The Cobus Quadrant™ Of Conversation Design Capabilities
∗ This is part one of a two part series, please also take a look part two, the Cobus Quadrant of NLU Design.
Most Conversational AI platforms support two main types of capabilities:
Conversational Design, which includes flow branches, slot filling, etc.
And NLU Design capabilities which consider intents, entities, training examples, NLU training and improvement.
In this quadrant, I’m comparing different conversational AI platforms based solely on their conversational design capabilities…
What do I mean by conversation design?
The conversation design process usually begins only once you know the user intent or customer journey which you want to support.
Conversation design of course includes building and managing the various states of conversation, and requires defining the flow-logic, response management, slot-filling and often understanding the business logic integration required to ensure a great user experience.
Strongpoints of Voiceflow are shared workspaces, realtime collaboration, commenting & feedback. Voiceflow is focussing on dialog design, development and integrating with the existing ecosystem of an organisation. Hence from a solely visual design canvas perspective and ease of use, Voiceflow is right up there.
Voiceflow started off as a design tool that required a 3rd party platform like Dialogflow, and so they don’t have all the bells and whistles for going all the way to production — but that this is likely a direction they are headed.
Google Dialogflow CX
Dialogflow CX is enterprise orientated with Google Cloud Integration & Cognitive Services, also leveraging the power of Google for Speech To Text.
The chatbot IDE does not have a steep learning curve, managing context within each page and flow is very intuitive and to some degree digression can be accommodated. As mentioned before, being able to leverage the Google Cloud environment contributes to Google’s completeness and scaleability.
In general Dialogflow has shown its versatility as a conversation dialog manager via the NVIDIA Riva integration leveraging the dialog management capabilities of Dialogflow. The web development console is very flexible, uncluttered and clean.
Boost AI has a visual design canvas approach which seems very standard and tightly coupled with their training data management system and NLU.
The conversation design capabilities are no-code orientated and most probably not as feature rich as the likes of Cognigy, Kore.ai or even yellow.ai.
Kore.ai is experimenting with the automation of conversation design. They have an approach which is referred to as a conversation-driven dialog builder.
This builder automatically converts the mock-up conversations into a dialog flow. The idea is that designers can focus on visualising the end-user conversation before building the dialog. This does remind somewhat of Rasa’s Interactive Learning and Conversation Visualisation.
The Yellow AI design canvas is very responsive, not as intuitive as other frameworks and lacks in richness and features compared to Cognigy or Kore.ai.
Cognigy is one of the leaders in terms of a complete solution and innovation. They are fusing dialog flow design with other elements like NLU. Flow Nodes, this is a new and innovative approach to dialog state management. Flow nodes combined in a certain way, can help create dynamic, flexible, interactive conversations. To some extend this loosens up the state management process.
Cognigy is also pioneering the idea of an extension marketplace of re-useable dialog components.
IBM Watson Assistant
IBM has a very complete solution whilst not having a visual design canvas approach. Their Dialog Skills can be described as a flow state configurator with enhanced capabilities in digression and disambiguation.
Dialog Skills are complimented by Action Skills, which can be used in a stand-alone mode, or as an extension to Dialog Skills.
Oracle Digital Assistant
The graphic design canvas (Visual Designer) is not fully interactive. The ODA canvas reminds of how Nuance Mix implemented their dialog design canvas. With the nodes not being fully interactive in the sense of dragging and dropping elements.
The mediums available for deployment are substantial, but pales in comparison to other frameworks like Microsoft, Cognigy, Kore AI and more.
A few detractors of Nuance Mix is the fact that the design and development canvas is not interactive, the menus and settings on the side need to be used to build out the dialog.
It does feel a bit debilitating not being able to drag, drop and manipulate nodes in the canvas by moving them.
There is a dialog-message abstraction layer, which allows for bot messages to be managed separately, this reminds much of Microsoft’s Composer dialog management.
Microsoft Power Virtual Agents
Power virtual Agents is very similar to IBM actions skills. Power Virtual Agents are in essence a simplified dialog state configurator focussing on simplifying the bot development process. Scaling and complexity will be amongst some of the challenges.
Amazon Lex allows for multiple approaches within a single framework. For instance Amazon Lex V2 has a dialog configuration approach, which can be superseded by a design canvas, and in turn Lambda functions can be used.
The new Lex Visual Conversation Builder is a huge leap for Lex in terms of a more visual and intuitive design canvas, but scaling and collaboration need to be tested on an enterprise level.
Here you will have to make use of a Python pro-code dialog state management framework. But Cisco MindMeld makes it really easy with existing code templates which can be used to fast-track conversational AI dialog development.
Managing the installation and resources will have to be planned for, as there is not a SaaS plan available. MindMeld is open-sourced, which does lent a certain level of flexibility.
Microsoft Bot Framework
The Microsoft Bot Framework can be used as a pro-code implementation, or extended with Composer. Integration options to Microsoft’s Cognitive Services is made easy and there is a certain level of flexibility with a pro-code approach.
As noted previously, Microsoft’s ecosystem is more fragmented and it is up to implementers/organisations to constitute the configuration of the Microsoft services which will serve them best.
Rasa has pioneered the idea of Machine Learning stories, and has illustrated flexibility with the NVIDIA Riva demo which makes use of Rasa ML stories for dialog management.
Within Rasa, policies can be used to bring a balance between a machine learning or rule based approach; where appropriate.
But managing and scaling a Rasa environment is a technical challenge; Rasa is attempting to change this with an enterprise version which is a more no-code canvas approach.
Amazon Lex / Lambda
Amazon Lex dialog state and response messages can be managed from a pro-code Lambda function. Amazon Alexa has a few Lambda functions available in pre-sets / templates. And I would love to see a clear path in Lex V2 on how to extend into Lambda functions with templates and pre-sets.
DeepPavlov is a highly technical pro-code environment which will demand very specific technical capabilities for production deployment.
Currently, DeepPavlov supports two ways to define domain model and behaviour of a given goal-oriented skill — RASA (domain.yml, nlu.md, stories.md) or a DSTC2 format. Read more here.
The Cobus Quadrant™ Of NLU Design
NLU design is vital to planning and continuously improving Conversational AI experiences.
These are the platforms I’m most familiar with — but there are certainly a number of other platforms, for instance Botpress and others, that provide similar capabilities — my goal is more to showcase a few different companies and highlight what makes their particular implementation of conversation design interesting.
If you have any comments or corrections please let me know here in the comments! Also, if you have any recommendations for other Conversation Design tools, I would love if you could highlight those to me. 🙂
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.
No-code tooling for NLU
The complete productivity suite to transform natural language into business insights and AI training data
Get an email whenever Cobus Greyling publishes.
Get an email whenever Cobus Greyling publishes. By signing up, you will create a Medium account if you don’t already…
Eliza Language Technology Community — Language Technology: Conversational AI, NLP/NLP, CCAI…
ELIZA — Where language technology enthusiasts unite.
Five Approaches To Managing Conversational Dialog
And Which Elements Can Play A Supporting Role
Different Approaches To Automation Of Dialog Flow Development
As Conversational AI develops and our digital assistants become more complex, the manual development and rigid process…
Dialog Management Considerations for Chatbots
Chatbot User Experience is Determined by Dialog Management Efficiency