The art of being hospitable

Conversational AI in the travel & hospitality industry

Conversational AI chatbots, voice assistants, and interactive voice response (IVR) systems are used widely in the Travel and Hospitality industry. For travelers, they can provide a better customer experience prior to, during, and after a journey. For employees in hospitality, they can help improve the quality of work and getting jobs done.

It all comes down to being hospitable — providing a better experience through conversational AI.

We had the opportunity to discuss conversational AI in travel and hospitality with three industry leaders:

• Manuel Nalda Castro, Sr Digital Innovation Mgr, Aeromexico
• Michael Nelson, Strategic Partnerships, Vonage
• Sherry Comes, Managing Director of Conversational AI, PWC

Watch the video and read highlights below

Meeting the customer where they are

Conversational AI solutions, like chatbots, voice assistants, and IVR enable travel and hospitality businesses to provide a better customer experience — and be more hospitable. As Sherry Comes of PWC indicates, these solutions help with meeting the travelers where they are, and letting them interact on the channels they prefer, 24/7 — whether it is talking, typing, or tapping.

Travel and hospitality businesses have to deploy omnichannel to support a wide range of customers. As Manuel Nalda of Aeromexico explains, anyone can buy a plane ticket — customers come from all different countries and have different ways to communicate. Hospitality companies need to deploy in the channels customers are interacting — and depending on the country or channel, adapt their services. It is about being part of the daily life of the customer.

A challenge today is that some of the experiences may be more multi-channel, rather than omnichannel, given the complexities in back-end integrations. As Michael Nelson of Vonage explains, in some cases the communication is in silos. While an enterprise may use multiple channels, it may not necessarily be a unified communication.

The tools and technologies are available now, however, to consolidate communication so that context can be shared amongst channels. In this way, if a user is escalated from a text SMS notification to a live agent call, the context can stay intact.

One area of particular interest is true multimodal communication — wherein a voice experience guides a user through a web or mobile experience — taking full advantage of a device’s capabilities. For example, imagine calling an airline to change a seat. If the IVR said seat 20D is available, what does that mean? If, however, the IVR sent a seat map to the phone at the same time, one could see seat 20D is an aisle, and perhaps select it right on the phone.

Multimodal solutions can also be used for channel shifting — i.e. starting in one channel and moving to another. Nelson experienced this first-hand on a flight that was cancelled during boarding. When he called the airline, he received the option to chat via text, instead of waiting on hold, and was able to rebook his flight via SMS in a matter of minutes.

Comes refers to this type of multimodal, channel shifting as the “zombie channel.” As a user switches from one channel, that interaction dies, and comes to life again on a new channel.

Channel shifting can also be used to leverage the capabilities of the modality for additional functionality. In the case of Aeromexico, they leverage a web experience to augment WhatsApp when it comes to entering payment information. The context is shared to enable a smooth, seamless transition for the user.

How an enterprise uses multimodal solutions depends on the company and its customers. As Nalda explains, companies need to experiment depending on what their customers are looking for, to see if a mix of channels will work, and if so, enhance their systems to support the interactions.

As Comes adds, being multi-channel is about meeting users where they are and where they want to interact. If a user is driving, they can speak but not type. If they are in a loud place, they can type or tap, but not speak.

These multimodal, omnichannel solutions can be highly effective. In one example, Comes team created an experience on a voice channel that would send users a link to a web form to fill out. The team had found some people just prefer to pick up the phone and call, instead of looking for the form to start. This deflection bot saved over 2,000 calls a day and led to high customer satisfaction.

Improving both customer and employee experiences

When it comes to use cases in travel and hospitality, there are solutions for both consumers and employees.

At Aeromexico, the goal of the conversational AI experience is to be a travel companion — before, during, and after the journey. The team looks for small tasks along the customer journey that can be automated or improved with conversational AI. For example, customers can check-in for a flight via WhatsApp and have their boarding pass in a minute. If after a flight, a customer has an issue with baggage, they can make a claim via the chatbot and quickly receive compensation.

