to better track competitive prices and offers, while also enhancing their flexibility to respond to changes in buyers and competitors behavior. Their subsidiary Rimi Baltic has chosen Revionics Price Optimization to improve their interaction with customers and influence their business.
"We are pleased to further expand our productive partnership with Revionics by using a data-driven approach to analyse the effectiveness of our advertising," said Anders Nyberg, Managing Director of Apotek Hjärtat. "This is critical to our ability to provide insight, structure and analysis to support strategic decisions and ensure a high return on investment for our company". By adopting a science-based approach to pricing to gain retail customers confidence, Apotek Hjärtat is now able to move on to analysis of promotional campaigns to focus on those that generate more interest. Revionics allows them to detect targeted promotions that take into account seasonality and attract customers, while avoiding disregard for other profitable products, including private brands, while improving business efficiency.
In addition, Hjärtat used artificial intelligence technology to promote products that help people quit smoking. In a square in Stockholm, the company installed a screen with built-in smoke detectors. Every time a smoker shows up in the recognition area, a young man on the screen starts coughing. The effectiveness of advertising has been controversial, but it is clear that the use of different sensors is the future of Digital Signage (a technology for presenting information from electronic media installed in public places).
At Costco retail company machine learning is used to maintain the productivity and sustainability of its fresh food department. Costco donated all of its unsold or damaged products, and therefore produced more than needed fresh food.
They have worked with SAP company to tackle this issue using a demand forecasting algorithm that assists managers in making sure the right quantity of fresh products is always in stock. Costco's bakery managers must be able to predict the demand for each menu item they need to produce on a daily basis. Prior to the SAP solution, these managers had to establish a production plan on paper by reviewing sales and trend reports. They also had to review local events and the past background of their colleagues. Costco calls it "tribal knowledge". The plans were updated daily on the basis of previous day's leftovers, damaged or destroyed products, before the baking teams could start their work in the morning-time. The development of this solution was led by SAP AppHaus, a branch of SAP solutions.
When Costco was ready to start work on the new solution, they invited Jeff Lyons, senior vice president of Fresh Foods, and several managers and supervisors from their bakeries. These bread experts at Costco are end users of the solution, so AppHaus asked their feedback on when customers purchase specific foods and how frequently they do it. They interviewed each employee extensively and watched some of them during their work. When SAP had enough information about the bakery manager's daily tasks, they managed to create a new "bakery of the future" with Costco employees. It is a tablet application that displays data and ideas for bakery managers and digitizes manual processes. The application uses machine learning to provide a planning forecast for each item on the bakery menu. This forecast determines how much of each item is to be baked and is automatically adjusted for the passing residue and damaged items.