This is a contributed piece from Scott Crawford, VP of Product Management at Expedia
They’ve been calling it ‘the year of the chatbot’ and 2016 saw an unprecedented level of interest in conversational applications designed to help consumers. But chatbots aren’t exactly new; the concept was first explored by MIT computer scientist Joseph Weizenbaum, who created a program called ELIZA in 1964. The bot parsed inputted words and replied with responses that would often rephrase the user’s statement as a question, but didn’t truly ‘understand’ what it was being told.
Recent strides in AI technologies like neural networking and natural language processing have moved chatbots beyond the realm of scientific speculation, and there are now thousands of consumer facing bots. The big tech firms like Facebook, Google, Apple and Amazon have all launched their own offerings, with applications ranging from helping consumers buy groceries to select new tracks on Spotify.
Despite the meteoric rise of this new tech, it is still very early days for brands trying to determine how chatbots can be applied to their sectors. At Expedia, we view product innovation through the lens of solving a particular consumer problem or need, and we believe chatbots have the potential to be transformative for both booking and customer experience.
Travel: the ultimate chatbot challenge
Travel isn’t the first sector you’d think chatbots would be best suited to: purchasing holiday packages, flights or hotels is invariably highly complex. Travel purchases have a long ‘funnel’, with consumers at the top unsure of the details of their trip and potentially looking for flights or hotels in various destinations. Users further down the funnel might be business travellers or regular users who have an idea of where they’re going, but need to work out the details, while users at the end of the funnel are ready to part with their money.
Customers at each stage of this funnel have very different requirements, and chatbot technology needs to adapt to reflect this. If a customer is in the early stages of the funnel, without basic information such as target destination or time-frame, a straightforward booking chatbot will not be very helpful. It’s important to remember that chatbots are still machines and their outputs are limited to the data and information available to them. Put simply, chatbots still can’t help a customer in the research phase of their trip. The good news is that rapid advances in technologies, like natural language programming, are making it progressively easier for chatbots to handle these kinds of open-ended queries.
But for today, chatbots become exponentially more useful the further down the funnel the customer has progressed. New companies such as Mezi, KimKim and Pana have sprung up to help facilitate the booking process for flights, hotels and entertainment: once the customer has initial information and search terms it becomes much easier for chatbot technology to help by automating key steps in the booking process. Some of the big players in the travel sector are branching out into chatbots too, with Expedia and Uber both launching Facebook-based chatbots this year to gather booking information and hail cabs respectively.
What can the travel sector teach us?
Product managers and technical leads need to work more closely to map out their customer funnel and really understand at which stages chatbots could be helpful. Despite the huge technical advances that we’ve seen in AI technologies over the past few years, current chatbots thrive in environments where the task is clearly defined, such as searching according to clear parameters, or requesting specific information from users.
At Expedia, our Facebook chatbot is specifically designed to prompt the user to input relevant data points in order to complete a search on our platform. Clearly defining what role the chatbot will have in the booking process is key to retaining the user’s attention. Businesses need to start with thinking about a specific customer problem they want to solve: once this problem is thoroughly defined, it will be far easier to design a chatbot around it that effectively delivers a solution. Too many companies start this process in the reverse, with the result being a general-purpose chatbot that delivers sub-par solutions to many different queries.
We’re also acutely aware of the risks associated with deploying a chatbot too early in its development. Error-prone chatbots can easily create negative brand experiences. Many people are becoming increasingly frustrated with the dreaded “sorry, I didn’t understand that”, and it’s easy to see how this can quickly impact how people view a brand. Businesses need to be clear with their customers about the purpose of the chatbot: if users understand what the chatbot is for and understand the commands and sentence structures that the chatbot will be able to understand, a productive resolution to the problem is more likely.
Treat chatbots the same way you would treat your customer-facing employees: if they’re not ready for prime time, then keep them away from the customer.
What lies ahead?
For now, humans remain a key part of the purchase process for many industries. It’s hard to imagine chatbots replacing human salespeople in sectors like marketing, IT or retail. However, we’ve already seen rapid advances in natural language processing, which have enabled computers to understand human conversations more than we have ever thought possible. The key now is to build on what’s already been achieved: there are already over 11,000 chatbots on Facebook Messenger, and the more experience data that is available to developers and data scientists, the smarter tomorrow’s bots will become.
PREVIOUS ARTICLE«Rant: Superb Twitter merits public service status
Mark Chillingworth on IT leadership
Phil Muncaster reports on China and beyond
Kathryn Cave looks at the big trends in tech