Use Case Paradigms for Enterprise Bots


Historically, organizations have relied on mechanization and automation to improve process throughput, process efficiency, and process defect rates, amongst other organizational or process performance metrics. Some organizational metrics such as capacity, efficiency and quality have been the mainstay of improvement teams and executives, although of late, rapid developments in artificial intelligence and bots are ushering in a brave new world of corporate productivity. Bots and automation have been viewed by some sections of industry and some academic commentators as harbingers of a new industrial age, where humans are replaced by robots not only for trivial, repetitive tasks, but also for the tasks and activities that have in the past attracted human ingenuity and intelligence.

Given the accelerated pace of improvement in artificially intelligent systems, the notion of an artificial general intelligence, that can “learn like humans” is not far-fetched. In a survey of experts, cognitive science and associated technologies, along with deep neural networks and developments in hardware were touted by most experts as the best enablers of a human-level machine intelligence (HLMI). Despite these notions, organizations have been slow to warm up to the possibilities of adopting AI and automation by bots, due to diverse reasons. In this post, we will explore a few broad patterns in how executives can think through the use of bots, artificial intelligence and automation. The purpose of this post is to aid executives in organizations that are considering bots and AI as enablers of productivity and as harbingers of new capabilities. Here, The Data Team provides a framework with which to evaluate the nature of problems, and associate those problems with automation solutions that fit these problems.

Organizations are Warming Up to Bots and AI

Amidst these doomsday commentaries, organizations are slowly evolving to face challenges from their competition that will adopt automation to improve organizational performance at many levels. Workforces will doubtless evolve to develop the skills that organizations will require from their human resources, since organizations enabled by AI and bots will constitute significant capability leverage, especially for tasks where human intervention isn’t really required. We’ve seen the inception of such a paradigm in the automated vehicle technologies, where repetitive jobs such as truck driving will be seen as a thing of the past in perhaps a decade’s time. Organizations such as Mack, Volvo and Mercedes have evaluated automated driving technologies for trucks that provide more fine-grained route and configuration control over multiple trucks as a caravan, for instance.

Organizations will naturally begin to expect higher value from their employees, and the ability to manage people, processes and technologies are likely to be part of this revised expectation.

The Four Broad Paradigms

As executives warm up to different bot and AI use cases and possibilities in their organizations, they are faced with improving organizational efficiency in two typical ways:

  1. Improving the extent of process automation (% of automated process time in total process time)
  2. Increasing the complexity of process automation to lead to real process enhancement (simple task level automation, to complex automation akin to expert systems)

These two broad ways of thinking about bots and AI generally distill down to the following paradigms.

These four paradigms are:

  1. Full automation: replacement of entire workflows, filled with simple, often chained tasks
  2. Partial automation: replacement of parts of workflows
  3. Quantitative (simple) enhancement: Improve workflow performance to allow higher quantitative performance (e.g., throughput, quality or error rate)
  4. Qualitative (complex) enhancement: Improve workflow performance by adding bot driven capability that is not possible with humans

We’ll look at each of these paradigms in more detail below.

Full Automation

Some data-intensive and interaction-intensive workflows that require already-shared data can be automated by bots to a very large extent, thereby codifying the value addition in an interactive process. This kind of bot-driven automation is valuable in some financial transaction settings, and in some business reporting settings. Automation has been possible in the past wherever clear data and routines are established. With bots, automation can provide natural language interactions.

Examples of this include:

  1. Cancelling an ATM or Debit card with a prompted bot
  2. Paying your general-purpose utility bill with a prompted bot
  3. Fetching a certain standardized report from a server

While the same activity could have been done by the user or a human agent in the course of a few minutes, the interactivity and on-demand nature of the bot transaction makes the process very efficient, and hassle-free. What’s also attractive is the fact that such bots can be built as a service and sold as APIs.

