The AI-Powered Enterprise. Seth Earley
Читать онлайн книгу.
Problem One: Friction and Siloed Communication
Companies, especially those born in the pre-digital age, are created not holistically but in a piecemeal fashion. Departments and functions operate independently. When a signal comes in, it doesn’t instantly go everywhere it needs to go; it must be routed from one place to the next. Internal communications, databases, and other systems act as messengers to relay a signal throughout the organization. Imagine if the brain told the legs to run, then the legs had to go and ask the adrenals for some energy to execute the movement, and the adrenals had to then go and verify that the decision was authorized by the brain. Before you could move, you’d be run over.
In most companies, there is a lot of friction along the way. Challenges like differing or competing priorities, incomplete understanding of the big picture, dropping the ball, and double-checking all create drag as information makes its way through the organization. That’s not fatal when the information being handled is human brain–sized, which is why manual systems and processes have worked for companies so far. But it collapses under the weight of the enormous data sets required for AI and high-tech interventions and for the rapid response required in a digital economy. The key to making AI work is reducing that friction and speeding the flow of information.
Problem Two: Incompatible Data and Language
From time to time, every organization needs to do housekeeping on its digital working files, archived data, and other forms of information. But how do you do that consistently and cleanly when every process or function uses different terminology or a different way of organizing things? Too often, the various groups within an organization are speaking their own languages—or at least their own dialects. Information is stored differently, and teams use different terminology that speaks to their needs and processes but that doesn’t consider the broader organization. This leads to manual workarounds and to information getting lost in translation. As I’ll show in more detail in chapter 8, technology companies designed collaboration tools to make it easy for people to create and use data, not to make the data effective in the context of a corporation. The result is an information environment full of poorly designed, fragmented, and disconnected systems. It’s not a surprise that many of these systems don’t share a consistent set of data language.
Problem Three: Junk Data
Entropy is another fact of the physical world mirrored by the digital one. Every system tends from order to disorder, and reestablishing order requires energy. We all experience this in our day-to-day lives: our desk gets messy and we need to put energy into organizing it. The house gets dirty and requires energy to clean. Information gets messy, too, and organizing it also takes energy. For example, we have to delete or label our email messages, put files into folders, or cleanse a document repository of any out-of-date material. Productive activity seeks to reduce the amount of disorder and therefore reverse the entropy of a local system through the application of energy.
Most organizations today are drowning in junk data as the incredible volume of digital information produced massively outstrips the energy available to manually practice good data hygiene habits. This book will show you how to get your digital house in order so the robots can keep it clean for you.
TAGGING UNCLOGS THE FLOW OF INFORMATION
Tagging is a central part of what makes ontologies able to speed the performance of enterprises.
A manager’s objective is to give employees direction; provide resources and the information necessary to solve problems; and allow creativity, hard work, and expertise to generate solutions that have value to customers and the marketplace. All of this activity is fueled by knowledge, and the way that knowledge flows through the organization’s networks is key to its efficiency and effectiveness. Important information needs to be flagged, tagged, and held up as meaningful. This information includes, for example, the needs of customer segments based on market research, solutions to engineering problems, the current quarter’s strategic objectives, and the features of a new product and how it is different from the competition. These are all signals that need to be separated from the noise of day-to-day communications.
The separation comes from tagging that identifies the information as important. Then someone can take that important piece of data and use it to solve their problem. (As I will describe later, that tagging, or separation of signal from noise, can happen at multiple levels—from manual information and data curation performed by humans through AI and machine learning approaches.)
Not having the right tags causes meaning to be lost—the noise drowns out the signal. Inefficiencies in information flows, lack of consistent terminology, and systems that don’t talk to one another bog down operations and create waste. We’ve all seen what this looks like and felt the pain: for example, folder structures on shared drives that are redundant or nonsensical, or that contain labels meaningful only to individuals (“Joe—Important docs”), along with excessive translation and manual manipulation of data because systems do not use the same terminology or data standards.
Constraints can be liberating. The enterprise adapts and thrives when simple tagging rules that speed information flows create value through emergent behaviors within the system. Artificial intelligence is a mechanism that, properly designed and applied, speeds information flow and enables those emergent efficiencies, including tagging. It can therefore help every part of your organization do its job more quickly, efficiently, and consistently. To succeed over the long term, you must allocate resources in a sustainable way to build AI-ready assets of increasing value—including the ontologies and data structures that will serve all aspects of the business. The business needs to use continuous feedback to develop these structures, apply them to the production of value, fine-tune systems and processes, and adjust and make course corrections.
What does such a solution look like? There are three areas in which your company needs to rethink its resources, focus, and attention so it can exploit the power of artificial intelligence. My systematic plan that you can follow to prepare for an AI-powered enterprise includes these elements:
1.Data. This includes how data is architected, managed, curated, and applied. You must wrangle your messy and inconsistent data into a refined asset for high-precision, high-leverage activities. The organizations with the most agile architecture, highest-quality data, and best algorithms for applying that data to address customer and employee needs will win.
2.Technology. To deliver personalized experiences to customers (whether internal or external), appropriate technologies must scale processes by relying on detailed enterprise knowledge. They must also remain adaptable as tools, approaches, and information sources evolve and change.
3.Operationalization. Just as with the reengineering transformation of the ’80s, these improvements require a commitment to new forms of organizational discipline, with new accountabilities and metrics. This includes rethinking how the organization delivers value through end-to-end digital processes.
Let’s examine each of these three requirements in detail.
Data: The DNA of the Organization
Most executives reflexively nod their heads when discussing the value of data. “Data is extremely valuable to our enterprise,” they say. “We are undergoing a data-driven digital transformation.” According to this thinking, good data is good, and bad data is not good. But when it comes to creating ontologies—which means fixing the foundational data issues in a sustainable way—and they see the price that the enterprise needs to pay, these same executives deprioritize projects like fixing the data-quality issues. “It’s not that valuable” is the message that is communicated to the organization. Because data issues are not addressed properly at a fundamental level, tens or hundreds of millions of dollars will be spent on digital transformations that will fail in the long run.
High quality, findable, and usable data is an essential part of the AI-powered enterprise. Machine learning can find patterns in unstructured data, but it cannot make sense of information that is of poor quality or missing. Data