Mid-term goals

1. Visual programming language for conversational-AI

Development of dialogs with human-like performance starts with creating an initial “happy path”. It is followed by a continuous iterative process of improvement based on QA and production traffic analytics. We believe that in order to make conversational bots viable and scalable all of that has to be intuitive and manageable by one person - Dialog Owner. That is why we are creating our own visual programming language which is super light for users and allows building and maintaining very complex dialogues.

Why? We’ve seen many dialog building approaches/tools across Conversional AI companies and we are convinced our approach is unique and better.

2. In-depth analytics

Production data and feedback trigger iterative work on dialogs. Visual editor allows to manage the explosion of combinations and staying in control will require more insightful, deeper analytics integrated into it.

Why? Dialog Owner has to be super efficient in maintaining dialogs and monitoring performance. Other benefits are: knowledge extraction and possibility to create sophisticated classification rules for QA.

Insight: Only 0-5% calls in human call centers are audited and there’s very little control and knowledge extraction. Vast majority of speech analytics systems are keywords based and provide limited insight. That’s one of the reasons virtual agents are huge paradigm shift.

3. Adaptive omnichannel dialog workflows

Another crucial part of our quest to make conversations with bots outperform human agents. Most conversations are just a part of a long running process. For example, customer’s request during an inbound call needs to be followed by notification upon resolution a couple days later. Follow-ups result from pieces of information exchanged during preceeding contacts. Managing that in a vertically integrated system gives unparalleled opportunities. We also see a lot of use cases in omni-channel workflows such scheduling call time via SMS.

Why? Conversional AI from some of the vendors works fine and businesses are starting to implement them but they still qualify for a very narrow set of use-cases. Processes are managed in people-oriented tools like CRM or Contact Center platforms. We will change that by providing a radically new way of designing interactions.

Bonus: Our visual programming language is being developed with Step 3. in mind and workflow building will be based on the same philosophy as dialog development.

4. Transforming Quality Assurance and Testing towards Software 2.0 paradigm

Scaling iterations requires maximum automation. Firstly, qualification for workflow/dialog testing should not require analytical human touch. Secondly, QA process should incorporate data tagging. Finally, NLU models should be retrained automatically.

5. NLU development

Multi-language models, mini-ASRs, higher generalization, intuitive builders deeply integrated with Studio

About Mid-term goals NLU Manifesto Team Join us