Dreamforce 2018 – KLM customer service takes off with AI
- Summary:
- A crisis that grounded hundreds of KLM flights led to the airline making a greater use of social media. It now believes that you have to be where your customers are. But it also needed a helping hand from AI.
Back in April 2010, the skies over much of Europe fell silent. The hundreds of planes that normally criss-crossed overhead each day were all grounded, unable to take off due to a huge ash cloud that had descended over the area, the fallout from an eruption of the Icelandic volcano Eyjafjallajökull.
The ash cloud lasted from 14 - 20 April, and saw about 20 countries close their airspace and around 10 million travellers affected. For Dutch airline KLM, it had a more positive and long-standing impact, as it was the event that directly triggered its big push into social media.
During those six days, KLM couldn’t reach the tens of thousands of its passengers who were having problems, stranded at various airports and locations around the world. Staff noticed that people were posting up messages and questions on Facebook to get information, and soon the whole team were on the social network, contacting customers to offer advice and information.
The ash cloud experience resulted in KLM investing in Salesforce Service Cloud to offer a better social media customer service experience, rolling it out in five days, using its out-of-box integration with Facebook and Twitter.
The airline’s social media operation has expanded since that initial foray eight years ago. KLM now has 300 dedicated agents, who only handle social media cases not calls, and 30 staff in the social media development team. The firm gets around 180,000 messages a week in nine languages through platforms like Facebook and Twitter, and it answers all of them.
KLM has basically followed its customers, according to Joost Oremus, Product Owner Social Media Development at KLM Royal Dutch Airlines. The world is now dominated by big tech companies, like WhatsApp, Facebook and WeChat, with their huge audiences and billions of monthly users; in comparison, Oremus noted, the KLM website has an annual audience of 142 million.
We figured out that if you want to talk to your customers, you have to be where your customers are and apparently they're living in this big world around us. We decided to add the functionality to work these platforms to service our customers.
KLM started to offer functions like booking confirmation, tracking information, check-in, boarding pass and flight status updates on social channels. Currently around 10,000 of these automated messages are sent out on a daily basis. 40 percent of boarding passes now go through social, along with 50 percent of booking confirmations, and 60 percent of all flight and airport updates.
Last year, KLM added WhatsApp to its customer service channels, which has led to a huge spike in messages over chat from under 100,000 to more than 200,000 each week. Oremus said:
There’s a challenge, as when you bring this functionality to your customers, you need to be ready for this.
The other side to this is that customer expectations are getting higher and higher. You may be remembering the days that they said, okay, we will reply to your email in a week and now what we see is, they’re getting very angry if they have to wait more than 20 minutes. So we’re opening up these the channels and customers are getting services, but on the other hand there are rising expectations and rising volumes.
A bit of assistance
So how can firms cope with these heightened volumes and customer expectations? Artificial intelligence is the answer, according to KLM.
You need humans and personal touch for your interactions with the customer. But we also saw that with the rising volumes, it's very hard to do that with humans alone, so you need technology to assist us and handle the volumes and give speed and correctness. So now we’re making the leap towards AI. We really try to leverage AI to help a customer better.
KLM takes a dual approach to its use of AI, using it to give ‘super powers’ to its human agents, and then ‘bots for speed’ to directly communicate with customers. The firm is using DigitalGenius for the staff superpowers element, which provides AI for Salesforce.
KLM has been running its AI customer service technology for two years now, using it to answer certain basic questions with templated answers via a bot, and also prompting agents with possible replies. The AI has got more intelligent as the agents have used it, learning from their responses, so now when it fills in a field with an auto-reply, the agents can have a high confidence level that this would be the right answer.
Learnings
However, KLM’s agents have had their own learning to do to get the most out of the system. Agents can sometimes mess up the AI learnings, in cases where they see a correct prompt but then decide to edit it anyway, as the AI then thinks it was incorrect.
KLM also had to educate its agents on when to use the tool or not. It found that its more experienced agents, who had built up their own list of template answers stored in a Word document on their desktop, were already using this kind of functionality. Rather than carry on filling in replies using that document, the firm asked those agents to share those templates so the answers could be added to the system and would therefore start automatically appearing in the text fields.
With agents who were new to the company, it was the opposite problem – they needed to be slowed down on turning to the AI for everything. They were overusing the tool, even where there was still a need for the human touch.
Oremus was keen to note that the introduction of AI has not been about replacing its human agents, more about helping to deal with the increase in volume and difficulties around customer service complaints. He cited the example of a ‘traffic controller’ working in a KLM contact center back in 2016, whose job it was to go through the list of all customer queries, find the urgent or sensitive cases and route them to agents.
With AI classification, this task is done automatically, and the controller is now helping out in more difficult cases and making sure the workforce is distributed properly.
His job got much more interesting by bringing this technology to his work. I think that’s a great example of how AI can get rid of tasks that are not really meaningful.
Oremus finished by sharing some key learnings from KLM’s AI project:
- Respect the conversation - you’re in a conversation with your customer, you really need to do this well, don’t bring something that doesn’t work. This is the challenge but if you invest in it, it really pans out well.
- Be where your customers are - don’t force them into a channel that you’d prefer
- Use the best of humans and the best of technology - the human aspect is so important, and also in making AI better because the AI learns from the human
- Give your agents AI super powers - you need to help them out in dealing with these volumes
- Make smart choices with technology – you don’t have to innovate everything from the ground up