Chatbots for Travel Brands

Get in touch

From Conversational Interface To Conversational Intelligence

A maturity model framework for great customer experiences.

When Facebook opened up its messaging platform, Facebook Messenger, to developers in 2016, conversational interfaces such as automated agents and chatbots became mainstream. Almost immediately, a wave of press releases were issued, heralding a succession of travel and airline chatbots for Messenger. There was a land grab to be the first – what it did, the value it brought and the rationale behind it was secondary. When something is only weeks old, there is no roadmap or best practice to refer to.

Chatbot Mobile Messaging

Since then, consumers and businesses have rapidly taken to messaging. As of January 2019, Facebook Messenger had 1.4 billion monthly active users, with WeChat coming in at just over one billion, according to Statista. In an enterprise context, every month Facebook Messenger handles ten billion messages between people and businesses.

The rate of adoption in so short a time frame means that the maturity model for conversational intelligence and automated agents is still developing. A maturity model which identifies three stages is an effective way for travel brands to gauge where they are today and where they can go tomorrow. It can serve as a de facto road-map, a framework for growth offering a way to progress from one stage to the next.

These three stages are: interaction, intelligence and integration.

Roadmap for conversational intelligence

Roadmap conversational intelligence in messaging and chatbots

‘Interaction’ is the simplest aim for a conversational interface , effectively automating the FAQ (Frequesntly Asked Questions) page of an airline site but with the ability to pass the conversation over to a human agent if necessary. This is where the term ‘automated agent’ acts as a very powerful descriptor of conversational AI capabilities at this entry level.

‘Intelligence’ involves connecting the chatbot to some airline systems, but also ramps up the AI and natural language processing (NLP) so that the chatbot can understand or infer what the customer wants and respond accordingly.

airline automated agents messaging

‘Integration’ is the most advanced. The challenge is interpreting the intent of a message before connecting in real-time to live business-critical systems to access the appropriate response. The response then needs to be returned to the traveller, instantly, in a way that allows the conversation to continue or an action to be completed by reconnecting back to the airline’s back end.

Each of the three maturity stages – interaction, intelligence and integration – has not only specific benefits for airlines but also different consumer propositions, and there are degrees of maturity within each of the stages. The cost to travel brands is reflected in the amount of work required but at this stage it is important to note that even the simplest chatbot can deliver value from day one.

Chatbot interaction

At this first stage of the maturity model, the conversational interface will support basic conversations and only understands predefined sentences. It sources answers from a database of FAQs, in turn pulled from the website or contact centre request sheets. There is an element of natural language processing involved in converting a query into something that is compatible with how the answers are structured in the database.

A typical question which can be answered by an entry-level automated agent or chatbot include “what time does check-in open?” The interface will have been programmed to understand the different ways this question can be asked. The content of the response is standard, as it is sourced from the FAQs from the website. The interface can also hand over the conversation to a contact centre agent when it does not fully understand the question or when the response required falls outside the pre-defined parameters.

Chatbot interaction messaging

However, there are scales of compliance within each of the maturity stages. While the FAQ responses are static, a simple chat chatbot can also access dynamic information, pulling data from some live airline systems outside the FAQ.

A question such as “is my flight on time?” is relatively straightforward to answer if the chat chatbot is told the flight number and can access live information. The chatbot will not, at this stage, need to connect with the entire departure control system for the airline – it can be developed on a need-to-know basis.

This level of maturity can be achieved within a short timeframe, even this simple automated agent can add value from day one. Using the automated agent to answer FAQs enables agents to handle value-adding revenue-generating requests which reflects their expertise and their value to their employers.

chatbot AI automation

A key takeaway here is that chatbots or automated agents should not be seen as a way to reduce the number of live agents. Instead they can contribute to a more engaged workforce while also helping airlines establish a better relationship with customers.

Indeed, research by Hubspot found that 40% of humans don’t care if they are dealt with by a human agent or automated agent – what matters is that they get the help they need. Similarly, Facebook also found that 53% of people are more likely to shop with businesses they can message.

Messaging Intelligence

This is the stage at which automated agents and chatbots can handle complicated queries, moving beyond a predefined FAQ script and becoming context aware. The response system is better able to understand the intent of the entire conversation as the chatbot can access a wider range of natural language processing techniques which combine to establish what the end-user wants.

An ’intelligent’ conversational interface will also connect with a selection of other IT systems, depending on the specific use cases identified for this leg of its conversational intelligence journey.

chatbot intelligence inference intent with NLP and AI

A typical question for an ’intelligent’ conversational interface would be “I want to sit in an exit row”. The intent of the question is clear. In order to get to where the end-user wants to go requires access to the PNR (so that the original booking can be accessed and updated) the ancillary merchandising platform, booking engine and seatmap (so that any available exit seats can be identified and priced). The payment can be taken via a link sent to the end-user.

