Dec 6, 2022

20 min read

The Future of Conversational AI

Trends and predictions from the experts

As 2022 comes to a close, I had the opportunity to ask leaders in the conversational AI space for their opinions on the biggest trends of 2022 and predictions for 2023.

The experts included:

Alan Nichol, Co-founder / CTO, Rasa
Andrei Papancea, Co-founder / CEO, NLX
Angie Talley, Practice Manager, Natural Language AI, AWS
Braden Ream, Co-founder / CEO, Voiceflow
Cathy Pearl, Conversation Designer, Google
Hanlin Fang, VP, AI-powered conversation platform, ServiceNow
John Kelvie, CEO and Co-Founder, Bespoken
Laetitia Cailleteau, Global Lead for Conversational AI, Accenture
Maaike Coppens, VP of Design, OpenDialog
Philipp Heltewig, Co-founder / CEO, Cognigy
Richard Smullen, CEO, Pypestream
Sherry Comes, Managing Director, PWC

Trends of 2022

Conversational AI is integral to effective customer service

One of the most common trends the experts saw this past year is how integral conversational AI has become in customer service. As Laetitia Cailleteau of Accenture points out, the adoption of conversational AI is “pervasive” in customer service. Angie Talley of AWS adds that the tight integration of conversational AI with the contact center was one of the biggest trends.

While reducing costs and increasing containment are often cited as key reasons for automated solutions in contact centers, it really comes down to providing a better customer experience. As Andrei Papancea of NLX points out, there has been a shift to focus on customer satisfaction as the top goal of automated customer service, as opposed to just cost cutting. As Maaike Coppens of OpenDialog adds, conversational AI “makes all the difference” in serving customers in an efficient, meaningful way. Hanlin Fang of ServiceNow echoes this sentiment in seeing conversational AI as the “core story” in providing a better customer or employee experience.

As Phillip Heltewig of Cognigy points out, the weakening economy and global pandemic has resulted in enterprises realizing the contact center is one of the most important places for businesses to interact with customers.

The underlying conversational AI technologies continue to improve

The underlying technologies used in conversational AI, including Natural Language Understanding (NLU), Automatic Speech Recognition (ASR), and Text to Speech (TTS) all continue to improve. There has been quite a bit of venture investment in the space in the past year, especially with NLU and ASR services.

The maturity of these underlying services is bringing about better experiences and new opportunities.

As Heltewig points out, the maturity of NLU, ASR, and TTS has enabled enterprises to automate large parts of their customer service, without compromising on quality or customer satisfaction. Fang adds that the maturity of these services, as well as customer requirements for multi-language support, has led to large enterprises implementing third-party and open-source solutions.

The maturity of the underlying services also leads to new ways of leveraging them to optimize the experience. As John Kelvie of Bespoken points out, there is a need for orchestration to leverage the best NLU or ASR for a particular use case or domain.

Along with the maturing of the underlying technologies, Coppens saw significant progress being made in the interoperability of conversational systems — working towards “conversational harmony.”

The maturation of enterprise strategy and need for domain expertise

As conversational AI continues to advance and evolve, the enterprise approach to development has also matured. Many of the experts pointed out the shift in how enterprises now view conversational AI solutions as vital products, as opposed to proof-of-concepts or innovation projects.

Developing effective conversational AI solutions requires domain expertise. Given the free-form nature of communication, it can be challenging to build a solution that not only understands a user, and responds appropriately, but also responds in a way that satisfies the user.

Some enterprises are turning to vendors for this domain expertise. As Richard Smullen of Pypestream indicates, he saw a significant trend in IT teams realizing they are not experts in conversational AI, and thus need the help of vendors to deliver solutions.

Fang adds that “time to value” is also a critical, deciding factor for enterprises in choosing a vendor. A vendor can help reduce the time to market, so enterprises can start seeing returns on their investment sooner.

For those enterprises developing internally themselves, there has also been a maturity in the approach. As Alan Nichol of Rasa describes, the trend is for enterprises to no longer treat conversational AI as one-off projects, but as real products. They are staffing the initiatives with multidisciplinary teams of developers, designers, product managers, and content specialists. He points out they are also following best practices seen in traditional software development including prototyping, testing, and CI/CD deployment processes.

