The Current State Of Conversational AI
Here I discuss what is currently not working in CAI, drivers of market fragmentation, NLU Design, Voicebots and more.
What’s Wrong?
Currently in Conversational AI there is an overbearing focus on multi-turn dialog conversation design. While simultaneously elements like the long-tail of intent distribution is neglected, together with NLU Design.
Add to these ailments the siloed approach to Conversational AI, CCAI and CX within enterprises.
Another problem is single vertical vendors attempting to address all enterprise conversational AI requirements.
The light on the horizon, is the emergence of market fragmentation, this is where organisations are adopting the best-in-class tooling to address CAI requirements. The fragmentation is perpetuated by the introduction of CCAI use-cases and voicebots.
NLU Design is also emerging as a methodology for astute data practice and to effectively detect signals within existing customer conversations, and turn those signals into NLU training data. NLU Design can be seen as an accelerated process of intelligent intent detection and management.
With NLU Design converting unstructured ground-truth customer conversations into NLU Training data, a sustainable process can be established where Customer Intents are informing the CX roadmap.
➡️ Please follow me on LinkedIn for updates on Conversational AI.
Drivers of Market Fragmentation
The image below shows a breakdown of the current Conversational AI technology landscape. Consider that the breakdown below does not even include CCAI, ASR and Speech Synthesis. ASR and Speech Synthesis are obviously two vital requirements for voice enabled digital agents.
However, what is evident is the market fragmentation & emergence of best-in-class horizontal tooling for enterprise success.
Contributing to fragmentation is the abstraction of the chatbot messaging layer for improved response design. And Large Language Models (LLMs) which are popularising Generative text for bot responses. LLMs are also playing a key role in the automation of PII redaction.
Multi-Modality & Orchestration
Orchestration will become a critical capability for a number of reasons.
One reason is the ability to leverage existing chatbot effort as much as possible to fast-track voice enablement of digital assistants.
Secondly, multi-modality will be driven by the idea that multi-turn dialogs are not the solution to all customer intents. Each conversation will have to be customised according the the medium, modality and the conversational components available.
To this point, consider the following customer intents, and how a multi-modal approach will work best:
- Request a seat change on a plane
- Getting directions
- Flight check-in
- Car rental
- etc
NLU Will Inform The CX Roadmap
The process of NLU Design will uncover important signals expressed by customers, service representatives etc from historic and real-time conversational data.
NLU Design invariably yields CX improvements due to actionable, qualitative and quantitative insights. Spinoffs of this process are improved automation, personalisation, customer support success, and data for AI-powered features.
➡️ Please follow me on LinkedIn for updates on Conversational AI.
CCAI and Voicebots Will Drastically Improve
With the current focus on CCAI and voicebots, user experience is set to drastically improve. Technology providers are focussing on elements which empowers CCAI and voicebots.
Innovation in Advanced Speech Recognition (ASR) is addressing impediments which seemed insurmountable in the past, like phone based voice quality, multiple speakers, background noise, etc.
In the past ASR and Speech Synthesis adoption were severely impeded by the scant representation of minority languages, regional accents and locales. This has been addressed recently, most notably by Whisper and Microsoft.
Chatbot & Voicebot Technology Landscape
As a reference, below are two technology matrixes which cover both chatbot and voicebot implementations.
➡️ Please follow me on LinkedIn for updates on Conversational AI.
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
