Kiwi.com thought a chatbot powered by artificial intelligence would help cut customer service costs, but it wasn’t that easy
Travel can surprise and delight, but it can also stress people out. Flights are cancelled, delayed or overbooked, bags can go missing and thanks to bad weather or unexpected circumstances, for example, they may be unable to travel. At EyeforTravel, we do hope that you have got away this festive season without any of the above happening to you!
In scenarios like this, as Kiwi.com knows only too well, these requests from people are often complicated. So while the data-driven Czech born OTA has been able to automate most things with 90-95% accuracy, experience has shown the human touch is still needed in many different situations.
Speaking at EyeforTravel’s recent Amsterdam conference, Kiwi.com Chief Data and Automation Officer Martin Ratoliska explained that the company, which launched in 2012 and has grown into a business employing 1,500 people in 10 countries, processes a huge amount of data. To put this in perspective he is talking:
· 400+ data sources
· 10,5 terabytes of flights
· 50 million searches
· 65 billion potential combinations
…and with an average API response time of 0.3 seconds
As a tech-driven firm, Kiwi.com does everything from machine learning to dynamic yielding and analysis of sales in real time. But on one thing Ratoliskaremains clear: “Until we have artificial intelligence that is so good that it can calm our customers down, we’d rather support our agents in doing their work more efficiently”.
Until we have artificial intelligence that is so good that it can calm our customers down, we’d rather support our agents in doing their work more efficiently
The company learnt this lesson the hard way, by thinking it would be easy to create a chatbot that lower the high-cost of internal customer support. “We thought it would be easy,” says Ratoliska, “but the reality was very different”.
It’s different because travelling is a physical activity and queries can be complicated, especially when travel plans blow up! At times like this, people are “desperately looking for help, comfort and hope and they want it ASAP!”
A chatbot at this point in time simply cannot deliver. So, Kiwi.com decided to approach things slightly differently. It has developed an AI-driven chatbot that is integrated into the interface used by the human agents, who are tasked with calming down disgruntled customers.
This means that the agent is able to choose from the best AI-driven answers, or if the circumstances require it they can continue chatting human-to-human. Every time an agent chooses something from the AI-led list of responses, there is an adjustment, which the chatbot learns from.
A range of tools
An AI-fuelled chatbot integrated into the agent’s interface is just one way that Kiwi.com is helping its agents. It has also developed other tools to help agents do their jobs better:
· Banana Plugin - Often agents have to go to the airline’s website to find, filter and enter passenger details. Kiwi.com has automated a function that shows agents passenger details from its own database. This automatically checks whether agents are entering the correct data against what is in the database. If an agent enters the wrong date, for example, they are alerted to this.
· Flanger – or flight changer, is a tool that allows agents to quickly offer an alternative flight when something goes wrong. From Kiwi.com’s database agents are able to compare the pricing from available flights and process a new journey in just one call.
· Voidatron – When a customer calls to get a refund or void a ticket, the whole process – including the deadline period for voiding a flight, and what the refundable amount is, is automated.
· What bot – With this function, agents no longer have to remember all the IATA codes of airports, airlines and so on. Instead they just enter, for example: /what easyJet; /what Travelport and so on.
· Portalo – with this function, Kiwi.com has created a whole in-house interface for agents where it provides information related to a specific case. For example, if one of the flights from a reservation wasn’t processed, based on the existing logs an agent can knows if a price is still correct.
April 2018, San Francisco