Partial Automation

Replacing portions of workflows can make overall processes efficient in situations like medical diagnosis, tax planning, expense reporting, insurance sales, car service and checkup, etc., where the real value may be added in tangible ways due to real world human interaction. In the past, some decision support systems have substituted aspects of these interactions. With sensors built into systems, we can channel the data from patients, tax assesses, business travelers, insurance buyers, or car owners into bots that can reorganize this information to be looked at by an expert. In this sense, the bulk of the up-front data collection for some processes can be automated.

Again, the availability of natural-language-level interactions with bots can improve the interaction quality and reduce error rates in information collection in measurable ways. Key value additions might be:

  1. Information collection and organization from user specific to one user interaction/request
  2. Integration of new information from user and existing information from multiple sources specific to the request
  3. Report preparation for consumption by a human in the process
  4. Simple diagnosis or transformations that can be algorithmically done

Quantitative Enhancement

Some of these situations are where we may do things faster, better or with fewer defects due to intelligent automation and bots. Paradigms within this are:

  1. Finding information fast: This is another use case for bots – rather than open a browser, key in a search term and look for specific information we need, the Q&A format for finding information may be much faster and could allow humans to work faster.
  2. Scheduling meetings and events:  In this use case, quantitatively enhancing workflows could mean lower process times for scheduling meetings, automatically checking availability of participants, alerting about meeting conflicts using natural language, etc., – it is a case of using numerous smaller capabilities to deliver a large improvement in meeting scheduling process time
  3. Reducing mistakes and erroneous input: Measurement aids can improve input, auto-correction, automatic and custom formatting, can all be done faster. Integration with devices that can provide images, sound and video can enhance text information and reduce reporting time.
  4. Sending information or fast notifications: Sometimes, in situations where we’re monitoring infants, heart disease patients or the elderly, we may require specific kinds of information to be sent fast. This is done efficiently by melding together features from chatbots and IoT enabled systems. Such technology could also be used for automatic food or water replenishment systems, turning on and off specific household devices such as water pumps, or kitchen devices etc.

Qualitative Enhancement

In the examples given above of financial transactions, medical consultations and product sale or service, the current workflows may provide sources of customer dissatisfaction, such as:

  1. Avoiding information overload and ambiguity: with multiple outlets for information (multiple shopping apps, for instance), the absence of a higher-level interface is increasingly a problem with many customer interfaces.
  2. Consistent, higher-level design and user experience: Conversational style apps that work in a natural language abstract away the need to learn about user interfaces or graphical workflows. They can provide a higher quality experience in some ways
  3. Accessibility improvement: Chatbots that are NL and voice enabled may be able to interact with a wider range of users than tech-savvy smartphone users. This will allow service penetration to happen faster in bot-based services such as mobile shopping enabled by bots
  4. Virtual Concierge: With the advent of bots, general purpose concierges may replace a diversified tool set meant for numerous day to day activities. Whether booking cabs, ordering goods online, or setting up appointments, they could become a single interface
  5. Smart device interface: With the advent of sensor-embedded systems and IoT, a bot interface would be a more convenient way to reach out and control different devices of different kinds
  6. Timely natural language recommendations: Based on physical and digital contexts of the users, and their preferences, could recommend specific products or services to them, a la The Data Team’s Cadence, which is a robotic data scientist for enabling marketing functions in completely new ways. Except now, with natural language interactivity, where we can find out more about the offer, etc.

Natural Language Processing for Human-Bot Interactions

In this section, we’ll briefly discuss how rapid improvements in applied natural language processing and understanding are enabling fine-grained interactions between humans and bots in organizations.

Traditionally, problems like text processing and conversational engines with many rules have driven development of some applications in the human-computer interaction space. With the advent of machine learning driven natural language processing and natural language understanding functions, there is an opportunity to integrate many new capabilities which improve error rates for natural language based learning models.