Conversation design is critically important

Conversation design is a critical component of conversational AI development. Designing flows that are intuitive, concise, empathetic, and inclusive — while also leveraging context and personalization — requires skill. It is about the power of words.

As Cailleteau points out, enterprises have learned investing in voice user experience (VUX) and conversation design is essential for success. Talley adds, an increased focus on conversation design as part of the development process is key.

As the founder and CEO of Voiceflow, the leading conversational design platform, Braden Ream sees firsthand the importance of conversation design. The trend he saw this past year is the continued rise of conversation design (CxD) and NLU design as a sub-discipline within CxD.

Multimodal and multichannel solutions are on the rise

Enterprises are increasingly looking for multimodal and multichannel solutions.

While multimodal can refer to the handoff from one channel to another, an example of a true multimodal experience is where an interactive voice response (IVR) solution guides a user through a web or mobile experience. These types of multimodal solutions take full advantage of the device capabilities. Imagine if you were calling an airline to change your seat. If the IVR said seat 20D is available, what does that mean? If, however, the solution were to send the seat map to the device, while the IVR said seat 20D is available, you could see whether it was an aisle, middle, or window to help decide whether to select it.

According to Papancea, NLX is seeing over half of potential customers are looking to leverage a multimodal experience as the first implementation. They are also seeing 90% of customers leveraging two or more channels — often starting with voice and then moving to Whatsapp, web chat, or SMS. Cailleteau is also seeing enterprises starting with multimodal by default as one of the top trends this past year.

The richer user experiences are enabling more advanced use cases. As Smullen points out, the move from basic FAQ style chatbots to more transactional use cases is made possible by these richer user experiences that are more visual, and enable swiping and clicking.

Experimentation with large language models, synthetic voices, and avatars

Our experts are seeing enterprises experiment with newer technologies in the conversational AI space.

One of the more exciting areas is with Large Language Models (LLMs). Cathy Pearl of Google highlighted the trend of enterprises experimenting with LLMs to auto-generate conversation responses. As she points out though, while there were some interesting explorations, we still have a long way to go before this is truly useful.

While LLMs may not be quite ready for production deployable, natural language generation (NLG), there are some interesting applications for them. One area is in building NLU models — using the LLM to generate potential training phrases for an Intent. A related use case is to leverage an LLM to improve fallback handling. For example, if a user utterance does not match an Intent, instead of sending the fallback response, use the LLM to expand the user’s original message to see if any of the generated messages match an Intent.

Coppens also saw a trend in synthetic voice and virtual avatar experimentation. As she points out, these open up new possibilities as brands and users venture into virtual worlds.

Predictions for 2023

The demand for conversational AI will increase, especially in the contact center

Our experts see the demand for conversational AI will continue in 2023.

As Heltewig predicts, conversational AI solutions will be the most sought after feature in contact centers. Enterprises will continue to look for multimodal solutions for consumers as well as AI powered “agent-assist” functionality for contact center agents.

Cailleteau adds, conversational AI will be the new system of engagement for brands — to provide “voice of the customer” insights to enable a 360 degree view of the customer and help drive strategic designs.

Sherry Comes of PWC, takes it a step further in predicting enterprises may no longer give customers an option to do anything other than self-service, in order to stay competitive on price and profit. She predicts this may actually lead to better overall service, by allowing consumers to serve themselves.

Coppens also predicts 2023 as a fertile opportunity for new ground-breaking use cases. As she indicates, periods of economic downtown are often when innovation is at its best. Now is the time for enterprises to invest in automation initiatives with conversational AI to improve the customer experience.

There will be an even greater importance on conversation design in a user-focused approach

As mentioned earlier, enterprises are implementing chatbots, voice assistants, and intelligent IVR to provide a better customer experience.

Our experts predict a further focus on the user and customer experience, through effective conversational design. Pearl predicts a return to focusing on what users really need and want, and not just which technology is the most exciting.

Effective conversation design includes following best practices and taking into consideration empathy, inclusion, and accessibility. Cailleteau predicts a rise in ethical AI design. It is not just for text and voice, but for non-verbal signals as well, she adds.

