Making Business With A Chatbot

Drew Bates

Chatbots are not a new concept. They’re not even from this century. Alan Turing’s 1950 paper “Computing Machinery and Intelligence” laid the foundations of computer conversation via the Turing Test. Almost 30 years ago (1979, the same year Sir Tim Berners-Lee invented the World Wide Web), Moviefone began its interactive telephone service. The first real and widely used “chatterbot,” named Smarterchild, was powered on in 2000.

These fundamental and literal milestones along the chabot revolution make us question why it has taken so long. The answer is quite straightforward and grounded in reality. Chatbots, despite their on-paper potential to represent computer intelligence, have really just been an alternative way to search for information or complete simple, routine operations.

Voice and text commands add value by using their infinite patience to channel your path across a known set of choices with less deviation and more speed. These principles have been adapting quite favorably into customer service. Gartner predicts that by 2020, 85% of all customer interactions will be handled by chatbots.

Most companies say their top three customer questions represent the lion’s share of request volume. One of China’s largest food delivery companies, Eleme, receives 200,000 support requests between 11 a.m. and 2 p.m. on weekdays. Ninety percent are “My food hasn’t arrived yet” and “My food fell out of the container.” These questions are becoming robot-answered with some version of “We’re sorry. Nothing can be done right now. Here’s a discount code for your next order.”

It is no surprise, then, that the upward creep of chatbots has primarily been absorbed by this form of direct questions-and-answers. What we have recently reached could be bluntly described as a glorified FAQ around a tight decision tree. Now, thanks to rising network speeds, edge computing, and service integration, as well as humans’ near-readiness to talk to computers, the real revolution is beginning.

Enter the conversational user interface (UI). This is a more general term for natural-language based interaction with computers. Chatbots are a key part of the interface. The bigger picture includes text, voice, visual patterns (like cards and buttons), cross-platform activities, and a new way of getting complex things done.

The power here is to make interaction natural. In our early years, humans experience implicit cognitive development as they learn to communicate with other people. Everything else is explicitly learned; using a mouse, typing on a keyboard, browsing an OS, or putting multi-currency partial refunds into the booking system for an old sales order. Some are more natural than others. And many require specific training.

But what if this wasn’t the case? What if we could make our everyday actions happen naturally – without training, errors, irrelevance, or mashing buttons – in order to reach the desired outcome with the least immediate pain.

This is a significant dimension beyond looking up movie times or complaining about late food delivery. When it comes to getting things done, we are playing on a new dimension of business solutions.

8 things to keep in mind when building a conversational business UI

We’ve been spending time here in the SAP Innovation Lab understanding what conversational UI means to small and midsize businesses (SMBs) as they conduct day-to-day business. What we’ve found has been very revealing. With the right mindset comes the potential to redefine chat-based interactions. Here are some of the key points to keep in mind when building solutions designed to help businesses operate:

  1. Business vocabulary is already outside the natural sphere of conversation. Buzzword and industry-specific jargon are interspersed among seemingly general sentences, making them particularly difficult to extract. Common, abstract nouns like “opportunity” or “refund” often become concrete or proper nouns to the user. They refer to a specific type of business object and perhaps a specific instance of an object. Expect to define a dictionary of business actions and build well-defined intents across them to accurately detect sentences like, “I have an opportunity.”
  1. Once the user’s intent is understood, magic can happen. Typical business interfaces involve lists, forms, mandatory fields, default values, and dependencies. With a conversational interface, all of this must be maintained entirely behind the scenes. Assume that missing information can be requested as part of the rolling interaction. In this way, users are not overwhelmed with clutter and are instead getting precisely their intended job done.
  1. Businesses are complex. Trying to handle all possibilities in a single thread can be both difficult to implement and potentially dogmatic to use. In the Lab we are building a family of ERP bots that are experts on their own domain and act as concierges to their service; creating opportunities, handling customer records, or summarizing information. By openly talking to each other within the UI, they provide handover in a more natural-feeling way than typical interfaces. As humans, we are actually quite comfortable with talking with different people about different topics.
  1. In business, interactions are rarely discreet. A huge milestone in chatbot development is nested conversation threads. A prompt to the user such as: “Tell me the customer’s name” may very likely be answered with a question: “Who are my customers?” As tasks become complex and potentially lengthy, define a strategy for dealing with interruptions. Do this either by visibly parking the existing thread, handing it over to a new agent, or indicating that a delay has happened. Messages in the log like “cancel,” “get me out,” and “forget it” are common indicators that users are getting lost and want to start over.
  1. Success is a very specific outcome and failure is not an option. This is business. Users have to get their job done, and inputting a sales order has no margin of error. They don’t have consumer-style flexibility to try a different provider or come back later. Business interfaces are particularly downward-sticky. Once an action fails, trust is broken and spreadsheets are opened. Ensure that the interface provides commitment, confirmation, and feedback when performing actions like creating and modifying records.
  1. Extensibility and underlying complexity cannot be ignored. As a general rule, the conversational UI should be able to handle the same degree of customization as the standard UI, whether that means user-specific metadata or client-specific dependencies such as mandatory fields or tax calculations. It is okay to narrow the scope and push/refer to existing methods if they are too complex to handle right now. It is not okay to ignore them and essentially throttle back your business software to a rudimentary version for the sake of adding a chatbot.
  1. Start with zero UI. The most common denominator for integration is pure text. While platforms may build on top of this with UI elements and widgets, there is no guarantee they can be directly transposed between platforms. Thinking text-only also ensures the basics of conversation are covered. The final solution may never be 100% text/voice (we believe it is more likely that smart speakers will converge with displays before too long), but starting this way ensures the fundamentals of conversation are covered.
  1. Natural language is synonymous with intelligent interactions. If the interface is human-like, then the brain should be human-like, too. This is one of the biggest areas where conventional chatbots fail to satisfy because we have such high expectations going in. In business, this becomes a prerequisite of business intelligence, whether that be smart-handling stock replenishment, identifying churn-risk support tickets, or simply automatically applying discounts. It may mean rethinking the underlying business-logic in your solution.

If it works, conversational business UI will become a seamless part of day-to-day life without even acknowledging the Turing Test. Users will become closer to your products than ever, benefiting from being better understood on their own terms. This is a new realm of chatbots which, if done right, will trigger many other questions and solutions which might just redefine the way business happens.

For more on the positives and pitfalls of AI in business, see Teaching Machines Right from Wrong.


Drew Bates

About Drew Bates

Drew Bates is responsible for SMB innovation. He writes from SAP Labs in Shanghai China on the topic of lessons learned whilst being on the forefront of modern technology.