Automated customer service chat

Agoda’s automated customer service chat allows travellers to receive quick, contextual help, while reducing the load on our customer service agents. My focus is on defining experience flows (e.g. a wish to waive a cancellation fee, a special request for a hotel), determining required inputs from customers, and writing the messages we send. More recently, I’ve been working alongside project managers responsible for our agent workflow, integrating our tools to further improve efficiency.


When chat was first introduced as a contact channel for Agoda customers, customers were simply connected to the next available customer service agent. Although this new channel improved our customer satisfaction score, the additional contact with agents meant increased operating costs.


Our data showed that many of the contacts made to Agoda's customer service team are about issues customers can resolve in our customer portal. If we could help customers self-serve through our chat channel, we might be able to reduce costs while further improving customer satisfaction.

We considered training a machine learning model to detect customer intent from free text, but ultimately opted to present customers with a list of options where possible. Aside from allowing us to move faster, this made sense because our emphasis would be on self-service options, which require specific inputs from the customer.

More recently, we've begun to integrate customer chat with our agent-facing products. By partially automating the agent workflow for issues customers are unable to resolve on their own, we expect to reduce the amount of time agents spend on these issues.


I've owned design for our automated chat from its conception, and collaborate with a product manager to define new flows and improve existing ones.

I also write all content for the messages sent to our customers, and work closely with our development team to understand tech limitations and ensure proper implementation.

Some milestones

End-to-end self-service journeys within chat

In the beginning, we shipped quickly by leveraging the existing self-service options in Agoda's customer portal. We gathered what customers needed help with, and provided links to the appropriate features within the portal, such as this cancellation modal below.

Once all available self-service options were supported with links, we improved the chat experience by eliminating the links and building end-to-end journeys for each option, prioritizing by potential savings in the time agents spend on cases. While our customer portal relies on more traditional forms for input, the chat format allows us to guide customers with messaging and instructions, collecting one input at a time.

The example below is of our simplest cancellation flow, when a booking is eligible for free cancellation.

A majority of customers who initiate the automated chat do not end up contacting our customer service team, and customer satisfaction was higher for those who were able to self-serve than those who were eventually connected to an agent.

Verification by booking ID

When our automated chat first launched, it was unavailable to logged-out users because we used only log-in for verification. To reduce friction, we introduced verification with booking ID and email, which is how customers are verified over the phone.

This increased usage of our automated chat significantly, reducing the total time agents spent handling customer issues.

Partial automation of agent workflow

When a customer is connected with a customer service agent, the agent has a specific set of operating procedures to follow. Previously, customers who interacted with our automated chat before connecting to an agent were not distinguished from those who had not. This meant agents were spending time collecting information that was already available, and customers were sometimes asked to provide the same details twice.

To eliminate this redundancy and further reduce the time agents spend handling issues, we're in the process of partially automating our agent workflow by collecting as much information as possible with the automated chat, and having agents skip some of the steps they're otherwise required to go through.

Actions highlighted in the agent journey below will be automated by the chat system, and agents will proceed from the appropriate step in the issue resolution process.

For this, I collaborate with the project team within Agoda's Customer Experience Group (CEG) responsible for guided agent flows. Each step of an agent flow requires specific inputs, and is often written in Agoda-specific jargon. I write our customer-facing prompts, and work with the project team to ensure the input we collect can be mapped back to our agent-facing product.