While Artificial Intelligence (AI) still has an air of mystery about it experiments with machine intelligence tools and AI applications are underway on many fronts. Their motivation stems primarily from the increasing number of opportunities that AI is presenting. As the technology improves, consumers are becoming increasingly acclimated to voice interactions and more savvy about using virtual assistants in day-to-day tasks. Virtual assistant and chat interactions will be more commonplace in the corporate setting and be the preferred mechanism of interaction with enterprise applications. In our opinion, the AI label is becoming less meaningful. We believe that organizations will be discussing specific business capabilities enabled by AI technologies, rather than focusing on “generic AI.”
In the January issue of Harvard Business Review, Tom Davenport and Rajeev Ronanki discuss three kinds of AI projects that they have identified from their study of 152 projects at organizations exploring the value of AI for the enterprise . These include:
Robotic process automation (RPA), initiatives in which repeatable, unambiguous processes are enabled that replace humans performing rote tasks such as copying data from one application to another. [Took out the verb to make it consistent with items 2 and 3]
Cognitive insight, applications that help organizations make sense of large amounts of data – referred to as “analytics on steroids.”
Cognitive engagement, in which humans interact with systems using natural language or have more personalized interactions through recommendation capabilities.
In the case of RPA, the processes are well defined and the outcomes predictable. RPA is to administrative work what manufacturing automation is to an assembly line. Some advanced automation requires the ability to deal with less predictable circumstances (such as picking parts that could be in different orientations). However, factories have been using automation for repetitive, defined motions and tasks since the introduction of the assembly line at the start of the last century. Similarly, RPA is not typically used in unpredictable situations or where judgment is required and therefore many RPA tools are based less on machine learning algorithms and more on rules-based applications. Rules can capture and codify human judgement based decision-making processes, but within a more limited context and narrower domain of possible choices. For example, incoming email for a customer service organization can be processed with RPA by classifying the email by inquiry type – based on rules – and then extracting customer details for entry into a CRM system.
AI technology can also work under less predefined situations, such as making sense out of large amounts of data and uncovering patterns and new (cognitive) insights. However, this scenario works only if a human sets up the technology in a way that enables these outcomes. An analyst setting up a machine learning project to analyze customer data needs to have a sense of what is important and what they may discover. The analyst also needs to know what to look for, and set up the tool in such a way that it allows for validation of a hypothesis, rather than simply analyzing the data. Typical questions might be: What insights might the data reveal? What parameters does the data model include? What inputs and data sources are being leveraged? Advanced machine learning platforms will enable this discovery more readily by preprocessing information. They may also suggest various features for exploration, but understanding of the customer allows for interpretation of patterns and recognition of insights.
The third area identified by Davenport and Ronanki – cognitive engagement – encompasses a broad range of applications and makes use of predefined rules as well as machine learning capabilities. In this class of AI, the goal is to make the interaction with a computer easier, more efficient, and more closely aligned with the needs of a user. Rather than requiring a user to search through a long list of documents for an answer, for example, a Q&A system using this class of AI would combine additional signals from the user context to provide a specific answer to a question.
Sometimes the interaction is through a chatbot or intelligent virtual assistant. The interaction is more natural and the goal is to make things easier for the user – to reduce the cognitive load on the user. This facilitation can be in a customer service scenario or product selection scenario. The recommended answer, or recommended product, is presented based on understanding something about the user that can inform the algorithm. That might be the user’s past purchase, interests, the pattern of site interactions, or other signals. Natural language is richer in meaning and context than a simple keyword. Machine learning is used to understand the user’s objective, goal, or intent. and then to respond to that intent.
The ability to provide this kind of high quality interaction depends on the same things that any customer experience capability requires – understanding the user and their needs based on personas, scenarios, use cases, and tasks. It also depends on the availability of the correct content, product information and data sources needed to support solve their problem or meet their needs. In AI parlance, developers need “training content” that allows the AI to process user signals and retrieve the right information. In addition, they need customer data models and correctly architected information that provides the input to drive engagement.
Davenport and Ronanki caution against “moon shot” AI programs – the large, costly, ambitious projects that attempt to solve very difficult problems using these new technologies. Instead, they observe that many applications that are more narrowly focused are more achievable, provide tangible benefits, and help move the organization efficiently along the learning curve.
In the past year, EIS has worked with a variety of organizations on these types of realistic and practical AI programs and helped them take initial steps into powerful, exciting new capabilities. In some cases, we began with a challenge that is common to any company with products – that of understanding, modeling, classifying and cleansing product information. We helped apply machine learning to content programs to improve search and information retrieval and access. We also developed conversational search and chatbot proof-of-concept and prototypes using various technologies on the market from Microsoft, Google, and a number of startups. Finally, we evaluated readiness, and developed strategies, roadmaps and architectures for their next iterations of AI programs. Each of these projects moved the organization toward a future state of operationalizing AI while solving pressing problems in the near term.
In our next Earley Executive Roundtable, we will be describing our work in enterprise AI and the prerequisites for successfully implementing AI. If you are interested in exploring the potential of AI but are not sure where to start, this discussion will be very helpful. Our client partners will discuss projects that EIS has supported, and areas of opportunity for organizations in 2018. Be sure to invite your colleagues, managers, and executive leadership to learn from front lines of enterprise AI.