Cast your mind back to 2019. Game of Thrones came to a close, royal babies were born, and everybody hated chatbots.
Well, not everybody. But, according to Forrester Research, most consumers did.
Indeed, the analyst found that 54 percent of US online consumers believed that interacting with a chatbot had a "negative impact on their quality of life." Yikes!
Thankfully, conversational AI has come on leaps and bounds since then and interest in voicebots specifically has spiked.
Just take a look at this Google Trends graphic, which shows search volumes for “Voice AI”.
Much of this interest likely stems from the hype currently swirling around ChatGPT and generative AI.
Indeed, such technologies already augment many enterprise conversational AI platforms.
Nonetheless, many more technological advancements have contributed to the customer-friendly, value-adding voicebots of 2023.
Here are five examples of these advancements that have made the technology a much more viable option for businesses of many shapes and sizes.
1. Speech-to-Text
In March 2020, Statista published a study that found the average accuracy rate for speech-to-text automated transcription models across various industries was 77 percent.
In other words, the average model could only transcribe 77 words out of 100 accurately.
However, fast-forward three years, and the technology is much more advanced. Indeed, Microsoft and Amazon have reached 95.9 and 95.6 percent accuracy rates, respectively.
Still, a one-in-20-word error rate may sound bad. Yet, it is not bad at all. As Pierce Buckley, CEO & Co-Founder at babelforce, explains:
“Customer conversations will likely involve simple language, not tricky terminology from a science paper or complex book – which the AI will struggle with.”
As such, voicebots for simple use cases – such as gathering customer information upfront before a live agent interaction – now achieve well over an 80 percent success rate, according to Buckley.
In addition, much of that remaining <20 percent will fail because of issues detached from the capabilities of the conversational AI model.
For instance, perhaps an integrated system could not locate the customer ID number, or the customer has no existing record. Such problems are most likely to cause voicebot failures.
2. Text-to-Speech
Think of how we speak, as humans, constantly changing our emphasis - in very subtle ways - to convey tone and often meaning. For a voicebot, that is much more tricky.
Conventionally, voicebot vendors will take one of two approaches to overcome that issue.
First, they may employ a predictive, statistic-based neural network or stochastic model.
The alternative is human-written rules, where developers can use control mechanisms within the voicebot to place emphasis.
Yet, as voicebots advance, vendors are finding a happy medium. Sharing why, Buckley says:
“Imagine writing out a sentence that has two places for emphasis. You want to be able to tell the bot to do that. You don't want to wait six months until the neural network has enough data for that type of request, so it does it automatically. It should go live on Friday.”
As such, interfaces will allow developers to add points of emphasis – or "prosodic markers", as linguists would say – so the bot says something in the desired way.
Nevertheless, neural networks will run within the bot, continuously learning and improving – across each language it speaks – so future generations require less manual programming.
3. No-Code Tools
Within 30 minutes of first working on a voicebot, a business can have its first flow up and running while planning an A/B test for the following week.
Such speed to deployment is relatively new, and much of this stems from the development of low-/no-code interfaces.
These interfaces make the experience of building a voicebot similar to playing a video game, as developers navigate drop-down menus, connect dialogs, and select various tasks and actions.
Now, with LLMs, brands are taking this further. For example, Google is using its Bard-powered App Builder to plot conversational flows and tweak the design automatically for IT teams, utilizing natural language prompts alone.
