A peek at selective new customers benefiting from Microsoft AI and Machine Learning.
Cami Boosts Customer Engagement at Dixons Carphone
Based in the United Kingdom, Dixons Carphone is a major electronics retailer employing over 42,000 people in 11 countries, and providing consumers with products and services that help them lead seamlessly connected lives at home, in the office, and on the move. Like most retailers, Dixons Carphone has had to adapt its business to changing consumer buying patterns, including an increase in online product research and shopping. For instance, 90 percent of their customers start their shopping journey online, and a full 65 percent use their phones to assist them while shopping in-store. This resulted in a comprehensive review of their existing customer experience strategies.
Dixons Carphone partnered with Microsoft to research ways to transform their business through better customer engagement. They also wanted to and provide their store colleagues with tools to optimize the time spent with customers. They determined that artificial intelligence could be the key to offering customers a differentiated and personalized service. Specifically, Dixons Carphone investigated the capabilities of the Microsoft Bot Framework and Microsoft Cognitive Services in the context of customer interactions. The Bot Framework helps companies build, test and deploy intelligent bots that are capable of interacting with customers in a conversational way. In the case of a retail business, bots can be programmed to answer questions about product options and stock availability, for instance. The Bot Framework works in tandem with Cognitive Services, a collection of intelligent APIs hosted on Azure that provide the underlying language and image recognition capabilities that power the bots.
Dixons Carphone spent a fair bit of time deciding on the branding and personality for their bot, and finally landed on the name Cami – a mildly geeky and confident personality, designed to help customers navigate the world of technology. Cami currently accepts text-based input in the form of questions and accepts pictures of products’ in-store shelf labels to check stock status. The bot uses the Cognitive Services Language Understanding Intelligent Service (LUIS) for conversational abilities, and the Computer Vision API to process images. Dixons Carphone was able to simply reuse existing information available on their online buying guide and from their store employee training materials to train Cami.
Cami is not just for customers – she will be assisting Dixons Carphone store colleagues as well, for instance to do stock checks the same way customers do. In addition, the company’s research showed that customers who research products online before coming to a store are often frustrated that they need to start from scratch when getting assistance from a store colleague, and Cami helps bridge that gap through a Wishlist feature. As customers add items to their Wishlist, Cami saves the search criteria they used, and store colleagues can pull up that information in-store, to see what the customer was looking for. They can then direct them to that specific product or related products which may suit their needs.
Dixons Carphone has been testing Cami before rolling out the functionality to customers, and the results have been positive. Once Cami goes live, Dixons Carphone expects the bot to be a new source of business intelligence. The company will use the Cognitive Services Text Analytics API, Azure Application Insights, and a Power BI dashboard to review which products customers are looking at, the sentiment of their interactions, and the questions they are asking. Understanding the questions that customers are asking and analyzing their interactions with the bot will help the company improve their communications and messaging as well.
You can learn more about the Dixons Carphone story here.
Arvato Bertelsmann Protects Online Merchants from E-Commerce Fraud
An estimated 70 percent of online sellers in Germany have suffered fraud attempts, but only 14 percent of them use any safeguards today. Although e-commerce merchants recognize the importance of protecting themselves from online fraud, many of them lack the resources to manage this risk efficiently. What’s more, criminals are quick to adapt their fraudulent methods, so any solutions used by merchants need to learn and adapt as well.
Arvato Financial Solutions, an integrated financial services provider, offers vital services around ecommerce safety for some 2,000 odd customers. One of eight divisions of Bertelsmann – the German media, services and education giant – Avato has over 68,000 employees and delivers a rich portfolio of payment-related BPO services. Arvato recently partnered with Microsoft, inovex GmbH, a cloud and big data specialist, and a few of Arvato’s e-commerce customers to create a fraud detection solution using Microsoft’s big data and machine learning offerings.
By combining Azure services with the open-source Storm and Hadoop frameworks, Avato built an integrated cloud-based solution that uses a modern lambda architecture to process massive data quantities using both batch and stream processing. The batch path transforms existing data using Hadoop. Then, by applying machine learning algorithms, the solution develops self-learning analytical models from past fraud cases, for early recognition of any new fraudulent approaches. The stream-processing path captures incoming real-time transaction data via Azure Event Hubs. It then analyzes the data with the assistance of Storm and Azure Machine Learning to uncover fraudulent activities as they happen.
