Voicebot Design Needs to Be Simple, Gridspace Execs Contend at CX Connect
When building a modern voicebot to handle customer interactions, artificial wisdom is far better than artificial intelligence, Gridspace corporate executives said during the opening day of the 2022 CX Connect virtual event, which focused on speech technologies.
Unlike AI, which is limited, artificial wisdom provides a “compendium of knowledge and a store of accumulated memories,” said Anthony Scodary, co-founder of Gridspace.
Artificial wisdom, Scodary continued, makes systems smarter with less human instruction, makes automation worth the investment, and makes contact center automation “maximally uaseful.”
“Artificial wisdom is very critical to building systems with a very high ROI,” he stated.
In general, for contact center automation to be worthwhile, it needs to work quickly, intelligently, and independently; should be less of a burden for agents, customers, and programmers; and should be inexpensive.
It also plays a role in automated systems’ ability to get information into and out of data repositories. “Ultimately, you want to be able to provide information to end users and to inject data collected back into the system,” Scodary said.
Additionally, building a bot with artificial wisdom should ideally involve document crawlers and indexing; online learning and adaptation; specialized supervised speech models; and distributed authoring, which Scodary said is critical, enabling teams to collaboratively encode their knowledge of their particular niches of the business into the final design.
Artificial wisdom was a central component of Gridspace’s Grace voicebot and its Sift speech analytics product. The company also baked in several other layers that Scodary said are critical to successful voicebots today. They are recognition of named entities, semantics and meaning, emotions, demography and roles, dialogue activities, speech, speaker identity, and voice biometrics.
The secret to a good bot, he intimated, was doing less authoring, less testing, and less work, instead focusing on the specific needs of the business.
But beyond that, good voice automation blends right into operations on Day 1, added Cooper Johnson, lead voice user experience designer at Gridspace.
Good conversation design also allows the voicebot to repeat what it heard and offer a warm transfer to a live agent when it cannot handle the request.
Other keys to successful voicebot design, according to Scodary and Johnson, include a feedback and tracking mechanism to know whether the bot is working as planned, adaptability so that the bot can dynamically adjust the dialogue if it encounters words or phrases that are outside of its original programming, and the ability to supplement word recognition with emotion recognition so the dialogue can be adjusted in real time.