The transition from ‘interactive’ to ‘intelligent’ brings many new layers of complexity into the equation. The type of complexity is driven by the nature of questions that are asked.

Finding the right natural language tool, or combination of tools, is the building block for an ’intelligent’ conversational interface because without understanding the intent of a query it is impossible to respond. Different open source natural language processing tools on the market have specific capabilities, so travel brands and their conversational interface development partners need to decide between a full solution or mix and match.

Having decided on a natural language processing tool, or combination of tools, you need to train the conversational interface to be, well, ’intelligent’! Industry-specific models are available and can improve the quality of the conversation. The ability for developers to build chatbots that learn from previous interactions and improve over time the the most important learning to come out of artificial intelligence – and it is at the core of why travel brands are getting behind artificial intelligence (AI).

The lesson here is that the technology, the business use cases and consumer usage patterns need to be continuously evaluated and adjusted.

Most natural language processing programmes have artificial intelligence capabilities built in, which means the language capabilities of the system improve the more it is used. Queries that cannot be related to an intent can be identified and translated into something which makes sense. This can be done by artificial intelligence (AI) or manually, where human adjustments to the model are more effective.

intelligent conversational interface for airlines and travel brands

The transactional possibilities of an ’intelligent’ conversational interface creates value propositions for revenue. Customers can have private conversations, in natural language, with the ’intelligent’ conversational interface helping them through the check-in/manage my booking flows, allowing the automated agent to upsell or cross-sell additional services. Before the check-in window opens, an ’intelligent’ conversational interface can proactively contact the user with specific messages – ‘Online check-in opens in 24 hours’ – or with targeted up-sells – ‘secure your seat’.

By helping users check-in, an ’intelligent’ conversational interface can be part of a frictionless and fast customer experience. With people encouraged to check-in online, there is less footfall at the airport terminal desks, ensuring that staff on-site are prioritising customers with the most complicated and time-sensitive queries.

Messaging Integration

The next stage of the conversational intelligence maturity model is known as ‘integration’. This is the most advanced phase where personalisation becomes possible. For an ’integrated’ conversational interface, the response system needs to access and assimilate a lot information about the end-user.

These insights can be held within the CRM system and/or the loyalty progamme, and include previous interactions to form a basis for historical analysis. Some of the data can be manually analysed and fed into the conversational interface training model, while some fields are immediately absorbed by the machine learning modules and become part of the decision-making process.

The conversational interface system is fed with information from the natural language processing tools, various content systems and the current and historic CRM data. With AI running in the background, the more data that comes in, the better the answers the chatbot can provide to users. Permission-based access to third party data sources can also contribute to the conversational interface being able to personalise the experience a customer.

Integration in action - chatbot booking crm

One challenge here is that the artificial intelligence algorithms and machine learning business rules need to be decided in advance, by humans. Finding a way to combine all sources of information is only part of the difficulty: you also need another layer of involvement to determine how much autonomy the ’integrated’ conversational interface can have when dealing with end-users.

The typical question for an ’integrated’ conversational interface might be along the lines of “I am flying to London tomorrow but want to change my seat.” Many systems need to work together for this to happen, and it is worth remembering that not so long ago this simple request would have been fulfilled manually by the passenger calling an agent. Today’s chatbots are already performing sophisticated tasks better than humans, reducing the workload for staff and making the pre-flight experience better for customers.

As automated agents get more sophisticated, there are many more business use cases worth considering. But the more sophisticated the interface, the wider the scope of the integration work. It is worth repeating that conversational interfaces are not a single deployment on a job sheet – they require commitment from a number of departments. On an ongoing basis, the machine learning outcomes need to be monitored and adjusted, new data sets considered.

Multichannel helpdesk cross platform
Cross platform messaging to continue the conversation

Case studies:
Chatbot implementation in action

One mantra of all travel retailers is that brands need to be where the customers are and currently, customers are on messaging platforms. OpenJaw are aware of 35+ airlines around the world which are using conversational interfaces and automated agents to help with customer enquiries.

But adoption is set to rise: SITA’s latest Air Transport IT Insights found that 85% of execs questioned said that they were planning to have AI-driven virtual assistants and conversational interfaces in use by 2022.

The problem:

customer support chatbot messaging

OpenJaw has a number of in-market implementations of its conversational interface platform, OpenJaw t-Social Working with TravelSky. OpenJaw developed an automated agent for China’s largest airline IT business. TravelSky is a distribution network of more than 70,000 sales terminals across 8,000 Chinese travel agencies linking directly to global distribution systems (GDS) with access to the inventory of nearly 140 commercial airlines.

