CEO Neil Tolley describes our AI Journey with EVA – Our AI Powered Virtual Agent Solution for Hotels
There’s a lot of talk about AI at present, with ChatGPT rightly getting a lot of plaudits (and some concerns) for the simply mind-blowing things it can do. We have already had it help us with some documents and possibly the odd piece of homework for some of our teams’ school age children! Read on to learn how the technology behind ChatGPT is transforming automated hotel guest interactions, boosting satisfaction levels for guests and staff alike.
What is ChatGPT
ChatGPT has been built on the GPT Large Language Model, soaking up billions of lines of code, websites, books, and articles to learn the skills it now has. There are very few companies in the world that have any hands-on experience with a Large Language Model.
Most are in the category of ‘big tech’ – Google, Microsoft etc. Their focus is broad, rightly so. They want to build a large language model to do everything.
What is a Large Language Models?
If you are unsure what a ‘Large Language Model’ (LLM) is, head to the bottom of this post, I asked ChatGPT that very question and have added its answer below.
However, what’s more important to us, is a Large Language Model, that can do 1 thing very well – customer service over the phone. You might be waiting for 1,2 or 5 seconds or longer for a response using chatGPT. That’s an unacceptable latency for human conversation over the phone.
The LLM behind EVA, our AI-powered virtual agent for hotels, has been optimised for real-time speech and has been specifically designed for customer service. It’s why EVA is so amazing.
Customer Led Voice Assistant
EVA is a customer-led voice assistant for hotel guests. You can ask it what you want, when you want and how you want to – you can even ask multiple questions at the same and EVA will manage that,
You will have spoken to voice assistants that are bot led and undoubtedly been asked a series of questions, one at a time, which can be slow and frustrating. This is because bot-led agents, including our first-generation EVA, use Google Dialogflow as the engine behind it. Dialogflow searches for keywords to hang on and is very linear in how it operates and can only handle a single request at a time.
LLM v’s DialogFlow
EVA understands your requests much better than say an agent built using DialogFlow. In our tests, EVA understood simple requests 28% better than a DialogFlow based agent. A simple request may be, can I get directions or what time is check in.
EVA handles more complicated requests with ease where keyword-based agents simply cannot handle them at all.
An example could be ‘do you have a table tomorrow night at 8pm for 4 people?’ EVA would understand that request with ease, an alternative AI agent would have to ask (once the requirement for a booking has been established) when would you like to come, what time and how many people. Another example could be ‘do you have a toothbrush and toothpaste I can have’ or ‘could I get a burger, fries and a coke’ standard stuff in any hotel – EVA understands these requests with ease. Just like a human would.
The Evolution of EVA
The evolution of EVA since 2020 has been quite something to see and hear. In the early days during Covid, just the simple service with a more robotic voice but integrated to hotel systems was amazing and gave great results. Not least due to our brilliant AI team, who spend their days not just building agents for new customers but continuously reviewing and tweaking to make sure EVA gets better and better.
Now we have human voice actors in place for multiple brands and in multiple languages, with integrations to a growing number of other systems in the hospitality ecosystem. Hearing the interactions with callers and EVA never gets old as it constantly amazes.
Our story at Fourteen IP since 2020 shows just how far AI has come and we are still in the early days. What will AI be capable of next year or in 5 years and what will hotel guests expect from it? Looking at the market today, I anticipate callers will expect longer and increasingly natural, human-like interactions and to be able to do not just everything they could when speaking to a human but more – Humans don’t API into other systems, yet!
We are already extending EVA to text messaging, to allow us to keep up with those that prefer to let their fingers do the talking.
EVA Deployments
EVA has been deployed by multi-national and regional chains as well as individual properties and ranging in scale from super-economy to luxury to resorts both small and (really) huge – the results are pretty much the same across every property type and across multiple languages. For those of you still unsure about AI answering calls or dealing with guests via any channel, I get you. Some of them drive me mad and more so now, I know how good they can be. EVA handles millions of calls now both internal and external. and we have not had a single guest complaint – not one.
EVA lets the staff hotels have, focus on giving great face-to-face service or spending more time on the phone with those guests that need it. We have also deployed EVA to provide helpdesk support, accessing a knowledge base of known problems for a Point-of-Sale upgrades, providing assistance and creating tickets in ServiceNow when needed.
EVA Integrations
When you see the work that goes in to making the experience for the guest as great as possible, the integrations with 3rd party systems like HOTSOS, Sevenrooms, Salesforce, Jazzware and numerous others, you can see how, done properly. AI powered systems can make a big difference to the guest experience and help front of house teams focus on giving amazing face-to-face service to their guests.
I can’t speak for others of course but I can confirm that our AI is not going to attempt world domination any time soon. It is however, bent on dominating the hospitality AI space and I for one wouldn’t bet against it.
What is a Large Language Model?
Answer from ChatGPT
A large language model is a type of artificial intelligence (AI) model that is trained on vast amounts of textual data to generate human-like language responses. These models use deep learning algorithms to process and understand natural language patterns and structure.
These models are able to generate text that resembles human-written language in terms of coherence, grammar, and meaning. They can be trained to perform a variety of language-related tasks such as text classification, question answering, machine translation, and more.
Some examples of large language models include GPT-3, BERT, and XLNet, which have been used in a variety of applications, from chatbots to language translation tools.