“Always design a thing by considering it in its next larger context - a chair in a room, a room in a house, a house in an environment, an environment in a city plan.” – Eliel Saarinen
Humans rely on established knowledge, cultural awareness, and implicit clues communicated through tone and body language in order to establish the context of nearly any conversation. Context shapes the rules of engagement: the questions we ask and the answers we proffer. We are rooted in context, and when shared experiences are absent from our interactions, we lose a sense of connectedness, trust, and clarity.
Humans provide the context that intelligent virtual assistants (IVAs) need in order to successfully navigate human/machine conversations. The problem with most IVAs, and AI technologies more generally, is that they are contextually challenged. As a result, end users bear the burden of explicitly describing the context governing most machine/human interactions. IVAs are only as effective as our ability to interact with them, and an IVA that asks irrelevant questions or solicits answers that could be inferred without input is bound to frustrate and alienate end users in a matter of minutes.
The primary challenge of “conversational AI” is to infer context from the appropriate mix of information, experience, and user input.
So how do we build contextual intelligence into an IVA? We construct industry-specific language models, and we prioritize specific KPIs when customizing IVAs for individual businesses. We value experience: our interfaces ship with the capacity to learn from each and every interaction with your customers or users. We integrate with your systems of record: rather than burdening end users for information, we rely on what you already know about them to drive a more natural conversation.
We’ve previously talked about how we design industry-specific solutions and build language models. Here, we’ll expand on integrations and extensibility. It’s easy to take these features of a platform for granted, but they’re critical for contextual awareness, and they're key components of any successful IVA.
1 – Extensibility makes it possible to deliver an IVA via a website, mobile app, text/SMS system, and more. Our IVAs are truly channel and platform agnostic because they’re designed for extensibility. We’ve even had an IVA answer questions on Facebook – the first to ever do so. Cross-channel delivery is table stakes today. Imagine a bank that didn’t have online banking; they would churn customers at an unbelievable rate because they’re not set up to deliver services where the users want to engage. The same logic applies for IVAs: don’t force users to conform to the single channel that you prefer.
2 – Because our IVAs can be fully integrated with your site or application, it’s possible to create granular contextual awareness. Our IVAs are capable of knowing what page or screen your end users are currently viewing. Even monitoring user authentication and knowing something as simple as whether a user has logged in can drastically improve the customer experience. It’s all about expectations – users expect you to know this information, and if you have to ask them to get it, you’re shifting the burden onto them.
3 – Integration with your systems of record makes it possible to tap into the rich context of your previous interactions and conversations with users. Imagine if each time you talked to an office mate they forgot everything you previously discussed. It would be infuriating to reassemble context for every new conversation you need to have. People don’t do that, and IVAs shouldn’t either.
In drawing an analogy to Eliel Saarinen’s sage architectural advice, we propose that a successful IVA is like the chair in Saarinen’s room. It belongs in the context of its surroundings: in the room that is your business, in the environment that is your industry, and in the plan that is your vision for the future.