Amazon's Flywheel Approach: How The Tech Giant Uses AI

Updated: 

April 22, 2022

downloadable version of blogdownloadable version of blog

The Glossary of Amazon Listing Errors

Want to save this for later? Download a PDF of this blog for easy reference later on.

go to top

Amazon's Flywheel Approach: How The Tech Giant Uses AI

Unless you've been living under a rock, you know how Amazon has been an early adopter of Artificial Intelligence (AI) and Machine Learning (ML). It is a company that has reorganized and restructured itself to benefit from these technologies in multiple areas.

From its warehouses full of products to your Echo smart speakers, Amazon has been utilizing AI and ML across its many divisions to drive internal processes efficiently and boost customer experience. It is a classic example of successful digital transformation.

Wait, what is digital transformation?

It is essentially using modern digital tools and technologies to transform traditional and non-digital methods or create more efficient ones to match the evolving market and customer expectations. And that is precisely what Amazon has done with its "flywheel" approach to AI!

This article dissects how Amazon continues to digitally transform itself with AI and the areas where it has made considerable progress in the past decade.

The meaning behind "flywheel" and how Amazon uses it

"Flywheel" is an engineering concept that describes the way businesses can conserve energy and keep up the momentum. The flywheel maintains a constant flow of energy, spreading it to other areas of a machine.

Let us take the potter's wheel as an example. To kick-start it, it needs an initial burst of momentum, or the wheel will never move. You can keep pressing on the pedal and run the wheel consistently by adding a bit of energy at a time.

source

Similarly, using AI takes a lot of energy to get started. However, once the wheel starts turning, it is much easier to keep it going by continuously giving the wheel small boosts.

By leveraging AI's momentum, Amazon ensures that the data is shared across the organization and not stacked in one department. Therefore, AI innovations in one department or team can be transferred to other areas for business growth.

For instance, if a customer visits the Amazon Go store to get a few groceries for dinner and asks Alexa to look up a recipe, its AI model can derive that the customer likely needs to purchase a specific type of saucepan.

Five areas where Amazon uses the flywheel approach for AI integration

Amazon is dedicated to offering the best possible customer experience. "Start with the customer and work backward" is its motto.

When the founder of Amazon, Jeff Bezos, was serving as the CEO, he would famously leave an empty seat for the customer at every meeting. This may sound cheesy, but it was a way for the company to include or think about the customer in every discussion.

Even after Bezos' departure, senior-level employees continue reserving a seat in pivotal meetings. That is how dedicated Amazon is to customer experience. In the following section, we will discuss five areas where the company uses the flywheel approach for AI integration:

1. Chatbots

Research shows that chatbots will power 95% of all customer support interactions by 2025. This is not surprising since 62% of customers are open to using AI for improving their shopping experience, and 88% expect companies to accelerate their digital initiatives.

Naturally, Amazon has made great strides in utilizing chatbot technology. Simply put, it uses a template ranker, where an AI model controls the chatbot's vocabulary by choosing among hand-authored response templates.

These templates are, in fact, general forms of sentences with variables for product names, delivery timelines, dates, prices, and so on.

Let us take the example of order cancellations. In this scenario, the AI model receives the dialogue context and information about the customer's account profile.

On receiving a customer response as input, the template ranker applies an attention mechanism, a technique for utilizing the most relevant parts of the input sequence flexibly. This helps determine words in the response particularly useful for ranking it.

source

The AI model incorporates new templates with little additional work because of its pre-training on a data set of interactions between the customer care reps and the customers themselves. 

Given how chatbots enable two-way communication with the customer, it is evident your eCommerce business can also benefit from using AI-driven chatbots in many ways:

1. You can record real-time interactions with online store visitors and improve your services as per their input.

2. By deploying a shopping bot on your storefront and your official Facebook Messenger, Telegram, and WhatsApp accounts, you enable potential customers to easily browse through your catalog and make purchases directly on the platform of their choice.

3. A chatbot can instantly tackle FAQs and guide store visitors to relevant pages on your website and boost engagement.

Sephora’s Kik bot, for instance, shares makeup suggestions by asking questions (on preferred makeup brands, occasion, age) to gather information and then redirects the user to the Sephora app or mobile site for making the purchase.

2. Product recommendations

Since its earliest days, Amazon has applied AI to derive product recommendations based on what customers already said they liked. It is by far the most sophisticated element of the company's eCommerce efforts.

You will find product recommendations across every page, channel, and device of Amazon. Research shows that 45 different recommendation widgets are visible on the app homepage alone, representing Amazon's devotion to promoting product discovery.

Over time, the algorithms have undergone a massive transformation and become more dynamic. Many variables come into play in product recommendation algorithms, from location and recent purchases to saved items/lists and user reviews to purchases other customers have made after viewing similar items.

The technology helps Amazon fully understand the context and intent behind customer search queries so that they know why visitors are searching for specific products and accordingly puts together a list that will drive higher conversions. No wonder today, Amazon's product recommendation engines are driving 35% of total sales.

To drive higher sales, you have to bank on product recommendations to engage and convert your eCommerce store visitors. Research shows that 75% of today’s consumers expect a personalized online shopping experience. Here are a few ways you can incorporate the technology like Amazon:

1. Showcase your best-selling products on every category page to offer social proof about what similar shoppers like and what the visitor might also like.

2. Have a classic ‘recommended for you’ section featuring product suggestions - on the homepage - based on what the visitor has previously wishlisted or bought from you.