Nichol also sees a greater importance placed on conversation design, with a greater empathy for end users and more emphasis on discovery, prototyping, and iterating based on user feedback. Comes echoes this with a prediction that the “age of Intent driven analytics, diagnostics, and design” is here. She adds that involving a multidisciplinary team that includes designers, data scientists, and business strategists leveraging data will result in much better experiences. She concludes, the focus will shift more to the people, rather than the tech itself.

Coppens adds, as synthetic media, human-like avatars, and language models continue to evolve, human-like conversations will be an attainable goal, provided there is an increased focus on the conversation design and user experience.

Personalization will increase

Leveraging context and personalization can make for a great user experience. The more information you know about the user, the less questions you have to ask.

Papancea predicts even more personalization in contact center use cases — especially across channels, modalities, and devices. An increase in cross-channel personalization will enable enterprises to automate more, and provide better customer satisfaction.

The good news is we see users are more than willing to share personally identifiable information (PII) in conversational AI solutions. AWS ran a survey of consumers’ feelings towards automated customer service. The survey results indicated 70% of consumers are comfortable with automated solutions using existing PII, and 67% are comfortable providing new PII to an automated solution.

An increase in richer, multimodal, multichannel experiences

We are moving past the days of basic FAQ chatbots to more advanced implementations that are multimodal and incorporate additional AI beyond NLU.

Papancea predicts we will see even more multimodal solutions in the coming year. Smullen also predicts users will seek out experiences that are more visual and immersive as well. These more advanced, multimodal solutions will enable conversational AI experiences to take on even more complex tasks.

Fang also sees a future convergence to omni-channel solutions, with vendors providing a more holistic solution.

We are seeing enterprises implement experiences that incorporate additional AI, such as for document processing or image recognition. For example, there is a car insurance provider that incorporated text and image recognition into their chatbot to process drivers’ licenses and other documents. Instead of typing a name and address, the user can take a photo of their driver’s license in the chat. If they have an existing insurance policy, they can take a photo of it to extract relevant information. These additional AI components have resulted in reducing the sign up process from multiple days to minutes.

Talley predicts even greater integration with additional AI/ML services in the future. The greater integration will enable conversational AI solutions to tap into a wealth of additional information.

The build versus buy debate will continue

The question of whether to build versus buy is an ongoing debate.

Developing conversational AI solutions can be challenging for some enterprises given the free-form nature of communication. It can be hard to predict all the things a user may say, or how they may say them.

There are multiple aspects to developing effective conversational AI solutions — including the use case selection; the NLU model creation with Intents and training phrases; the conversation flow design; the back-end integrations; and the testing, monitoring, and measurement — all in a continuous iterative cycle.

Whether to build versus buy can depend on the skills of the team, the resources available, and the time-to-market goals.

For some enterprises, there is a need for low-code or no-code tools. Fang predicts enterprises will seek low-code solutions, given limited budgets and the ease of deployment.

On the other hand, some enterprises do invest in building out internal conversational AI teams. Ream predicts a move to in-house conversational AI teams and a further consolidation of vendors. Kelvie echoes this move to in-house development, with enterprises implementing “best of breed” architectures grounded on the principle of picking the right tool for the right job. With this, comes the need to leverage data from testing, training, and monitoring to optimize the solution on an ongoing basis.

Related to the “build versus buy” question, is the trend in vendor consolidation. As the space continues to mature, more consolidation and acquisitions may occur. As Comes predicts, there will be a thinning out in the space, given the landscape is saturated with similar tools and platforms.

Conclusion

The demand for conversational AI continues to grow and the underlying technologies continue to mature. As the technologies continue to improve and more enterprises implement conversational AI solutions, it will be interesting to see what the future has in store, and what new use cases and experiences we will begin to see.

In their own words…

Below are the responses from the experts.

Alan Nichol, Co-founder/CTO, Rasa

What were the biggest trends in chatbots and/or voice assistants in 2022?

Increasing maturity in how enterprises develop AI assistants. Rather than treating it as a “project” / one-off initiative, staffing multidisciplinary teams of developers, designers, product, and content specialists, and truly approaching AI assistants as products. That means bringing in good practice from other kinds of software development, all the way from discovery and prototyping, through to user testing, and on to a mature delivery pipeline leveraging version control and CI/CD.