Learn more about our exclusive online community for CX Leaders at top corporate brands.
LEARN WITH US
Browse our member blogs for just a taste of what’s to offer inside our community.
Are you an end-user CX Leader? Request to join our community for exclusive access to the best minds in the business.
Explore participating service providers’ emerging technologies, industry research, and underlying approach to CX solutions.
Browse blog articles packed with the latest ideas and inspiration to raise your CX game.
Access our growing library of downloadable materials typically gated to the general public (online community members only).
by Execs In The Know
Using this approach, next-generation, conversational, effective self-service can be launched in days, not months — which can either immediately complement a brand’s current-state automation/chatbot investments or replace them.
Artificial intelligence (AI) has undoubtedly advanced rapidly and transformed the way we both live and work. For years, AI has listened to our commands, directed us to customer support agents, and answered or resolved our basic queries. Chatbots can now handle both simple and complex requests, commonly taking the place of agents in conversations that would have previously required human interaction.
In recent years, the growth of conversational AI has been fueled by the emergence of messaging applications and the development of AI technologies, such as natural language processing (NLP) and machine learning (ML). Heightened consumer expectations are being set by voice assistants like Alexa, Siri, and Google Assistant. This is the technology consumers are interacting with every single day, and they are coming to expect this same type of rapid response and interaction when they communicate with businesses. Given this, solutions such as conversational AI will soon be less of a luxury and more of a requirement for every brand’s customer engagement strategy.
According to most projections, conversational commerce is poised for significant growth. However, there is still a major gap that exists between what businesses are currently offering their customers and what consumers prefer when it comes to conversational AI and self-service options.
So, what does the future hold for conversational AI? Let’s take a look.
A growing trend of conversational commerce has allowed brands to communicate more directly with customers at all stages of the purchasing journey and provide them with a more personalized experience. Some examples of conversational commerce in action include a website or messaging app-based chatbot, a voice assistant, or automated text messages. Nonetheless, current digital customer experiences and tactics are not delivering as anticipated.
In fact, according to our 2022 CX Leaders Trends & Insights Consumer Edition report, survey respondents exhibited slightly lower satisfaction levels for their most recent self-help experience, as well as higher dissatisfaction. “Very Satisfied” and “Satisfied” totaled out at the lowest rate
(44%) since 2019 (39%). At the end of the day, it’s clear that the adoption and increased use of self-help solutions hinge on ease of use and, perhaps more important, effectiveness. When customers visit a website to do a search for a product or service, there is often a high failure rate, leaving them sifting through various headlines and links to figure out if the right result was presented.
The introduction of chat and chatbots has not solved this issue, with many customers expecting to have a negative experience when they know a chatbot is involved. Conversational approaches of “I can help you with things like this. Tell me what seems to be the best fit” leave customers with much to be desired. The technology has offered big leaps in experience improvement, but also has big limitations.
“These previous generation intent-based models leverage natural language understanding (NLU) and are capable of doing a few things well. But as you expand on the vast number of requests from customers, it takes a long time to add new competencies and eventually, the models begin to conflict with themselves. This requires more engineers to rebalance efforts and lead times get longer,” says Jay Wolcott, Co-Founder and CEO at Knowbl.
“At this point, what we’ve been seeing is a lot of sophisticated brands recognizing plateauing containment rates and expanding agent escalation needs. We see a very expensive solution that has reached capacity and doesn’t appear to be the self-service dream everyone envisioned.”
These inconveniences have pushed consumers to want to get things done on their own online. Self-service is no longer a “nice to have.” It’s a necessity to provide positive CX. In our July 2022 Self-Help Solutions Consumer Research report, 22% of respondents indicated they were first-time users of the self-help solution most recently used. Even more interesting were the ramifications of first-time use, with new users far more likely to poorly rate the convenience and ease of a solution compared to more experienced users. If brands want to increase the likelihood that their customers find their self-help solutions to be convenient and easy to use, encouraging widespread adoption, as well as educating about such solutions, is essential. Above all else, a user’s first experience with the self-help solution must be exceptional.
