By Stuart Lauchlan

April 4, 2023

It goes without saying that conversational AI is one of the hottest topics in tech at the moment, but the abiding question that accompanies that celebrity/notoriety - delete according to prejudice! -  is where are the practical use cases of the technology in practice? 

One such exemplar can be found at loveholidays, the UK’s fastest growing travel agent, whose website boasts a claimed 500 billion potential holiday packages on offer. The firm has around 3 million customers and has ambitious growth plans to expand internationally and build on its status as the UK’s third largest travel firm.

Speaking at the recent European Contact Centre and Customer Services Exchange event, Eugene Neale, Director of Business IT and CX Engineering at loveholidays, explained: 

We sell travel, but fundamentally, we're a technology business. So, we've invested a lot of time and thought into building this platform technology that we can scale to multiple markets, that we can take the business and effortlessly open in different regions. We could also scale it to different products. We're choosing to sell travel at the moment, but we can add more products in the future. Conversational AI for us for the last few years has actually been a fundamental part of that strategy, a platform business that we can repeat.

The company deployed a conversational AI bot called Sandy, using the Dialog Flow tool within Google's Contact Center AI solution, initially to handle five percent of customer queries. But such has been the success of the technology in action, it’s now handling half of all contacts. Neale made no secret of his assessment of the potential of conversational AI as an “amazing cutting edge technology that actually really does serve customers in a meaningful way”. 

Sandy is now nearly three years old, he noted, and has changed the way loveholidays interacts with its customers: 

Sandy is the first point of contact for everybody who speaks to us via the live chat channels. This year that'll be about 2.1 million contacts we'll put through Sandy. We can now successfully process about 70% of all those conversations. So 70% of every conversation started with us will start and end in conversational AI with Sandy. You won't need to go on to human agent, which is fantastic.

Customer experience 

In terms of CSAT [Customer Satisfaction Score]  numbers, Neale points to a “very respectable” 55 score as a travel company coming off the back of the COVID pandemic;

We're not investing in conversational AI at the expense of customers, and the customer experience. That's always been key to us. It's been part of our vision. I actually wanted to use conversational AI to serve, to get to people super quickly. With a wait time of 96 milliseconds - which is nothing,  less than a second -  and an average handling time to resolution of less than a minute, that feels like incredibly good customer service.

He added: 

As you build conversational AI in a business, you start on the easiest topics and you leave the harder and harder topics to the human agents, and you leave the emotive topic to human agents as well. So as we kind of advance our conversational AI,  we are changing the dynamic in our contact center for what the agents  need to deal with. Sandy herself doesn't actually come at any cost to the customer in terms of a poorer experience; if anything it's better.

Customer feedback certainly appears to back that up as Neale noted: 

We get hundreds, if not thousands, a week of comments from our customers. We ask directly for CSAT after every interaction. I personally find it amazing, and I think it's kind of a testament to where we are with technology, that people want to give verbatim feedback to a chatbot.

Doing something right

So what is it that loveholidays has done right with Sandy that so many other organizations have not? As Neale observed: 

A lot of businesses have tried building chat bots over the years, and chat bots don't have the best of reputation. I remember having conversations with various different bodies about the fact that chat bots are actually 'the most evil thing in customer service' not that long ago. It's because they were very generic years ago. 

A number of conversational AI projects have also failed because they haven’t been integrated fully into the business, he added: 

The whole business needs to pivot to support having an AI strategy. That's very key, but I think it's also about the intent of contact centers, who've used lots of tools over the years to deflect or, in chatbots, you call it containment, which is not letting customers out. I think if that's your intention, you're going to frustrate customers. And I think we've had that era up till now, because actually fulfilment was hard, resolution was hard, but it's got easier. So I think it's a mixture of intent, and also priorities in businesses. Hopefully now we're starting to see that change.

Trust

It’s still the case that there are a lot of chat bots out there that ask the customer their name and look like a pre-contact form. They’re almost “orphaned away from the whole experience”, suggested Neale. That was where loveholidays started from, but then evolved into a more conversational style:

We're not just conversational, we're conversational with a lot of personal detail about the person, the customer. So, we greet them by name. We talk about the specifics of their product.  And this is not just reflected in the greeting. Anytime we talk to a customer, we talk to them about the particulars of their flight, the exact name of their hotel - we don't talk generically about a hotel - and that is huge. It changes the interaction with the customer from being some generic best fit answers, which don't really drive confidence in customers, to the answer being given is exactly about me. So, it's the correct answer. So, the confidence of the customer goes up dramatically. Really, that's why we've hit the 70% conversion rate or resolution rate for people, because people have faith that this feels like a really engaging process for them.

Sandy is also getting past the vexed question of consumer trust. Neale argued: 

I hope we all know, the Large Language Models can't be trusted. They actually make stuff up, that's what they're doing. They're generative. They're making a pattern that you think feels like the right answer, but unless you asked it to be seated in fact, it won't be fact.  But actually with conversational AI, and using the contact center, it will be effective. We're not generating the answers, we're giving them out. The customers take them as that. So when Sandy talks to somebody, you can see it in interactions - people take it as the voice of loveholidays that is 100% accurate. They trust Sandy as well. It's not trying to get off the contract with somebody. There's no bias, there's no fake reassurance. People really take it as a concrete fact.

Underpinning all this is data and that has meant hard work for loveholidays, as Neale recalled: 

We spent two-and-a-half years changing our structure of our data so it was available in very near real-time for our conversations. You can't wait for data during the conversation, that doesn't make it work. Actually structuring our data systems in such a way that they allow these personal conversations has been a huge effort. You're going to need a combination things - charismatic conversation, great data, but you also need technology.

The tech in this case was Google’s Data Flow CX, which has been key to the success of Sandy: 

We can now have a conversation where we can remember the beginning of the conversation and the middle and use it the end of the conversation to change context, to actually change the response type for different people and avoid that kind of traditional criticism of a chatbot which just loops and says the same thing over and over again. We just don't need to do that anymore. We can learn as the conversation goes. A stateful conversation is effectively like having a conversation with someone with a short term memory and from interactions with bots, you know that that hasn't felt like the case with other bots.

In terms of ROI, Sandy is also paying its way, with a cost per contact of around 2.3 pence, which has mounted up over the past year to around £1.7 million of savings. But conversational AI has also been turned into a profit center, said Neale: 

As we developed our customer service capability with conversational AI, we realized people wanted more than just help with problems, they wanted to add to their holidays...So we experimented with this kind of idea of, could we turn a conversational AI into a profit center, or at least a cost neutral center, with an experiment to see if people buy ancillaries?  It turns out the conversion rates on doing this are very good. But actually the feedback from customers is very positive as well, because we can make it again a very integrated and easy experience for them. So it's a very interesting discovery. We want to build on it more in years to come.

It’s taken about 2.5 years to get to this stage, but Sandy’s capabilities continue to expand, with the bot now being tasked to handle and resolve its first voice calls. Neale explained: 

Voice is a very different channel to doing it in chat, you don't have the same kind of guardrails, but we expect to get up over time to very similar handling rates.

At the same time the tech has evolved, so to have consumer attitudes, with the pandemic driving people more towards a digital-first life, reckoned Neale, boosted by again by the rise of ChatGPT over the past six months: 

Now people actually actively want to seem to want to seek out machines to talk to. ChatGPT was the fastest growing product, fastest rolled out product, in the history of products. If you think about it, all it was was the ability to ask a machine to write silly poetry or share a made-up story with you. That kind of willingness of people to actually talk to computers, talk to machines, is very powerful. This journey wouldn't be possible without consumer behaviors changing.