While we often think of the consumer, there are plenty of opportunities to leverage conversational AI solutions for employees too. Comes’ team sees a lot of use cases for employees, behind-the-scenes — including for pilot scheduling, logistics, maintenance, and more. Nelson also sees significant interest in digital assistants to improve the employee experience for pilots and flight attendants.

The employee facing experiences, however, can often be more complex. The exception handling when there is a weather event, or disruption, can be very complicated. Pilots may need to be moved around. A flight may need a new crew. There are a lot of issues that arise. Fortunately, employees can use conversational AI interfaces to help get answers or resolve issues.

In addition to customer and flight crew use cases, there are opportunities to use conversational AI to improve the contact-center agents’ experience as well. One of these key use cases is “agent-assist.” When a customer calls into a contact center, the call can be transcribed in real time with automatic speech recognition (ASR), and run through a knowledge base or chatbot to provide the agent with answers or the “next best action.” Nelson sees a lot of opportunity for this in the travel and hospitality space.

Agent-assist functionality can have a significant impact on contact-center operations. For example, Comes’ team implemented an agent-assist experience that resulted in increased customer satisfaction and increased agent retention. They found callers were no longer yelling at the agents, as the agents were better equipped with answers via agent-assist. Agents were happier so attrition went down; the costs of hiring improved; and training time and costs were reduced.

Context and personalization are key to hospitality

Context and personalization go hand-in-hand with hospitality and effective conversation AI.

The hospitality industry is all about personalization — even to the point of hyper-personalization. Luxury hotels, like the Ritz Carlton, know your preference for pillows, the cookies your kids like, and the events you like to attend. High-end hospitality is about providing a great customer experience.

Similarly, with conversational AI, context and personalization are key. The more information you know about a user, the less information you need to ask, and the more you can customize the interaction. For example, when calling an airline, the IVR greets you and asks if you are calling about the upcoming flight to your next destination city. It knows who you are, and what you are most likely calling about.

There is an opportunity for even more, hyper-personalized experiences before, during, and after an interaction. As Comes points out, for the most part, everyone gets a similar IVR experience. However, if the system knows more about the user before the call occurs, it can provide even more personalization. For example, if someone over 80 calls into the IVR, the system can adjust to speak slower, use gentrified language, speak in a frequency one can more easily hear, and offer repetition, to provide a better user experience. During an experience, the automated solution can also take into account sentiment or tonal analysis and respond appropriately. After the call, all the data and learnings can be applied to the next interaction, to make it better.

Context and personalization can help enable a more natural, human-like experience as well. As Nalda explains, personalization needs to be a guiding principle in an effective, user-experience design-process — taking inputs from linguistics too. The more one uses these guiding principles, the more human-like experiences can be created. The challenge, as Nalda points out, is to avoid having a rigid interaction.

To enable more personalized, contextually aware interactions, it is important to have a single source of truth in terms of customer information. While the technology for personalization is there, the challenge is getting all the unstructured and structured data out of silos, as Nelson indicates. It is helpful to centralize and curate the customer data to enable chatbots, IVR, and voice assistants to all consume the same data, adds Nalda.

The role of generative AI

Large language models (LLMs), ChatGPT, and generative AI are getting quite a bit of press right now. ChatGPT, in particular, is helping drive broader awareness for LLMs and generative AI. It is an exciting time for conversational AI!

There are a lot of opportunities for LLMs and generative AI, in conversational AI. Two areas where some conversational AI platforms and enterprises are experimenting with generative AI and LLMs are in the natural language understanding (NLU) model creation process and intent fallback handling. When building a model, LLMs can be used to generate training phrases for an intent. When it comes to fallback handling, if a user’s utterance does not match an intent, an LLM could be used to “expand” the utterance to additional possibilities to see if those match an intent.

Two interesting areas where Comes is seeing opportunities for generative AI are in validating information and narrowing down the possible intents of the user. In the first case, generative AI can be used to help validate if the information a user submitted is accurate and makes sense — a form of Cloze test. In the second case, generative AI can be used to figure out what the general intent space is, and then NLU can be used to get to the more specific intent from there.

