The latest episode of the Shift Happens podcast welcomes guests Minqi Jiang and Peter Zakin (Hyper Travel founders and now employees at Tradeshift) who discuss their chat-based virtual assistant that helps power Tradeshift Go.
Here is the podcast transcript, edited for clarity and flow:
Christopher Jablonski: Hello and welcome to the Shift Happens Podcast. I’m Christopher Jablonski, your host.
On June 23rd, Tradeshift announced Go, the first virtual assistant for company spending and travel. The company also announced the acquisition of Hyper Travel, a chat-based travel assistant app. Go gives users at mid-sized companies an easier way to buy while providing finance teams with the transparency and control they need for better spend management. Go also comes with a travel booking skill enabled by the acquisition of Hyper Travel.
Here in the studio in not-so-sunny San Francisco, I have with me the two founders of Hyper Travel, Peter Zakin and Minqi Jiang, who are now employees at Tradeshift. Minqi will focus on design and end-user experience at the company, while Peter will be leading operations on the platform side.
Welcome Peter and Minqi, and congratulations on the sale of your company!
Peter Zakin: Thanks Chris, it’s good to be here.
Minqi Jiang: Thanks Chris.
Jablonski: Why don’t you guys provide a brief explanation of how Go works and what makes it different from other virtual assistants.
Jiang: Go is another entrant into this emerging space commonly called bots or chat agents, and there’s been this recent surge in interest among Silicon Valley entrepreneurs and investors in building these conversational chat-based experiences. What makes Go different is that, for one, it’s focused on business-to-business use cases. There’re a few startups that have raised quite a bit of funding around providing these concierge chat-based buying experiences like Operator and a few others around buying retail goods for consumers. Go, in many ways, offers a similar intimate user experience for enterprise users, in the cases where they want to purchase things that are out of their time or expertise resources.
The other interesting thing about Go is that it’s primarily not a chat-based experience. The chat-based side is a layer above of what is “core” Go. And the core layer is a virtual payments solution. Go started as and is in development as a virtual card solution. That itself is pretty unique in terms of all the other virtual card solutions that are out in the market right now. If you look at all the virtual card companies that are providing all of these virtual card provisioning services to enterprises, they’re actually focused on provisioning debit cards.
What is really unique about Tradeshift is that we have these strategic partnerships with pretty major card networks to enable Tradeshift to be one of the first major providers of virtual credit cards, not just debit cards. Credit cards have a slew of advantages over debit cards. You’re extending working capital instead of allocating a portion of it, limiting your existing working capital. You don’t have to open up a new bank account with an issuing bank. You’re earning rewards for each purchase. So, in many ways, the credit card solution is superior and Go is one of the first offerings for virtual credit cards.
On top of this core payments solution is where the concierge and the chat-based agent jumps in. There’s also a virtual assistant that essentially helps you both provision the virtual cards contextually in whatever mode of technology that you are using whether that be a web browser, or mobile or anything else. It also connects and forwards you like an operator to other partners and service providers all in the same chat context to help you to buy goods that are outside of your businesses expertise.
Jablonski: So it sounds like Go will elevate the technology that comes with Hyper, but you’ve already had some traction with the app itself since you launched. What does this mean for current users of the app?
Zakin: Current users of the app have obviously noticed that the skin and actual appearance of the app is now very different. We’ve rebranded the travel app as Tradeshift Go and so today in the app store you can find Tradeshift Go and in the current iteration you have access to what we are calling the travel skill, the first publicly available skill. Customers–existing ones and new ones–who are coming to the app can send messages to agents who help them book and manage their travel. For existing customers, we didn’t sunset any features. For the most part, it is a very similar level of service. The major change they’ve seen is just an alteration of the appearance.
Jablonski: But if they were able to convince, say their CFO to adopt the solution as it is in its current reincarnation, would it be easy to move forward to the virtual card aspect and use it for company spending?
