How EasyJet is scaling to new highs with AI

EasyJet’s digital-first holiday business was up-and-running for three clean months before the pandemic struck, which presented several challenges for the low-cost airline’s digital director Ian Chambers.

As passengers were stranded, flights were grounded and biosecurity started to kick in. The airline was hit by a “tsunami of questions” that EasyJet Holiday’s fledgling customer service centre was not equipped to deal with from a capacity point of view.

“We were also furloughing staff, but we needed to do something fairly quickly to deflect as many inbound queries as possible because we’re not delivering a good service otherwise,” Chambers recalls.

Pre-pandemic, Chambers had been talking to integration partner Sprint Reply, which specialises in the automation of conversational interfaces about the ability to scale customer service without having to use more staff.

But now, a Google-cloud based chatbot using natural language processing was being considered as the best option to answer EasyJet customers’ queries at the scale required.

The initial build took just three weeks, although Chambers added that it helped that the company was already on the Google Stack with Google Tag Manager and had the analytics in place – which meant that the digital team didn’t have to redevelop the entire site.

The customer service team worked up policy docs into Q&As to feed into the bot so that it could sufficiently answer customer queries.

“The challenge was that things were changing every couple of weeks with the traffic light system that the government introduced,” recalls Chambers.

“Each Thursday we would look at Grant Shapps’ twitter feed to see which countries were going to go on Green or Amber – as that influenced what the policies were going to be in those destinations.

“We had to make sure that the chatbot was up to speed. But it also meant that we could answer questions like ‘Can I still travel to Italy?’ – in real time and it was immediately impactful,” he added.

Natural Language Programming

As with many other types of AI, the machine which launched wasn’t as accurate as the machine that it became and continues to become. When it hit its stride, the figures were impressive, with the bot handling 5 million queries with a 99.8% accuracy rate.

“Customers don’t ask questions in the way that you expect. What we learned post-launch is that people are not exactly succinct. This is why the natural language model of tuning and training was critical to what people are looking for from an intent,” Chambers added.

“With NLP you can let your customers be chattier with the bot so that don’t have to narrow them down. It can take a while to train up, but because we were so focused on a certain set of content, the chatbot got fairly accurate quickly and it didn’t take us long to hit that 98.8% match rate so there were very few customers leaving without a resolution to their issues.”

According to Chambers, the chatbot is still being put to effective use following the operational challenges that many airlines have faced since the May half term with flight cancellations and strikes all over Europe.

“You can definitely see the pivot from Covid related stuff to operational related challenges – we’ve had to put more pointed content up on the site and update the chat bot to handle all those fresh intents that were coming through,” he said.

In a single week this June, the bot conducted 6,000 conversations from the airlines and holidays business combined. “Even if half have saved a phone call that’s a great cost saving,” Chambers says.

He adds: “We’re getting to the point now where we can confidently state how much revenue has been driven with people interacting with a chatbot. Because they’ve stuck on a site, they haven’t wandered off to another site or abandoned their journey.”

While Chambers acknowledges that so far the bot has been used for reactive experiences, he says that the digital team is in the process of “flipping it around” to become a more proactive revenue-generating venture.

“We’re looking at other ways to make money: focussing on using the bots to help drive bookings rather than as a post booking service,” he says.

“This may also involve training the AI up to use languages such as French and Italian to enable more rapid growth in new markets,” he added.

Data silos

To power the next generation of chatbots across the organisation and to offer a more personalised level of customer experience, Chambers acknowledges that any data shortcomings across the organisation will need to be addressed.

“We class ourselves as a data-driven airline, but we do still have data silos and we have got an aging backend solution that we’re going to address over the coming years.

“There’s a live project going on now which is looking at how we bring a lot more data together, pairing it better and fundamentally giving access back out again so that we can use that data on multiple platforms.

“To be able to use data to personalise experiences, to supercharge the customer service centre and to power the next generation of chatbots which can recognise booking references, identify customer types, suggest options and take you all the way through the process  – a fully transactional chatbot – that’s the eventual ambition,” Chambers says.

As a holiday company, easyJet is also working on another project using AI to power personalised search results and promotions on site, which is aimed at people in the ‘looking’ stage of their journey.

“We have a tech partner called BD4 which is on our site measuring the intent and signals that customer looking for holidays are giving them.

“We are trying to turn anonymous customer data – we don’t know who you are, but we know the type of customer you are – into more of an AI-driven personalised experience.

“So we are able to look at new customers coming in and say ‘You’re telling me you’re a family customer and that you are looking for a family holiday in the peak of summer in Greece – so these are the kind of hotels that will be perfect for you – and then we manipulate the search results order to show you the best converting hotels for that customer.”

Chambers adds that the tech can also be used to lure back visitors who (have a high propensity to book something but who later abandon their searches) with discount codes and added incentives to complete their journey.

EasyJet is also looking at its CMS to do some more contextualised personalisation for its users so that a city or beach customer, for example, can have content automatically positioned more appropriately for them.

“We can see that personalisation is a good driver to increase our customer satisfaction (CSAT) scores – it’s not something we’ve historically done that well at – so we’ve started on that journey now and Holidays has been able to take the lead on that.”