Designing Customer service platform in the middle of a crisis

01 Quick intro about Kiwi.com

Little bit of the context

Kiwi.com is a travel technology company and we are combining more than 800 different air and ground carriers into unique itineraries other search engines simply can’t see. So you can get from any A to any B in the world.

We process 100 million searches daily, sell 40,000 seats per day (pre-pandemic), and have more than 200 active distribution partnerships. Supporting this amount of workflow interaction every day is quite challenging. For this reason, we has built an in-house helpdesk platform. You can imagine it as the Jira, but very customized to our use-cases.

Once the pandemic has struck the whole travel industry had to adapt to new increased volumes of customers’ contacts.  I would like to tell you how we've changed our product & customer support experience during one of the biggest travel crises of the modern era.

Without any longer ado, let's jump 🦘into the details and see how the design team played a key role in the digital transformation.

02 Challenge

What would you do if...

In the normal situation if something went wrong, or if you needed some support, you simply sent the email to Kiwi.com or called the customer support team. This system worked. We had several contact centers around the world. The service was reliable no matter your current location or time. There was always someone available and ready to help.

But in early 2020 the crisis suddenly came and the majority of our contact centers, including centers in the Philippines and India, got closed.

What would you do if you had 7 times more contacts, but only 30% of the customer support workforce would be available because of the lockdown? How would you deal with thousands of emails every day with no clue whether it is spam or an urgent customer request? How would you help your customers if you knew that your hotline was busy, because of thousands of repetitive contacts caused by the 80% flight cancellation rate at the beginning of the pandemic?

03 Ideation

How we designed solution?

We are a product-led organization and our goal was to build an internal hybrid helpdesk platform to reduce the pressure on our support agents. With this solution in place we would be able to serve our customers again which means:

• We would be able to reduce the strain on our phone is driven customer support system.
• Prioritize communication-based on multiple aspects and customer situations.
• Reduce the wait time from days to hours or minutes.
• Mitigate the repetitive contact by showing the request status or related information.
• Completely eliminate the spam and bring transparency to our long contact backlog.

How have we done it? I've got a new message from one of Kiwi.com product managers — "Hey, stop everything you are working on and check this RFC doc". All right, I've messaged my design colleague and we've immediately started quick design ideation. Back then, we were shifting tooling from Sketch to Figma, luckily. So We opened Figma and started jamming on the workflow.

I've put together user flow for the internal part and Jardi came up with the customer-facing flow. We did it in two hours and then quickly reviewed it with the Leading product manager. The next morning, we scheduled the review with the whole feature squad composed of people from three tribes.

Connecting the customer

With the team we've decided to simplify the flow to focus on the core — send messages & respond to the customer. Then we immediately started design the interface. In the meantime, the development team started the Front-end and Back-end preparation to speed up the delivery.

What was the flow? In the nutshell, when a customer got a problem, he /she went to the mobile or web app and open the detail of the trip where was the option to send us a message. Then the system prioritizes the queue based on multiple factors. Then the Customer support agent reviews the customer request. If the agent needed more information he simply sends the question to the customer or just send a reply, that everything was solved and then he changed the status either to done or to waiting on a customer.

After that we have created a prototype, reviewed it again with the whole team and then we've run the moderated + unmoderated usability testing. Despite the minor usability issues we spotted two important findings. Our agents expected that the tools would provide a chat experience, which wasn't technically possible back then. We changed the UI that more corresponded with the Non-real time communication. And we also added the information about the average response time to set clear expectations with customers.

05 Delivery of language-agnostic platform

Parlez-vous français? Oui..., ¿Hablas español? También...

After shipping the design, we have carefully watched one of the most important development and release we've done for dKiwi.com digitalization and we were finally able to talk with our customers again.🤞

Wait, that is not all :-) So far we've relied on native speakers to translate customer inquiries. Our next priority was to build language-agnostic communication. We didn't have native-speaking agents during the lockdown who could handle the Japanese, Chinese, Portuguese and other languages. Finding, onboarding and training new native speakers would require six more months which we didn’t have.

So we've built the language-agnostic helpdesk platform together with Unbabel.com that worked almost in real-time. How? Now customers can send us messages in their native language — for example in Russian. This message is immediately translated to English for our agent by Unbabel's machine translation. Once our agent responds, the message goes to the machine translation again. But additionally, it is also double-checked by a human editor and delivered to the customer in a split dozen minutes. This ensures the native quality of the translation and keeps the cultural nuances.🇩🇪 🇨🇳 🇺🇸 🇫🇷

By the way, this hybrid solution was also requested by Kiwi.com because normally Unbabel.com translates both (customer and agent) messages by AI and by a human editor. We saved a lot of time with this behavior change. Implementing this language agnostic solution took us together with Unbabel.com approximately one month and they told us that they've never experienced such a quick and impactful implementation of their product.

We are constantly iterating on it. Recently we changed from reactive to proactive communication, we have implemented NLP to improve the sentiment analysis and provided valuable insights. And there is much more.

What next?

It has begun the foundation for the digitalization of Customer support in Kiwi.com and since the initial release, we have done many improvements to increase the communication quality and eliminate the mistakes agent did. But most importantly we implemented the NLP technology so now we are able to measure the customer sentiment and adjust the communication or the whole communication strategy based on that.

We also know what are the real contact reasons and react accordingly to eliminate the root causes. We are able to show or filter only relevant information the support agent cares about. That means if you are solving customer requests related to the baggage then you gonna have a special environment just for this task and you gonna see the communication relevant just to this task. This focus has a very positive effect on the information cognitive load and in the end also on the request handling time and agent satisfaction.

Another improvement we've prepared is the ability to send an email right from the helpdesk. Agents have to communicate often with partners and other airlines so having everything in one place connected to the customer request makes the platform even more powerful and moreover, this eliminated mistakes caused by lost emails.