Thinking about Virtual Agents
The work taking shape around virtual assistants (agents) is in many ways a look at the “deep end of the pool” for conversational UI. The emphasis is on conversational context, trust and delivering tangible value. Studying this helps explore the conversational user experience with greater perspective.
The Virtual Assistant Summit 2017 in SF was a good way to check out the latest in this space. A few noteworthy ‘agents’:
- x.ai uses email to sort out scheduling for you
- mezi is a personal assistant for travel and shopping
- Abe is an agent to help manage money
- NextIT helps businesses create messaging experiences for their customers
- Sense.ly provides a virtual assistant for healthcare assessments
- Joy helps mental health patients by conversing with them
- Ross helps lawyers find information relative to legal cases
It’s still early but some clear common themes have emerged.
The deeper the conversation — the narrower the scope
Virtual agents need a narrow focus: be good enough at a specific set of tasks to get customers to pay for them. Utilitarian turns out to be a good thing.
Amy from x.ai is a good example of an explicitly narrow agent objective: scheduling meetings via email correspondence. The pain of coordinating schedules for an event, gone.
The medium for this agent is your email and calendar: you give the agent access to these then cc: amy@x.ai, and the scheduling happens over multiple emails on your behalf.
What’s notable about this approach:
- no new app/website/etc. works with existing productivity tools
- an asynchronous messaging medium
- very narrow focus: do one thing well (enough) to get paid for it
The use of email as the conversational medium in this case has a ‘steampunk’ feel (old technology used in contemporary setting), but this is where most of the scheduling work happens.
All of the agents demonstrated had very narrow scope. The deeper their abilities seemed, the narrower their range of motion. This is a logical starting point: start narrow.
Conversational Blending: human and machine responses
Increasingly businesses need to converse with customers through a conversational interface. The conversation is ultimately a blending of human and machine with the goal of providing excellence in customer service.
This approach is an inevitability, we hereby give it a descriptive label:
“Conversational blending” — human and machine collaboration towards responses in a conversational interface
Businesses already have customer service representatives, and their efficiency and customer experience over phone is often lackluster. Customers are increasingly accustomed to messaging apps and expect high-quality service on their schedule —often they are willing to trade some asynchrony for convenience.
All of the agents on stage at this year’s VA Summit used human/machine blending on the back end of their response processing.
Sacrificing user experience for scale, particularly when you are selling a concierge service, is a really bad idea.
Noteworthy aspects of this ‘blending’:
- the machine-only user experience has serious limitations
- over time the machine can take on an increasing % of responses, this will yield to iterative learning
- asynchrony is unavoidable: human response (including routing to the right person) cannot be instantaneous
- the human expert may contribute to the machine response behind the scenes: validating machine selected information, disambiguation, etc.
This topic of human and machine collaboration on responses deserves more attention.
The conversation hinges on trust
The etymology of the word ‘conversation’:
“The future of business is a conversation with a machine that earns your trust” — NextIt.com
We, as social animals, engage in conversations that carry trust, specifically the trust associated with the context at hand. The travel agent is trusted with our itinerary, the executive admin is trusted with our schedule, the nurse is trusted with your health-care concerns.
If the scope is sufficiently narrow, as noted earlier, then it’s possible to deliver on the trust necessary to deliver value.
Conversational trust is always earned.
If the virtual agent is not already part of a trusted brand, it must build a baseline of trust with its new customer.
We will eventually see virtual agents collaborating and virtual ‘brokers’ in front of these.
But first — Baby steps.
