Data is everywhere. Data about customers. Data about products. Data about data. But all those bits and bytes (and petabytes) have little value to the enterprise if they exist in isolated silos like disconnected jigsaw puzzle pieces. Data scientists like myself and those on my team (Intel IT’s Advanced Analytics) bring the expertise to understand the data quality, decide whether the right data is already there, and provide guidance on where to get additional data if necessary. Our team helps build the connections that turn raw data into galvanizing insights about the business. The IT department plays a crucial role in empowering data scientists to extract the most value from enterprise data.
As described in our recent white paper, “Data Mining Using Machine Learning to Rediscover Intel’s Customers,” Intel IT’s Advanced Analytics group embarked on a project to help the Intel sales and marketing organization solve a specific problem: how to leverage the data they currently have to improve their sales strategies, gain efficiencies, and increase Intel’s revenue through better reseller engagement. Through our efforts, the sales teams have better access to past sales data, information about customers, membership programs, and so on. The machine learning platform we built provides the “big picture” of Intel’s customer base so that patterns and behaviors become apparent. Using this system, Intel call center agents and sales representatives can understand the customer and be prepared to talk about the topic that is most relevant to that customer.
Meeting Machine Learning Challenges
Intel is actively pursuing new markets, such as the Internet of Things (IoT) and cloud-based services. These new markets pose a challenge for Intel sales teams: finding the right customers to engage with and knowing how to communicate with them—what do they care about? What is likely to catch their interest? In this age of information overload, email messages, Twitter tweets, and LinkedIn posts are glossed over or deleted unless the reader sees something relevant to their immediate needs. The sales teams need more information so their communications are not considered spam, but instead pique readers’ interest and result in click-throughs.
As we built the machine-learning platform, we included customer training data, information about which areas of websites they visited, and past sales data. This gives a lot of information, but it’s only part of the picture. As data scientists, we understand that the sales and marketing people also want to find customers they don’t know about yet. We went about that by leveraging new data sources. One of the most readily available sources is Intel’s customers’ websites and other communication channels. These customers want to sell products to their own customers and are therefore communicating with their customers through web pages, Twitter accounts, and so on. We are harvesting that data -- listening to their voice. By doing so, we’re able to complete the picture and understand what Intel’s customers’ needs are.
For example, suppose the sales team wants to boost sales for Intel® IoT Gateways. An analysis of websites and social media content can pinpoint companies that are selling sensors, smart building management systems, fleet management products, or other IoT-based services. These types of companies -- which could be in many different verticals such as manufacturing, construction, or retail -- may be interested in hearing more about what Intel has to offer in the IoT space and are more likely to open an email message targeted specifically for them.
One challenge we encountered as we integrated the website data into our machine-learning platform was classifying websites. Several third-party suppliers offer website classification tools, but we found those tools too inflexible for our needs -- the information they provided wasn’t specific enough. For example, a tool might classify a website as relating to “IT hardware,” whereas we needed a far more granular classification scheme to better meet the sales and marketing organization’s needs.
With the machine-learning platform we built, sales teams can create ad hoc website page categories -- there are no preset classifications. This enables the teams to “define and refine” as they go along, constantly updating their classification to reflect their interest at a particular moment. The machine-learning platform is flexible and can quickly adapt to shifting sales strategies and targets.
Another challenge was how to deal with messy data. In an ideal world, data would be clean and well organized, perfectly set up for use. But in reality, enterprise data is often messy and disorganized. Despite such an environment, the Intel IT data scientist must deliver insights. This is done through careful problem formulation, exploratory data analysis, and data modeling. In a world where deep learning seems like the solution to every problem, the data scientist needs to know and understand when the “bleeding edge” of science actually doesn’t deliver as well as tried-and-tested algorithms.
That is not to say we are cavalier or complacent about messy data. In fact, we spend a lot of effort on cleaning and carefully pruning data, because the messier the data, the more “noise” in the analysis environment, which reduces the overall value of machine learning. But like most big enterprises we still are far from having clean, pure data, and therefore we always strive to use algorithms that are flexible and robust enough to deliver powerful insights even when the data is messy.
We are already seeing positive results with the machine-learning platform: click-through rates for email newsletters are now three times higher, and resellers contacted through the program are completing Intel training at a rate three times higher than the rest of the sales pipeline. We will continue to refine our algorithms and the sales and marketing machine-learning platform. We are confident that our efforts will be fruitful and will help drive tens of millions of additional sales each year.
Read the IT@Intel White Paper, “Data Mining Using Machine Learning to Rediscover Intel’s Customers,” as well as our recently published 2016-2017 Intel IT Annual Performance Report, “Accelerating the Pace of Business through IT Innovation,” to find out how Intel IT is using advanced analytics to transform Intel and create business value.