Posted June 23, 2020

Today, you really have four options to interface with an application: UI, formal language, chat, and voice.

Until very recently, a system designer was faced with an ogre’s choice between UI and formal language — make the UI simple and intuitive, and strip away the user’s ability to ask what they want. Or, provide a tremendously complex formal language interface that is unique to each application, and hope the user can learn enough of it to be able to use the system. Simple to the point of uselessness (UI), or complex to the point of impracticality (formal language).

Users want to talk to the applications and services they use. Or more concretely, users want to describe what they need and how they feel using the natural language(s) they use every day to communicate with other people, which makes chat and video the ideal interfaces. This isn’t just about convenience, and it isn’t just about comfort. Ultimately it is about expressibility and usability. Let me explain. Take the questions users type into systems, say a search engine, or perhaps a customer support ticket. Very often the intent behind those questions are unique to that user, or only shared by a small handful of users. Yes, there are often very popular questions (“how can I change my password”), but in cases we’ve studied as much as 50% of the inputs from users were unique even within the same app. And, of course, different applications have entirely different ranges of inputs from users.

Because we just say so many different things in so many different ways, natural language understanding (NLU) and dialogue management are incredibly complicated. And so, a chat or voice-based interface has meant relying on banks of people to interface with users, from support, to sales, to gathering feedback.

Advances in NLP — from its ImageNet moment to new benchmarks — over the last few years promise to change all of this. We’re now entering an era where software can interact with a human using natural language at levels of accuracy that dramatically change the usability of a system. The leading software project providing the engine to do that is Rasa, a horizontal tooling layer that provides an open source framework that developers can use to build incredibly sophisticated conversational AI functionality into their applications.

We’re now entering an era where software can interact with a human using natural language at levels of accuracy that dramatically change the usability of a system. The leading software project providing the engine to do that is Rasa.

And why go horizontal when vertical solutions, like a customer support chatbot, already exist? Because each vertical has its own language and dialogue logic. It’s not just different words, but the concepts underlying those words. In fact, most companies have their own languages (particularly large companies with complicated projects).

Companies that are inundated with millions, or even billions, of conversations every month — insurance, healthcare, banking — need a semi-automated solution. If an organization doesn’t build their own chatbot, it’s very hard to get the customer experiences right. Enterprises have started to build out teams to work specifically on chatbots, and they have realized the limitation of cloud offerings, which are not flexible enough to provide a good user experience. They are looking for a more customizable solution, and this is where Rasa shines.

The power of Rasa is directly visible in the size and energy of the community that has grown around the open source project. At a16z, we track many open source infrastructure projects and the following that’s developed around Rasa is world-class. For this space, we believe open source is very important for a number of reasons. First, the problem domain is incredibly complex, and so it requires a rich interface for integrating and extending. NLP and machine learning continue to evolve quickly and having a vibrant community around it helps the project keep pace with this incredibly dynamic space.

Beyond just having a large and enthusiastic developer community, Rasa is in wide use across the globe. We’ve been incredibly impressed with the adoption of the project across many verticals, from healthcare to banking to insurance to tech. Furthermore, Rasa the company has built a fast-growing business supporting those users and developing enterprise solutions to help companies build world-class conversational AI interfaces into their products.

There is, of course, a phenomenal team behind the success, the technical work, the community-building, and the business built around it. We’ve been speaking with Alex Weidauer and Alan Nichol, Rasa’s cofounders, for years. They are both very product- and community-focused founders with a deep passion for the space. While we were getting to know each other, the conversations we had with the founding team were some of the best I’ve ever had broadly in AI. We shared similar views of the problem space and the complexity of the problem, and agreed that the right way to build a massively successful effort was through a horizontal, open source software effort.

We’re very excited to be leading Rasa’s Series B round, joining the board, and partnering with the team to change how users interact with their applications and the companies behind them. We believe Rasa is destined to be an iconic company and the leading provider of conversational AI infrastructure.