The advent of AI (artificial intelligence) in human-machine conversation, along with the proliferation of robust digital channels, has greatly advanced the potential for high-quality automated interactions between the enterprise and consumer. This market opportunity has brought in a number of startups as well as big AI platform players, to collaborate with legacy companies in addressing the challenges of conversational commerce.
While AI helps with the challenges of understanding user intent in customer-facing applications, the construction of these applications to solve specific use cases has remained code-heavy and expensive. This is a problem that many players are seeking to solve with code generation approaches aka no-code or low-code tools. Within this blog, I will discuss the challenges and opportunities in this critical area but first let’s set the stage with a bit of background.
Where we started
It is instructive to take a look at the history of interactive applications for digital and voice customer service, as these applications sought to replace live conversation with agents. Interactive touch-tone telephone applications (IVRs) came out in the 80's and were enhanced in the 90's with speech recognition to improve and expand automation. In close proximity, the internet boom drove the explosive growth of digital websites which are still the dominant public face of an enterprise, providing, among other things, support for customer service transactions.
In the 2010's, we saw the advent of mobile touch and tap applications, followed by device manufacturers’ embrace of speech recognition and Natural Language Understanding (NLU) such as Siri, Alexa, etc., which lifted consumer expectations for the power of NLU. Now we are seeing the explosive growth of chatbots, to close the circle back to customer conversations leveraging natural language AI. Since these interactive applications collectively number now in the billions, we naturally ask if the efficiency of creating these applications has improved over the decades, in particular “have low-code and no-code tools reduced substantially the expertise and effort required to build such applications?”
Fast forward 25 years
Here, twenty-five years after websites were launched, extensive coding is still required for website development and only simple templatized websites can be created completely with no-code tools. And despite the massive growth in mobile applications over the past 10-15 years, production-grade mobile apps leverage some widget libraries but still demand extensive coding. While this seems consistent with our experience that comprehensive code generation of general-purpose programs from software flow models has not been successful to date, it unfortunately also bounds how efficiently interactive websites and mobile applications can be built while still meeting the business’s goals. Fortunately, there are indications that, due to the advances in AI, the subset of interactive applications that are conversation-oriented admit code generation, albeit with some limitations.
Taking a mixed-initiative approach to customer experience
The first factor in human-machine interfaces is where the initiative lies. For the typical website and customer-facing mobile application, the visual format and breadth of functionality places initiative squarely in the hands of the user, which in turn opens up an enormous number of navigational code paths that are difficult to describe or generate. On the other hand, for conversational applications, a mixed-initiative approach can be taken that allows the system to take the initiative in deterministic areas of the dialog (such as transferring funds between accounts) while allowing the user to take the initiative via AI in other open-ended areas (“How can we help you today?”).
The "system initiated" segments of the conversation have the potential to be described by a flow-graph specification that adequately covers the much smaller number of code paths. Meanwhile, the "user initiated" segments benefit from the advances in machine-learned intent understanding that can flatten a huge number of conversational pathways into a smaller number of intents expressed in natural language, thereby reducing the need for code-based navigation of those pathways. It therefore becomes realistic to build these applications through a combination of code generation and AI.
The other point of leverage for conversational automation applications derives from their roots in, and technological alignment with, the human-based contact center. For the most complex interactions, this application has the opportunity to fall through to a live agent. This opportunity is simply not available to websites and mobile applications, which are, and have always been, designed for completely automated user interaction without a means to integrate with live agents.
The above analysis indicates the feasibility of code generation approaches for interactive conversational applications. There are, however, some landmines which can mangle the machine-generated user experience. Here are three examples:
The evolution of modern digital conversations. The way users communicate has moved well beyond text-in and text-out. They now consist of an ever-evolving combination of visual, textual and haptic artifacts and formats - emoji’s, images, videos, audio, and so on. The modern consumer is familiar with these formats, and values them, thanks to their heavy use in social media. The code generator must be skilled in automatically leveraging these formats in the conversation, failing which one can expect text-heavy and laborious conversations to be generated, with lowered customer satisfaction and poorer transaction completion.
The effective use of AI and NLU. It is common in many AI approaches to rely on only the proximate text inputs from the user in order to understand the user’s intent and the entities that determine the next step in the conversation. This approach limits accurate understanding of the user’s intent, and thereby limits automation of conversational pathways. To address this issue, it is important for the system to capture broader application and user context and for the AI component to utilize these effectively.
The transition of interactions between automated applications and live agents. These types of interactions typically result in a jarring user experience due to poor integration between the two systems and this frequently undermines the value proposition of automation. This points to a larger issue and thereby an opportunity. Both during the conversation and outside, integration and collaboration between the automation software and the contact center software can lead to a smooth user experience and automation that is continuously learning from the contact center outcomes and therefore increasing in effectiveness. Conversely weak integration between the two sides results in poor customer satisfaction.
Delivering interactive enterprise applications of the future
Consumer expectations are growing at a significant clip which requires enterprises to have the flexibility and agility to adjust to those expectations. However, the staggering amount of code and developer time required to build websites and mobile apps, as well as make changes continues to be a significant obstacle for enterprises to overcome.
By focusing on integrating the right set of tools backed by the contact center, enterprises can overcome the burden of developing sophisticated code. For conversational applications, this means utilizing code generation for digital conversations with machine-learned understanding models and AI. By combining these technologies, barriers will be removed, delivering more transparency related to the customer journey and how to connect live agents to that experience.