Cue the transformer models and the new era of conversational commerce, which have arrived with an approach that has tremendous potential to drastically change the perception of virtual agents altogether.
Transformers continue to be the focus of ample research and innovation. Recent NLP models such as OpenAI’s GPT-3 and foundational models like Google’s BERT have pushed landmark developments in the accuracy, performance, and usability of numerous natural language tasks. These tasks include understanding text, answering questions, text generation, and performing sentiment analysis. In a call center, transitioning from previous-generation NLU technology to transformer models will be akin to comparing an infant with no language understanding to an adult with comprehensive language development.
“With previous technology, every brand was starting with a blank AI model that had no understanding of English or whichever language they were building for,” explains Dr. Parker Hill, Co-Founder and Chief Technology Officer at Knowbl. “Understandably, this was an extremely expensive and time-consuming process.”
When it comes to large language models built with transformer technology, brands will only need to teach the model their domain-specific information rather than an entire language.
“We’ve seen self-service via conversational AI reach much smaller brands, including those without technical teams, and have seen opportunities for large brands to build experiences that can outperform other support channels in cost as well as improvement in experience,” adds Dr. Hill. “Overall, transformers should be a first step in moving conversational self-service like chatbots from being one of those most dreaded forms of self-service to one of the most prominent.”
The team at Knowbl has had significant experience building and deploying virtual assistants for some of the world’s largest brands, though there came a point when they realized things weren’t going to end well with the current framework. So, they sought to explore new AI and ML models that could improve the breadth of self-service capabilities, while reducing the level of effort it took to build and maintain a quality experience and be agile enough to quickly react to a changing business.
The company’s core focus became building a virtual assistant platform that could provide speed, ease, and scalability. What they’ve invented is a scalable solution that can equally emulate human-level conversation. This will change the way Knowbl does business forever.
For brands that can replicate an in-person experience in a digital environment, it will allow them to rethink websites, mobile apps, and all these other interfaces that have been bound by the rules of play since they first originated.
How will this “Super Agent” allow for more natural and free-flowing conversations as compared to existing chatbots? During a conversation with Dr. Hill, he illuminated us by explaining the two main components.
“To build agents with robust natural conversations, you need conversational flows and the intelligence to traverse these flows. Using previous-generation technology, building conversational flows was an entirely manual process where the developer had to imagine all of the paths through the conversation and then implement these paths with AI building blocks. With this new tech, the AI understands general language well enough that it can infer the possible conversational flows between various topic areas (i.e., automatically retaining and utilizing conversational context). In addition to mapping out conversational flows, the AI building blocks employed to realize these conversational flows were a massive lift to implement. In the past, for each turn of the conversation, you would have to collect a data set of potentially hundreds of training utterances demonstrating what a user might say to go down that path of the conversation. So, even if a team managed to map out thousands of potential conversational flows, it was impractical to implement it due to the diminishing returns on the user experience versus the intractable level of effort.”
With this new technology, so little data is required for the AI to learn the domain-specific conversational flows that brands can leverage the terminal states (the content and responses for potential answers) to robustly train the context-specific understanding.
“Considering the pressure of increasing contact center volumes, depending on the specific intents, this approach can conservatively shift 7-10% of current contact center volume to self-service almost immediately, and deliver significantly decreased agent handle times
(AHT), which can significantly drive down operational costs, while simultaneously improving overall CX and revenue,” says Wolcott.
One of the most thrilling aspects of few-shot learning solutions, such as transformers, is that it removes most of the technical burden and barriers from the downstream brands wielding the technology. Previous technology that featured intents with hundreds of training samples required careful optimization and curation to achieve acceptable accuracy. Engineers and data scientists would have to work closely with businesses to convert domain expertise into something the AI could understand.
According to Dr. Hill, with this new technology, brands can now teach the AI their domain-specific content and leave their conversational AI vendor with the technical challenge of optimizing a general-purpose, pre-trained language model. For most companies, extensive learning will only be needed by those wanting to push the limits of what the AI can do. Most brands can likely get away with very limited internal technical AI support.
