Decision Intelligence For Dummies - Pamela Baker

Скачать книгу

      However, both data and AI democratization still require data professionals to develop more intuitive and highly automated software to remove barriers before non-data professionals can use the tools in ways that bring their own talents and skill sets to bear. Think of this as very similar to the path other software has taken. For example, Microsoft Office enables people to create documents, notes, spreadsheets, and PowerPoint presentations without knowing how to write code, what keyboard commands to give, or anything at all about how the software works. This is the path analytics and AI software are headed down now.

      So, who are these data professionals who are making and/or using analytics and AI to provide you with the pointers you’re currently getting from various analytics software?

      Typical job roles in data mining and analytics are data scientist, business analyst, data mining specialist, and data mining engineer (and variations of the same) to reflect a specific industry such as healthcare data analyst and risk-mining data scientist.

      In AI, job roles include AI scientist, AI researcher, business intelligence developer, robotic scientist, software architect, data scientist, and data engineer, among others.

      All these jobs will continue to be important positions in many organizations, and the demand for people with these skills will remain high for the foreseeable future.

      However, much of their work is also being automated as part of the data and AI democratization movements.

      As to specific examples of the work that these professionals collectively and individually produce for use in several business areas, below are some of the more common use cases for traditional data analytics and/or AI automated decision making:

       Anomaly detection, also known as outlier analysis, is a step in data mining (which can be aided by AI/ML or not) that finds deviations in the data from the norm, such as events (purchases on a charge or debit card in another country from where the cardholder is known to live or be, for example), and data point changes (attempts to change a social media account’s password via a device or browser that the true account holder is not known to use before, for example).

       Pattern recognition is the automated recognition of patterns discovered in the regularities in data. One example would be finding earlier signs of cancer in patient data than doctors and diagnosticians previously knew existed.

       Predictive modeling, also known as predictive analytics, analyzes historical patterns in the data using a mathematical process to predict future events or outcomes. One example would involve predicting when a machine part will need repair or replacement based upon its past usage compared to how long identical parts lasted under the same conditions.

       Recommendation engines analyze data to make recommendations or suggestions based on users’ past behaviors. Examples include analyzing your purchasing patterns in order to offer you a coupon for a grocery item you should be ready to buy again soon, or to recommend a movie based on movies you watched and rated earlier.

       Personalization systems use data analysis to customize a service, product, or automated communication. Examples include marketing emails sent to large numbers of customers, each personalized with the customer’s name and a custom discount offer for a favorite product or service.

       Classification and categorization systems automate the organization of vast amounts of data. Examples include sorting data files and data sets according to importance, topic, secrecy level, or other identifier; legal requirements governing the handling of specific data points (think of laws like General Data Protection Regulation (EU GDPR) which limit where personal data can be stored); and the nature of the data (such as structured machine data or unstructured Twitter posts). Data must be correctly classified and categorized for analytics or AI to work correctly. Automation is the ticket here because there’s so much data that it’s impossible to do it manually.

       Sentiment and behavioral analysis is contextual data mining to discover and analyze the subjective expressed responses (sentiments or feelings) about a brand, product, service, idea, political candidate, and so on in online conversations or customer channels (conversations and customer ratings found in texts, on websites and blogs, in voice recordings or streams during phone calls, and app rating systems. Did you rate that Door Dash driver’s service in the app? Yeah, that sort of thing!) Behavioral analysis can extend beyond sentiment analysis to include things like how long you spent reading a news article on your phone and how many times you return to a website, to what time of day and what device you normally use to post on Facebook.

       Chatbots and conversational systems frequently appear as a popup sales or customer service chat box on websites where you can ask questions about a product or service or your account and get an automated answer from the resident AI-powered chatbot. Some of these are so good it’s hard to tell they aren’t human customer service agents. Data on the user and on the stated problem is collected and analyzed to rapidly respond with answers the user needs. Examples of other conversational systems include every digital assistant you’ve ever heard of: Alexa, Siri, Google Assistant, Bixby, and Cortana. Each is a data king, with Alexa and Google Assistant reigning over two of the largest kingdoms in terms of technical and market prowess.

       Autonomous systems are actually a network or a collection of networks that are all managed by a single entity or organization. Data is live streamed and typically analyzed at the sensor or gateway level, although some data is often sent to a data center for additional analyses later. Think the Internet of Things, such as self-driving cars, robotic systems in manufacturing, and smart cities that use information and communication technologies (ICT) to increase operational efficiency, share information with other systems (such as self-driving cars), and promote sustainable development.

      There’s no question that the above list is populated with wonderous achievements that would not be possible (or at least not at such huge scales and fast speeds) without data, analytics, and AI. Nevertheless, the promised “actionable insights” produced by analytics and presented in many of today’s fancy visualizations and dashboards to business users are often merely pointers. They point to something you might want to use as a key factor in your decision, but they aren’t in a position to make that decision for you. You have to conjure some mad data interpretation skills and do some creative problem-solving on your part to figure that one out on your own.

      Making a decision

      Pointers (also known as actionable insights) are typically useful in so far as they go. The trouble is that they point to possible decisions but don’t suggest any. Users are often unsure about what action to take, or which option would produce greater value for the business. By contrast, the decision is the be-all-and-end-all of decision intelligence, and everything else in the process supports that decision.

      

Whether data driven or decision driven, in both cases humans are the decision makers in this context. It’s just that they decide at the tail end of the process in traditional analytics, whereas they decide in the lead position of the decision intelligence process. The starting point for the decision maker matters in terms of the level of control a person has over the impact and value. It’s hard to exert much control from the rear.

      A history lesson

      Disgruntlement with the limitations of traditional data mining is growing. Increasing frustration often leads to both the business side and the AI and IT sides starting to wonder aloud: “What’s the point?”

Скачать книгу