Google has been pre-eminent in the use of NLP for enabling intelligent text analysis for translation and for consumer-facing products such as Google Home and Google Now, with rivals like Amazon (with its Alexa suite) and Microsoft (with its LUIS engine) following close behind. Especially noteworthy is Microsoft’s focus on using NLP in business dashboards, such as the use of Cortana and LUIS in Power BI, to enable real time natural language querying (which, if you were to apply the aforementioned model, would amount to a qualitative enhancement for business analysts). On the other hand, we have frameworks that aren’t based on open technologies like Siri by Apple, which are more conversational in nature and narrower in their capabilities. The availability of open natural language processing libraries and their increasing integration with frameworks like Apache Spark offer organizations new possibilities.

Applying This Framework

So far, we’ve discussed a model to think about enterprise automation enabled by bots. This model allows us to situationally analyze different automation possibilities and solutions and characterize them. Once such characterization has been done, executives should plan for the development, deployment and change management aspects. We’ll briefly discuss these aspects below.

Development and Deployment

Bot development, like other system development, is an iterative process. Fundamentally, bot development is a question of balancing agency and complexity. While agency is the fundamental benefit of an intelligent system that is situationally aware (unlike other purpose-built systems that depend on specific or user input), complexity is determined by the range of rules, target use cases and engineering for machine intelligence that has to be built in. For organizations to practically benefit from the bot development and deployment process, a continuous deployment model, with iterative improvements to bot features should be evaluated.

A continuous deployment model for bot deployment would evolve through different stages of functionality. Early stage development would be focused on quick wins, that enable bots and automation to address a pressing problem with a short, focused scope. As the bot is being monetized in the business, mid stage development will begin to address quantitative and qualitative enhancements to the bot, that are again, continuously tested and deployed. Subsequently, the development activities will focus on adding more features to address as much of the feasible feature space as possible with the bot.

Change Management

Executives in organizations that are considering advanced automation to manage productivity have to be ready to manage change of various kinds. The Kotter model of change posits the formation of a guiding team that preaches the urgency of change throughout the organization, thereby allowing the business case for change to be made across the organization. This model also discusses the benefits of demonstrating quick wins, such as those you can demonstrate with continuous deployment and delivery. There are fundamentally three kinds of issues and grounds for resistance that change management teams associated with automation will have to address:

  1. Employees see automation in general as a threat, rather than as an enabler
  2. Employees do not believe that the specific automation solution at hand addresses the key problem at hand
  3. Employees (and managers alike) are resistant to change in general, and automation is one more change agenda they are slow to warm up to

Such concerns exist about any new technology in organizations, as they have existed in the past for a wide range of technologies. Change management is a classical hard problem in organizational management, and one that we at The Data Team believe we can help with. We believe that enabling the agents of change with data can bring a real benefit to discussions around technology-specific change programs.

Naturally, all the faculties of effective change managers such as the following will come into play:

  1. Prudency in addressing issues at hand
  2. Effective conflict management, especially starting by acknowledging conflict
  3. Using burning platforms and forums for change to address key issues, and
  4. Organizational messaging, for information, education and as calls to action

However, the value we add is in being able to get the most nuanced message across – thereby allowing change to be driven from the top, and with data and facts, thereby making a strong case for the messages and burning platforms we use as grounds for initiating change.

Concluding Remarks

In this post, we have discussed a broad swath of issues that address automation in organizations. Starting with the historical context, we have discussed a model to ideate automation paradigms, and have discussed full and partial automation, and qualitative and quantitative enhancements to processes and workflows in organizations and in products. We have touched upon developments in the natural language processing space that have driven technology companies along into new growth and value propositions. Finally, we have discussed some practical aspects of bot development and deployment, and a few essential aspects of change management for executives considering automation. In the enterprise world, understanding these bot paradigms and adopting effective development, deployment and change management strategies will surely lead organizations to better success along their automation journey.

The Data Team provides value added services in data strategy, in the development of purpose built bots, and in the deployment of data science for change management activities. Please contact us to find out more about our offerings.

Leave a Comment