Nalda sees an opportunity to leverage generative AI to enable users to have a more natural way to express themselves. Similar to Comes, he sees generative AI helping with intent detection.

As exciting as generative AI is, it is not quite at the stage for enterprises to use for responses directly, in production. As Nalda points out, we need to be careful how to use the response from generative AI, and not put it directly in front of the customer. Instead, we can take the generated response into consideration as the real response is being designed.

This is particularly important for compliance. For some industries, especially financial services, compliance is a key concern. One of the benefits of automated chatbots and voice assistants is that the solution can only respond how it was programmed, whereas agents can go off script and potentially say the wrong thing.

Nelson sees a great opportunity for generative AI for conversational designers — a cure for “writer’s block.” ChatGPT and other LLMs can help designers make their jobs easier by generating initial content that the designer can edit and fine tune.

There is also a lot of potential with LLMs in synthesizing information to aid in agent-assist solutions, Nelson adds. The technology can be used to synthesize all the knowledge across the organization, in structured and unstructured data, and use that information to provide potential solutions or answers for an agent. As Nelson points out, there is huge attrition in the contact center as it is not an easy job. The technology can be used to help make an agent’s job easier, and enable them to be more productive, and provide better customer service.

Measuring success

While travel and hospitality enterprises are implementing conversational AI solutions to provide a better customer experience, they are also implementing the solutions to achieve business goals.

Enterprises often look at the traditional contact center metrics like first call resolution, cost per contact, and other metrics around escalations. Some of these may not even be that useful — like the average handling time (AHT) as our experts point out.

There is a treasure trove of information though beyond these standard metrics — including understanding the reasons why customers are reaching out. For example, is there an underlying issue with the product or service?

For Nalda at Aeromexico, one of the key metrics is “jobs done.” Was the process successfully completed? Did the customer receive an answer?

In addition, customer satisfaction (CSAT) is also quite important for Aeromexico as it focuses more on the customer experience. As Nelson adds, anytime there is a positive experience with a customer, it increases brand loyalty and potentially increases the lifetime value of the customer. There is an opportunity to turn customer service interactions into sales opportunities. As Nelson explains, if you hear something or understand the context, you could turn that into an opportunity.

It is important to put conversational AI metrics in perspective. Comes raised an interesting point when it comes to metrics: we hold virtual assistants to a much higher standard than humans. As she explained, in a typical bank, agents in a branch may get half of the info wrong, but when we talk to a virtual assistant we are expecting 100% accuracy. We are measuring virtual assistants with different measuring sticks than we measure humans. People are critical of generative AI, when humans are probably a lot worse, Comes adds.

The future of Conversational AI

When it comes to the future of conversational AI, our panelists see a bright future.

Nelson sees the underlying technologies, including ASR, text-to-speech (TTS), and LLMs, continuing to improve and get better and more human-like. The adoption of multimodal solutions and digital humans will continue to grow. New opportunities will emerge too — like “model as a service.” For example, if you have a lot of data in a particular industry, like travel, you may be able to monetize and sell that.

Nalda sees the future of conversational AI evolving from rigid interactions to more natural ones, with more empathy and personality. He also looks forward to more proactive, and predictive conversational AI — virtual assistants that anticipate and initiate the interaction.

Comes predicts a major shift from IT and developers to business and designers. In the future, it will not be all about the technology, but how you make technology work for you based on business outcomes. It is about moving away from how we do things, to the why.

Given the advancements in conversational AI, especially in LLMs and generative AI, we can look forward to exciting times ahead!

Arte Merritt is the CMO of Bespoken, a conversational testing platform. He is a frequent author and speaker on conversational AI and data insights. Previously, Arte led conversational AI partnerships for AWS. He also was the founder and CEO of the leading analytics platform for conversational AI, leading the company to 20,000 customers, 90B messages processed, and multiple acquisition offers. Arte is an MIT alum.