The brief was for an AI-powered conversational interface solution for their travel agency call centre that provided an agent escalation system to respond to common travel related questions.

Like most airline contact centres, only a small number of customer service requests involve higher involvement actions such as cancellation or rebooking. Most requests are straightforward – ticket deletion, changing reservation, printing itinerary and boarding passes and refund problems.

The solution:

Building a conversational interface in Chinese to handle the low-touch queries freed up the agents to work on the labour-intensive, difficult queries. OpenJaw t-Social was able to automate standard requests and FAQs, leading to a huge productivity improvement on a standard call. Every call got an answer, using a conversational interface that automatically generated the answer using sophisticated algorithms and integrations with other systems.

The correct answer to customer queries came from understanding – or ‘teaching’ the automated agent to understand the intent of inbound customer messages direct to TravelSky, as well as contacts made through social channels. With machine learning built in to the automated agent, the quality of the replies is improving exponentially over time.

The result:

The result has been a huge increase in the speed of finding the right information, immediately. Second, as the OpenJaw t-Social automated agent gets smarter – thanks to AI, more and more conversations will be managed by it. Third, OpenJaw t-Social will allow Travelsky GDS to reduce the load on their agents, freeing time to focus on important or complex requests.

Your Conversational Interface Checklist

Chatbot Support Checklist

Coming up with a maturity model for a relatively immature concept such as messaging is not easy. The interaction/intelligence/integration phases are a workable construct and airlines can use this to benchmark where their ambitions lie.

Nevertheless, this maturity model exists alongside other considerations which need to be thought through before deciding to go live with even the simplest interface.

Conversational interfaces are only a few years old, so there are no established ways of doing things. Legacy-minded carriers who see this as a problem need to look at how creative forward-thinking airlines are approaching the theory and practice of implementation.

Here are five “checklists” to work through before making the leap into the world of conversational interfaces, chatbots and automated agents. They cover most of the topics any travel brand will have to deal with once they have made a decision to roll out any form of conversational interface.

As always, start with your strategic objectives:

  • What will be distinctive about the experience?
  • Why am I doing this?
  • Who are the stakeholders?
  • Who are my customer personas?
  • What is my brand personality?
  • What are the timelines?
  • What resources are available?

Next, ask what are your business objectives?

  • How will the bot improve customer experience?
  • Do my customers need another channel?
  • How can the bot improve the efficiency and value of our contact centre?
  • How can the bot help retain and engage higher value customer?
  • What benefits are there to using the bot to share information for customers?
  • Am I basing my bot strategy on the most up-to-date and reliable data
  • How deep do I want to integrate the bot?

Next, start to define the scope of your conversational interface: you need to find out how you can deliver to create a great customer experience?

  • How do I get the permissions needed to talk to the customer?
  • How do I decide the timing frequency and tone of the messages?
  • How do I guarantee that the bot responds quickly and accurately?
  • How do I guarantee the accuracy of the information the bot is accessing?
  • How do I build a bot that feels personal to the individual traveller?
  • How can my bot help position my airline as a retailer?

Every conversational interface needs to be trained. You now need to consider the type of personality your brand represents, the sort of questions your customers ask, and how to create conversation flows:

  • What is the persona and personality that I wish to convey?
  • How do I collect the questions that are typical of my customer queries?
  • How do I create the answers to match the questions in the right tone and ‘on brand’
  • How do I tailor the answer for the user so that they get a very clear action?
  • How do I create conversation flows that are natural and engaging?

Finally, you need to consider what internal resources you need to roll out your conversational interface:

  • How are the marketing and social media teams going to build a conversational interface that is on-brand?
  • Do I need specialist content writers to create the conversational flow?
  • How do I get customer support teams to engage with the project so that the interface has the right answers to the most asked questions?
  • Who will monitor the results and what if any KPIs do I need?
  • If we do not have cognitive engineering skills in house, who will oversee finding, appointing and liaising with the partner?
  • How can I get other staff with IT skills to engage with the project and in what context?
  • Who will have ownership of the entire project or do we need a different reporting process?

Conclusion

This article highlights how carriers can approach the new world of conversational interfaces. Interaction, intelligence and integration is the approach we have decided to share. It is based on our experience of talking to airlines around the globe and applies to early adopters and late arrivals.

The checklist of checklists serves as a handy cut-out-and-keep guide for the sort of questions airlines should be asking internally within teams, across the organisation and with third parties and partners.

Conversational interfaces are part of how billions of people live their online lives. Airlines need to see them as another way to engage with customers, just as the internet was new 20 years ago, as smartphones were ten years ago, as social media was five years ago. The big difference now is that new tech is adopted more quickly than in the past, so airlines need to make their first move soon.

business travel support messaging AI

Register for the Global Travel Summit 2019