An important goal of the project was to visualize and monitor the models, and Power BI serves this function, displaying datasets drawn directly from cloud sources, Azure HDInsight, and SQL Database on several large screens in Arvato’s monitoring center.
Avato’s investment in good cloud design is paying for itself, helping the company reliably fulfill SLAs using cloud services. Their flexible architecture enables rapid deployment, which is key for fraud recognition in an international e-commerce setting. Using Microsoft machine learning on big data, Avato has created an innovative e-commerce fraud recognition solution and built the basis for innovative financial BPO services based on Microsoft Azure.
Read more about the Arvato Financial Solutions story here.
Complex Networks Automatically Curates Personalized Website Content
Complex Networks is a youth culture media company catering to millennials and part of Verizon Hearst Media Partners, an American entertainment and media holding company. With a focus on music, style, entertainment, sports and popular culture and a robust portfolio of popular channels such as Complex, First We Feast, Collider, and Rated Red, plus a multichannel network with more than 100 websites, Complex Networks has built a powerful pop culture brand presence. Their sites feature over 54 million unique visitors a month and an average of 800 million video views per month.
The company’s main website, Complex.com, strives to present visitors with fresh, engaging materials to attract their interest. The current focus is around video consumption and efficiently showing visitors the right content at the right time, so they spend more time on the site, visit more pages, watch more videos and generate more ad conversions. The company wanted to improve engagement on its owned and operated websites, specifically around automating and personalizing the selection of featured site content. Historically, the company’s editorial department manually curated what content to highlight on the site, using factors such as what content was most recent, or trending, etc. While this took advantage of editors’ subject matter expertise, it was also inefficient, especially in contrast to the big social networks, which algorithmically recommend content to drive a ton of traffic.
It was clear to Complex Networks that they needed to invest in automation and the power of AI, to compete better.
The team initially considered crafting its own solution to add recommendation logic to the website, but ultimately decided that outside services would offer better results. They learned about Microsoft Custom Decision Service, an open-source reinforcement learning system running on Azure. One of the Microsoft’s Cognitive Services APIs, the Custom Decision Service harnesses the power of AI to help applications create custom experiences with adaptive, contextual decision making. The Custom Decision Service algorithms use observations of actual user behavior to train themselves to continuously improve and optimize decisions.
Using their existing video player setup, the team ran A/B tests comparing Custom Decision Service with other solutions and was pleased with the results. Aside from being easy to implement and understand, the Custom Decision Service was also customizable in ways that other tools were not. The input data and reward can be made to vary by decision, so the service can optimize the most important metrics. This makes it easy to implement business logic on top of the service.
Following the successful tests with the video player, Complex Networks implemented Custom Decision Service on the homepage of Complex.com. The homepage features three articles or videos at the top, and Custom Decision Service now chooses the content for two of the three from a pool of 10 to 15 items selected by the editors. When the team evaluated their A/B test results, they found that Custom Decision Service consistently generated more clicks than hand curation. What’s more, Custom Decision Service works around the clock, with the most sustained performance observed during nights and weekends. Since eighty percent of their audience is returning visitors, it was especially important to Complex Networks to keep their content fresh and relevant to their users’ interests, so they are motivated to keep coming back.
Custom Decision Service incorporates other Cognitive Services APIs to analyze page content and provide better decision making for Complex Networks. For instance, the Computer Vision API scans images on the page and determines what they contain, and can identify people, locations, objects, and the presence of any adult content. The Emotion API looks at people in the images and identifies any strong emotions being expressed. The Text Analytics API parses any language on the page and determines its sentiment. And custom topic models hosted on Azure Machine Learning extract topics for given articles.
Complex Networks has already begun looking at other ways to apply Custom Decision Service. The company is rolling out a new video product that presents visitors with one video at a time and lets them swipe right or left on the video, based on their feelings about it, and move on to another. They have built a prototype using Video Indexer – another Cognitive Services API – to index and analyze the video content, and are considering using Custom Decision Service to make the important choice of which videos a visitor is most likely to enjoy. By helping the editorial department curate and select pieces of content to be distributed through other outlets, and allowing editors to focus on creating the best content, as opposed to spending time ranking articles, Custom Decision Service can also help improve the editorial processes within the company. Because Custom Decision Service is constantly learning and refining its choices, the results just keep getting better over time.
You can learn more about the Complex Networks solution here.
ML Blog Team