A critical step in digital transformation is to enable the free flow of information throughout the enterprise. But various forms of friction can obstruct this flow. Friction is anything that slows down information access, information retrieval or information manipulation. One of the most obvious sources is the ongoing use of paper, which is still present in a surprisingly large number of business processes.
Other sources of friction include inconsistent data architecture and naming conventions, poorly understood or user-unfriendly applications, the need for manual conversions or manipulations, and answering the same questions over and over again rather than resolving them and operationalizing the answer.
Answers that are provided may not be well documented or easily findable, making it difficult for users to benefit from work already done. Information in general may be difficult to find due to poorly designed information architecture. Or it may be inconsistent across different information systems, resulting in the need for human intervention to resolve the discrepancies. Excessive communication via e-mail can also cause friction. People may miss a response amidst large volumes of email messages, or they may have to go digging through their inbox to find the relevant answer.
Finally, poorly integrated information systems can disrupt the flow of information and delay progress on actions that rely on more than one system in order to be completed. All of these sources of friction point directly to the conclusion that a solid information architecture and well-designed information management system, along with an overall vision for the digital transformation, are prerequisites for success.
The Impact of Friction vs Goals of Digital Transformation
Friction and its associated slowdowns impede the organization’s ability to act on information and insights about customer needs and understand dynamics of the marketplace. These obstacles can also block the ability to leverage new tools and technologies that can provide a competitive advantage.
Digital transformation initiatives seek to digitize end-to-end processes – certainly by removing paper from organizational processes, as well as removing unnecessary procedural steps, reducing manual intervention, and finally, integrating the disparate processes and systems that are required to create value, and deliver it to customers.
This is a tall order, and requires a comprehensive review of information management and processes; for example, the way information is collected, organized, and stored, and the purpose of each business process. The existing business processes need to be understood, and a future state envisioned. In addition, the underlying technology in use by the organization to be analyzed, and an assessment needs to be made as to what is needed for the transformation.
AI-based technology is often considered as a means to reduce friction. Organizations are experimenting with cognitive AI – bots and assistants that speed access to information by enabling conversational retrieval. Just like in Star Trek we will one day just talk to our technology. These conversational assistants can be used for internal or external audiences and processes, and can make tasks faster and more efficient. But understanding the abilities and limitations of cognitive AI requires the organization as a whole to understand the objectives, and to be aware of the key results and metrics that indicate progress toward the objectives.
Educating the Organization
The chief digital officer of a large chemical manufacturing company spoke in an interview about the need for foundational education at multiple levels. He noted that any AI initiative should begin with use cases and scenarios, identifying the various stakeholders and departments that are involved in each business process.
Executives who are less experienced with AI need to understand what is realistic to expect from machine learning and artificial intelligence and what they need to do to prepare for AI. For more advanced technical practitioners, the education needs to go a step further and identify the means to achieve business outcomes – how to apply technologies, what standards may be in place, what services architecture can be leveraged, and so on. The critical piece is to have realistic expectations about what the technology can achieve.
Many times, leaders placed in charge of large initiatives have knowledge gaps themselves about how certain foundational elements need to be developed in order to prepare for advanced capabilities. Because of the dependency of AI on core architecture, structured content, or quality product information, many aspirational outcomes are unrealistic, because those elements are not present. And sometimes the technical people can’t talk to the business people. The technical people may know how to solve the problem, but they may not fully understand the business problems they are trying to solve and therefore can’t fully explain the level of effort to solve a particular business problem.
What Cognitive AI Requires
One commonly used AI tool is the chatbot. Let’s see what we can learn from discussing a chatbot implementation in order to better understand how a business problem can be solved with this technology.
A chatbot is simply a personification of search; a tool that emulates the responses a human might give. It’s a channel for getting the information one needs in an easy format. To enable a chatbot, an organization needs some key ingredients:
- Use Cases – Who is the person that needs the information, what type of information do they need, what will they use it for, etc.
- Content – Essentially the answers to the questions that people are asking or information they are needing. Content must be created, edited, and published.
- Information Architecture – the scaffolding that connects all of the content. The content must be stored somewhere and referenced so that it can be found. Think of the old-fashioned library. Books had titles, locations, an indexed drawer that you could flip through by author or by title, or by subject that referenced the title, with a corresponding location. Information architecture includes elements such as taxonomies, ontologies, metadata, synonyms, etc.