We’ve had high school interns build experiences in hours with the new tech that brands found to be better than what their in-house engineers have developed for months using previous generation technology, says Dr. Hill.
For a lot of the most common queries, this new generation of tech can emulate human-level conversations. It allows them to state their needs and questions in the way they want, or they would with a human agent, no longer forcing them down a path of waiting, seeking, or interpreting an option that might make sense for their needs.
“Along with this ease of expression, they now have an option to self-serve for the fastest speed of answer we ever thought was imaginable,” Wolcott says. “This type of approach has tremendous potential to drastically change the perception of virtual agents, almost conversely having them seek out brands that have this type of conversational commerce capability and becoming irritated when they experience brands that don’t offer it.”
On the other hand, conversational AI still has a way to go before it’ll be on par with human-level understanding and response generation for arbitrary user input. However, this new technology gives us a big step forward in the right direction.
“For example, here at Knowbl, we can leverage documentation and FAQs to automatically build a virtual concierge configured to always respond with the language that the brand provides (i.e., avoiding compliance challenges and uncontrollability around customer support),” added Dr. Hill. “Since a large fraction of user inquiries are repetitive and simply require looking up a relevant answer, we’re able to contain most of the interactions. For the interactions that require a deep understanding of the domain or logical reasoning, the AI can hand the conversation off to a human agent.”
Like other investments, embracing innovation is a matter of understanding what is needed to commit, the risks of the commitment, and the potential rewards. Do you want to improve customer experience? Lessen the service burden on employees? Grow revenue by reducing customer churn? Increase customer engagement and loyalty?
For most brands, risks make the most sense in their specific competencies. The most practical approach to embracing innovation in conversational AI would be to find a cutting-edge, highly flexible product or partner. And if such a solution doesn’t exist yet, “then be on the lookout for new players in the space,” advises Dr. Hill. “Until recently, the off-the-shelf solutions for conversational AI required a massive level of commitment to wield — tens, if not hundreds, of engineers and data scientists.”
With the latest technology, this has changed. Now brands of all sizes have a compelling opportunity to partner with vendors that offer few-shot learning conversational AI technology due to the drastically reduced effort required to bring these experiences to production.
When it comes to making customers happy, communication has always been the core of customer care. Good communication leads to good service, high-quality issue resolutions, and satisfied customers. The innovations and inventions are here; brands just need to take their traditional approaches and begin to pivot.
Think about it this way: companies hire an agent and train them for weeks or months while relying on a knowledge base to find answers and recite them back to consumers. What if you could take all these workflows, training materials, knowledge base contents, and past conversations and upload them into a machine that could immediately and accurately represent that information regardless of how a person asks for it? Furthermore, agents could keep that conversation going so they could ask for, or even predict, what they might need. The transformer model can be used for self-service, agent assistance, and even quality control.
The need and demand for real-time communication with customers are likely here to stay, and conversational commerce can help businesses drive those real-time interactions.
As leaders in CX, we’re going to have to get comfortable with being uncomfortable. It’s time to stop and listen to learn vs. listening to respond. We must start listening to the tech, so we don’t get lost. Is your brand set up to embrace the latest innovations? Many have adopted new methods of technology to help them stand out from competitors and create a more engaging experience for consumers — anywhere and everywhere. By using this approach, next-generation, conversational, effective self-service can be launched in days, not months which can either immediately complement a brand’s current-state automation/chatbot investments or replace them.
Currently, the rate of technology change is accelerating at breakneck speed, the barrier of entry is lowering, and oftentimes the level of technical (no-code) skill is reducing.
These super virtual agents will never replace the need for human escalation completely, but imagine reaching a website that boldly presented just one option: “How can I help you today?”
Go Back to All Articles. Have a story idea? Submit to firstname.lastname@example.org.
Want to get this publication in your inbox? Subscribe here!