During the initial months of the Covid-19 outbreak, international travel dropped 70-75% globally in 2020. Travix was buried under a mountain of manual booking cancellations, changes and refund requests piled so high there was no end in sight to the number of backlogged needs our customers experienced during that first year of the Covid-19 outbreak. Our customer service team was confronted with their Goliath.
On a normal basis, our customer service team might have about 1,000 max. refund requests on our backlog. Within months, that backlog grew disproportionately large to over 500,000 requests, a 50,000% increase seemingly overnight. With call center volumes breaking record numbers, the question on everyone’s mind was: “How are we going to provide our customers with ‘the next journey at their fingertips’ now?”
Like David, our loyal agents and the Excellence team started picking up pebbles to attack the giant head on while delicately trying to balance the fact that at the end of this endless stream of requests was our valued customer. We needed to do more to protect the trust our customers had placed in us.
So, while we doubled our efforts towards the manual tasks, our Innovation Managers, Process Director and IT & Data teams got to work creating a scalable solution to position Travix to (hopefully) never have to face another CS Goliath like that again in the future.
It became crystal clear that we needed to aggressively invest in self-service that allowed for the number of immediate needs to be addressed in a timely manner. And this is when the idea of creating a customer service chatbot came into focus.
Here are 4 things we took into consideration when pursuing a digital-first approach to customer service.
Fortunately for Travix, we already had a nice working relationship with a company that supported our internal customer operations platforms. And it just so happened that Freshworks was embarking on a bot building platform that we agreed to pilot.
If this choice hadn’t fallen so gently into our laps, our first major decision would have been about choosing the right chatbot building partner. But with this decision quickly made, we began to wrap our heads around what it took to get into the mind of a bot while having the voice of a human.
It’s probably worth mentioning here that preplanning the content of your desired bot ahead of time will make your choice in deciding which type of bot your company needs much easier. There are several popular choices out there. The natural language chatbot is quite impressive, but requires a lot of development work in advance to create the acquisition of language necessary to be able to recognise and respond to your customers with expertise. This type of bot is based on AI algorithms that actually understands what someone is saying. It learns language like a child. It takes a few examples, makes the necessary connections to learn the meaning and grows its understanding eventually without being programmed.
On the other hand, a rule-based bot operates from a set of rules for the customer’s input. There’s no natural language processing (NLP) being applied here. These bots are pretty simple to configure and they say exactly what they are programmed to say. And it’s clear on the customer’s end what options they have in their interaction with this bot. These bots mostly use buttons as input options for the customer rather than asking open-ended questions that require interpretation from the bot.
Based on several factors (long-standing relationship with current vendor, budget, purpose of the bot, development resources), we followed our vendor’s guide to the ruled-based bot for our first CS chatbot project.
To better understand what our customers needed from us, we pulled data reports that indicated the top 6 reasons they reach out to our call centers. This revealed: cancellations of bookings, changing of bookings, refund status updates, online check-in, travel documents, & luggage specifications for the flight.
Another dataset we explored, which is especially useful if your company has multiple communication channels that customers use for service, was the incoming volume to the call center(s) on those other channels. We drew up a business plan that looked into how much of the current call center volume could be diverted into a chat channel once our chatbot went live. This analysis went a long way towards guiding us to the right agent resources to allocate to servicing our new chatbot conversations.
Doing the research, making predictions and strategizing our chatbot’s initial key performance indicators (KPIs) were a necessary starting point for our chatbot project. We captured essential data, like the number of unique users, the number of transfers to an agent, the transfer rate, the drop-off rate, the conversation statuses, the activation rate and a CSAT rating for both the agent and the chatbot alike, to help steer our chatbot towards success. Understanding the journey of the customer is like putting together the pieces of a puzzle. I hope you love puzzles! This is the part that gives you huge insight into how your customers are receiving your new chatbot. As we go, we keep adding data we find valuable to help us complete the full picture of the customer’s experience with our chatbot.
We also conducted interviews with each member of the CO team and the agents. Here we gathered information on how the team managed each need in practice and consulted our process flow charts to understand how we could communicate accurate and valuable information to our customers.
And because we wanted to adopt the voice of the agent in as much of our chatbot conversations as possible, we pulled conversation samples from our closed tickets and shifted through 1000s of real conversations; the successful ones, the complaints, the mistakes, and the ones where our agents bent over backwards to find solution after solution for our customers.
Based on this research, next we needed to determine what our chatbot would say!
Well, that’s not entirely true. We discovered, while servicing our customers during the pandemic, that no one wants to hear that their flight is cancelled/grounded/changed with no estimated timeline for a refund.
Yet, in the ideal world of chatbots, this is the place to iron out a reliable, accurate, seamless script that your customers will be happy to interact with. This script is a strong voice for your company and leaves a huge impression on your customer. So, circumstances aside, here’s the opportunity to get the conversation right!
One of the big benefits of using a bot is that what the bot says in response to a customer is almost always predictable. Some rules of thumb come to mind here:
So, I’m going to pull back the curtain here a bit. Chatbots are great! And an ideal way to deliver scalable service to your customers. They have come a long way in the past 10 years, especially the impressive NLP chatbots. But for the majority of your customers, what they really want when it comes time to make a decision or tackle a big task is to talk to a human.
It doesn’t matter how articulate your chatbot is or how many solutions it can provide as self-service, unless you spend time identifying transfer points to an agent, a human, customer service reviews will always contain complaints of not being transferred to a live person.
The day will come when this becomes less of an issue, but for now, just trust me, customers are looking for that option. I had no bigger indicator of this than when, in the midst of the pandemic, our beloved chatbot creation received terrible reviews as the travel industry came to a halt, but agents who received the transferred topics received a large proportion of 5-star ratings, even though we couldn’t satisfy the customer’s request as a result of grounded airlines.
Why? Because customers were so relieved to be able to reach a live person with their needs. Now, I’m sure there were other reasons, but this statement made by many of our customers didn’t get missed. So, do give this step serious consideration. At what points will your chatbot transfer the customer to someone live? And how will those service agents interact with your customers?
The chatbot script is, well, scripted. But the conversation with a live human is not. So, to keep a streamlined experience, spend some time thinking through this transfer and how your company would like to end the conversation. In this case, it could be a scripted ending by the chatbot or a live ending. Only one of them you have complete control over.
At the time of this article, our customer service Goliath has been pelted at for over a year now. In collaboration with airlines, we’re down to a que line of under 20k refund requests from 500k left to process. In a year, our Excellence team has tackled work never dreamed of alongside the rhythm of their regular daily tasks. Do our customers know this? I hope so, but our ratings don’t reflect it.
Thankfully our chatbot has made it possible for us to scale up our service to allow for these unusual tasks to continue to be accomplished. But there’s still plenty of work to do.
Next on our radar is opening up agent-assisted service through our chatbot to 24/7. At the moment, we’ve accomplished 24 hrs, 5 days a week on all of our affiliates (42 of them) except for the US. And we’re in the pre-planning stages of developing a chatbot to release on our social channels for quicker service.
But these steps take time and resources. So, while we’re very pleased with the steps we’ve taken to scale our service and offer quicker solutions, we’re also committed to pelting away at another giant: manual processes. Manual processes are a key element that prevent us from providing the service we think our customers deserve.
One thing is certain, none of us at Travix are interested in meeting another Goliath. And while that might have been in our past, we have no intention of it being our future. Our digital and scalable efforts won’t stop here.
Customer Care Innovation and Automation Manager