- Systems – Digital content must be stored in systems that must talk to each other. Which is the system of record? Where will the taxonomies and metadata be stored? What tools can we use?
- Processes – People have tasks they must do, and those are represented by processes.
- Governance – Who makes the rules around how information and processes will be created, reviewed, updated, etc.?
Each organization has many systems, processes, and perhaps millions of pieces of content that must be tagged, stored, and described so that it can be found. Each one of these categories can require an enormous level of effort to achieve “chatbot readiness.”
Let’s take a more detailed look at the first two ingredients–use cases and content–and then consider sources of friction and ways to reduce it.
How Understanding Use Cases and Supporting Content
Use cases describe the things people have to get done; the information they need to acquire and the tasks they need to complete. Some are obvious and explicit, but we can also infer the questions that are being asked through analysis of past conversations and content. An example of a use case is onboarding a new employee, or selling a product to a long-time customer. We can model large numbers of questions by defining archetype questions – the classes of questions that can be expanded using variables that cover a larger number of scenarios, which can then be handled by a chatbot.
For an insurance or financial services company, there may be unique answers depending on the state of residence. By using “state” as a variable, individual questions relating to state of residence can be represented by a single use case. This approach can be expanded to many types of use cases. The questions defined by metadata attributes are matched with answers that are tagged with those attributes.
Attributes and entities can also be are also mined from knowledge sources to identify content that meets the needs specified by use cases. Frequently, gaps in content are revealed through this kind of exercise, gaps which then require a plan for remediation.
Friction and the Customer Support Function
Consider the job of a customer service agent and where they might encounter friction. The agents have several screens in front of them and have to switch from monitor to monitor and application to application. They have to look things up in different sources, there are many things that are “integrated at the glass” – that is they do the integrating by going from system to system. In many call centers, there is no easy centralized way of finding information. The inability to find needed information creates a great deal of friction. When the process is delegated to a chatbot, the chatbot will have the same struggle unless the information is properly structured and retrievable.
Knowledge is context-specific, and we need to understand the user and their objectives in order to provide information recommendations: the next best action, next best content, a product recommendation – whatever they need to move them forward and achieve their objective. The customer journey is a knowledge journey.
In the typical enterprise, finding the right content for the chatbot is a very challenging problem, as content creation is siloed by business unit or department. Many organizations have experimented with knowledge management initiatives, but some have found it difficult to sustain value from such programs. Increasingly, knowledge programs are being driven by competitive pressures and the realization that cognitive AI is powered by knowledge.
Fragmentation Leads to Friction
In many organizations, however, knowledge is so highly fragmented that repositories rarely have consistent naming conventions, terminology for tagging, taxonomies, and metadata schemas. No one is responsible for knowledge across the organization and ROI is not clearly defined. Therefore the disordered data becomes a source of friction.
One way to remediate this issue is to establish centralized standards and competency that can support various departments. One organization that has successfully done so handles enormous volumes of knowledge transactions but without significant overhead and a large staff of content and knowledge managers. They do this by having a center of excellence that manages the platform for content operations, including standard taxonomies and content models, workflow processes and metrics-driven governance.
A critical ingredient is a fundamental shift in thinking about content development. Some organizations are creating “cognitive content” operations groups. The idea is that content needs to be designed for conversational access to support specific use cases and developed in “chunks” or components that answer specific questions. Large monolithic documents are broken down into pieces that satisfy process requirements but are still available as a complete document to meet compliance needs.
Rather than using this approach only for cognitive AI, it should in fact inform how all content is developed and managed. High-value knowledge needs to be structured for retrieval. Even if we use the most advanced AI machine learning we still must tell the tools the names of our processes and our products, the problems customers are trying to address and solutions to those problems. That information comes from human judgment – collaboration and communication. The key is to understand the specific usage of content and the needs of the users who need to retrieve the information.
Friction is a challenge to address but also serves as a signal as to where digital transformation is most needed. Identifying the sources can help the organization leverage innovation in process and technology to launch a more effective digital transformation. Organizations need to make a commitment to managing their knowledge resources in an intentional and disciplined manner in order to make the most of their investment in transformation.
 Cognitive AI includes classes of technology such as chat bots and virtual assistants that reduce the cognitive load on the human. There is no actual “cognition” in the human sense of the word though many vendors confuse human intelligence with AI. Henrik Hahn, CDO of Evonik has referred to AI as “augmented intelligence” rather than artificial intelligence – a usage I completely agree with.