3. Display the ‘recently viewed’ items that other online shoppers checked out. This can introduce your visitors to new products that they may not actively search for or know of.

There are a few AI-driven software you can use for deploying the technology. Dynamic Yield, for instance, recommends products across your eCommerce store — be it the homepage or product page. Similarly, Adoric is apt for recommending “most viewed,” best-selling, and “recently viewed” products.

3. Alexa-based voice shopping

Amazon is one of the first companies to foray into ML with the AI bot Alexa. The voice-powered virtual assistant, for instance, allows customers to find and purchase products on mobile and walk through the checkout with voice prompts instead of clicking or tapping on the screen.

When items are added to the shopping list with Alexa's help, the customer can access a text version of the list in their Alexa app and make changes later. Overall, the voice-enabled technology helps them have a hands-free user experience.

Similarly, you can get recommendations from Alexa. Amazon Echo, an Alexa-enabled smart speaker, uses built-in technology for matching what a customer speaks to the acoustic patterns of the wake word. It then sends the request to Amazon's secure cloud, where the wake word is verified and processed. After confirmation, an answer to the request is sent to the customer.

source

For instance, if you say, "Alexa, play No Time To Die by Billie Eilish," it will use the recording of the request and information from Amazon Music to play the song for you.

It is amazing to see how such technology allows consumers to search and buy products online with simple voice commands. You can get Alexa Skills, Cortana Skills, or Google Actions, easy-to-download apps for a voice assistant. 

For instance, grocery e-retailer Ocado uses Amazon Skill. You can simply say, “Alexa, ask Ocado to add tomatoes to my cart.” In this way, consumers can shop from other eCommerce businesses and not just Amazon via Alexa.

4. Product forecasting

Amazon sells 4,000 items every minute and caters to over 185 countries. However, the large volume of products makes it cost-prohibitive for maintaining surplus product inventory levels. 

Although historical buying patterns can be studied to decide on product restocking for household staples like trash bags or laundry detergent, most items exhibit variability in demand due to external factors — beyond Amazon's control.

For instance, the sudden rise in sweatsuit buying in early 2020 was hard to predict as no one foresaw the consequences of the COVID-19 pandemic.

Today, Amazon has progressed in fields such as image recognition, deep learning, and natural language processing for designing forecasting models that help make accurate decisions across various product categories.

Product forecasting is complicated, but when done right can fetch your eCommerce business useful information helpful in making the right decisions concerning product procurement, pricing, and market potential. Here is how you can do it better:

1. Retail AI software such as Vue.ai, Symphony RetailAI, and Peak AI take demand forecasting for products to newer levels of accuracy. Using Machine Learning, these tools analyze trends taking variables like location, seasonality, types of products, and competition in mind. With ML, accurately forecast demand by-SKU/by-store even without sales history.

2. Similarly, product images can help in demand forecasting. Measure demand for products, including those you have not even manufactured with 3D configuration platforms such as Threekit and Vendavo. For instance, if your green sofa is under production, you can still get a sense of initial customer interest by launching a dedicated page for that sofa.

5. Warehouse and delivery optimization

Amazon workers in fulfillment centers can skip manual item scanning thanks to AI. It allows them to store items that have arrived from manufacturers and distributors anywhere on a warehouse's shelves and record their location on a computer.

The workers can grab an item out of a box using a barcode scanner to scan it, place it on the shelf and then scan the shelf itself. The dual scanning process associates the item with its location. Besides, Amazon's fulfillment centers are automated with smart robots.

These facilities are highly structured — laid out more like a grid. The configuration enables computer vision systems to locate each order efficiently.

smart retail concept, robot service use for check the data of or Stores that stock goods on shelves with easily-viewed barcode and prices or photo compared against an idealized representation of store

source

Robots that carry out the orders are AI-powered, and algorithms help them learn details like where the order is, where it is supposed to be and how to fulfill it in the fastest way possible. All this happens in real-time.

Expansive spreadsheets remain the first choice for many retailers for running supply chains. However, this practice leads to unstructured and disconnected datasets, resulting in higher operational expenses.

Amazon has learned to leverage AI for redesigning supply and making sense of data, automating the process of predicting customer demand, optimizing delivery routes, and personalizing customer communications — all the while monitoring the whole supply chain.

The Internet of Things (IoT) has immensely transformed Supply Chain Management (SCM) by enabling machines to connect and interact with products and customers.

1. Smart glasses, for instance, enable warehouse workers to see text and images about a product in the warehouse. You can scan barcodes and use voice commands to confirm the activities on the go and aim at a handheld scanner for automatically entering critical information — thus making your job easier.

2. The vision picking system documents product lots on its own, completely removing the need for doing any laborious paperwork.

3. Employing collaborative bots (such as Mobile Rack GTP AMR robots) accelerates how a warehouse functions as the bots are faster than an average human being.

4. Simplify tracking your physical inventory at the warehouse with the help of RFID tags. Keep an eye on when a product expires to ensure the right items move to the end of the supply chain before they expire.

Over to you

When you think of online shopping, Amazon is the first name that comes to mind and you are likely to at least visit the platform once to browse products. As the e-retail giant continues developing fresh product categories, innovating new technologies, and driving new acquisitions, it is safe to conclude that companies of all sizes can learn a lot about how Amazon has transformed itself with the help of AI and ML.

Need Help with Multichannel?

Schedule time with a Zentail expert to see how we can transform your Ecommerce business.

More Blogs in

No items found.