Zakin: Absolutely. As we bring on more partners, we’re going to have our customers use the travel skill and the credit card provisioning skill so that they can more transparently manage spend in their companies.
Jablonski: I noticed in the press release and some of the articles such as VentureBeat, PYMNTS, and Finovate, lots of discussion around machine and human intelligence to provide a useful experience for the businesses that come through the chat interface. Just to be clear for the audience, we are not talking about a chatbot but a virtual assistant, which is a human who is relying on some combination of technologies to be more productive and be more efficient in their responses. What does that look like or how do they fulfill requests?
Zakin: One of the early design decisions that Minqi and I made was that we did not want our customers to have to speak a pseudo language using slash commands or [follow] the way that users today interface with Google. There’s a funny anecdote from a couple of months ago where someone’s mother searched in Google: “please find something…thank you very much.” That was her thinking of Google as a very natural interface, but that does not correspond with the way that most people use services like Google that are essentially just text boxes. There’s a general grammar to the way that we communicate with most software that we are using today.
In the last couple of years, there’s been an expansion of the types of languages that we are using to speak to different services. If you are using Siri, for example (and other natural language interfaces), although it is impressive technology, there’s a limited vocabulary/lexicon of things that you can have Siri do. If what you are requesting falls outside of that vocabulary, then it’s a terrible interface. That’s a long way of saying that today’s bots are good if you know exactly how and what to ask for. But for more expansive use cases, like things that are complicated such as travel, we knew early on that having a bot experience without any human intervention was going to lead to a disappointing experience.
Early on we decided that we wanted our customers to feel that they had a personal travel agent at their disposal. The gains that AI and machine learning provide to technologists are around efficiency. What we see is the opportunity to both provide a human type of experience to customers while also providing the same efficiency gains to our agents. What that means is that we expose our machine learning and AI techniques to the software to our agents, who really are the end users of a lot of those gains. When they are corresponding with customers, the customers have every indication that they are dealing with an actual person who can both understand all of their complicated requests and cares when things go wrong, which is not to be underestimated when dealing with a customer support experience.
On the back end, our agents can work much more efficiently than, say, call center-based agents who are on the phone and can only handle one customer at a time. These agents have the luxury of working in chat, where they can handle multiple conversations at a time, and they also have an extra advantage, which is that they are using AI and machine learning to help them draft responses to their customers. We’re talking about efficiency gains on the level of a few orders of magnitude than what their counterparts in traditional call centers have.
Jablonski: Down the road can a scalability problem potentially arise when you have a lot of users and a limited number of human agents? Do you see a virtual agent or algorithm helping to answer some of the questions or requests?
Zakin: That’s the question everyone has about these human-assisted experiences. On one end of the spectrum, say in the category of travel, there is a far less human expense in apps like Expedia than say apps like Hyper, which have much more human involvement. We’re looking at what we think of us the spectrum and move toward a design where humans are stepping in the use cases that call for humans that leverage the advantages of having a human on the other end. So when there are customer support problems, you really want a human there to help guide you through. There are other situations that I can go into but the way to think about it is that we want humans to be solving problems that humans are really good at solving, and we want user interfaces and conversational UI to be solving problems that can be solved efficiently and don’t require that same level of high-touch human intervention.
Jablonski: You mentioned that travel is the premier skill that we launched with; what other skills do you see on the horizon and how will they help with getting procurement and finance teams more efficient, and more helpful for employees?
Zakin: If we take a step back and think abstractly about what we are offering in our travel skill, the cool thing that we are doing is that we’re connecting customers with experts when they need expertise. Thinking about various business use cases, there’s a tremendous need for expertise. Outside of travel, there are workflows that are common among many businesses–particularly those that are going to be interested in our solution for spend management–there are going to be areas where we can connect customers to experts on demand.
The best way to think about potential partnerships going forward and other skill providers is to think of areas of expertise that are valuable to the types customers that we are dealing with. Whether we are talking about legal, human resources, IT, etc., there are many areas of expertise that we will do a really good job using our platform to support.
The other cool thing we are doing is assisted buying. Our travel agents are there to help customers when they are making purchasing decisions. If you think about the world of spend, we can inject a support agent into flows where a customer is thinking about buying something. There are different categories that customers are spending their money on and I think we will have a corresponding skill for each one.
Jablonski: And the benefits of that are not just increased control and visibility, but there’ll be productivity and perhaps gains in the expertise that will be added to that request. So instead of the employee going out and finding someone on their own to help them, they’d get best in class service through the virtual assistant.
Jiang: Yeah, in general, businesses are incentivized to do what they do best; wherever they have a sustainable competitive advantage. It’s in the best interest of every business to focus on that. What Go is providing as a platform is a way for businesses to connect into Go and freely offer what they do best to other businesses that need those services. It is a natural evolution of Tradeshift’s vision to connect businesses into a network that allows trade and services to be easily exchanged. A conversational user interface is a fluid realization of that. It’s the next step inside the marketplace for all of these business process outsourcing firms that are providing outsourced modules or groups of workers to provide skills and expertise that smaller firms might not necessarily have. For example, if you are Google, Facebook, or a Fortune 500 company, you most likely have a lot of specialized compartments inside your business that do different functions, like booking travel or doing accounting. But smaller companies don’t have the resources or the economies of scale to make that a good investment, so it makes sense for them to turn to outsourced solutions like BPOs. Go is making it even easier for any small firm to just directly access those outsourced services that they might need.
Jablonski: Going back, when you developed Hyper Travel, you probably didn’t foresee that you’d be selling it in roughly six months. In terms of how you view the evolution of your technology, your company, and how it is now folded into Tradeshift, is it what you expected? What is the ultimate outcome for conversational UI and your app?
Jiang: It is a great outcome. We started with attacking travel as the initial use case, viewing it as a great beachhead for providing more natural conversational experiences to all types of purchasing flows and potentially beyond that to accessing information as well. But we started with commerce because that was a good place to build the groundwork to build a business on top of, and we saw travel as the most promising commerce use case because a lot of travel purchasing among wealthier tiers of users and amongst business users are running inside of a delegated or assistant-driven purchase flow.
The technology that we are building; the agent dashboard, the network, dispatching, and logic is really agnostic to the use case. We can apply it to any purchasing flow that’s delegated to another agent that sits in between the buyer and the supplier, and connects the buyer to a greater network and helps to curate the best inventory. This is a terrific outcome that catalyzed our ability to take a big step forward and start working on expanding the scope beyond the initial use case and go to version two of that vision.
Jablonski: Speaking of that vision, when we announced Go, we were in New York City and Christian Lanng, CEO of Tradeshift, was speaking to analysts in the procure-to-pay space, presented his vision of the future that was absent of documents. I can foresee how a virtual assistant and conversational UI can annihilate the need for a variety of documents that are standard today for enabling transactions. With the virtual assistant, you have the absence of credit card fraud, maybe some expense reports, what else does this pretty much eat?
Zakin: It eats process. Back to what Minqi was talking about: the compartmentalization of companies that have different modules that perform different tasks. Taking a step back, I like to think about what we are doing as creating a platform for doing business. Companies have to make use of various categories of expertise and in order to take advantage of that expertise, they need to make purchasing decisions. And that is a large part of doing business–actually making use of those non-core competencies. We’re creating a platform for doing business more efficiently.
When I talk about annihilating process, it’s really about making companies able to function much more efficiently and leanly, such as not having to have an accounting department when they can outsource that into a separate module. The great thing about chat is that it’s agnostic. It can provide the foundation for a lot of expertise across a large swath of verticals.
Jablonski: Got it. Thank you very much Minqi and Peter for being on the show today.
Jiang: Thanks so much for having us
Zakin: Thanks Chris, this was fun.
About the AuthorFollow on Linkedin More Content by Tradeshift Editorial Team