Learn about the new features and enhancements included in v10.0 of Kore.ai Experience Optimization Platform, released on January 21st, 2023.
The v10.0 of the Kore.ai XO Platform comes with various innovative features that enable enterprises to continuously improve IVAs on auto-pilot and extend our open platform architecture. The release also includes exciting new ways to train your virtual assistants using Large Language Models and Generative AI technologies. Experience the future of automation with this release.
Key features and enhancements included in this release:
Design, Collect, and Analyze Feedback
The new Feedback Survey module helps you to capture feedback from the users about your products, services, and overall experience with the virtual assistant. The in-built analytics help you get valuable insights using NPS, CSAT, or any other metrics of your choice.
The survey builder allows you to quickly create surveys using CSAT, NPS, and Like/Dislike options to collect feedback. The platform can automatically create a Dialog Task with a flow to capture the feedback, or you can create your flow. You can collect the feedback after every task using the End of Task event or choose to launch the survey at select parts of your conversation. You can also use the new Feedback Service option of the Service Node to build custom survey flows and submit the feedback to your virtual assistant. Learn more.
The new Feedback Dashboard provides the survey results and insights. You can see the distribution trend of feedback scores and the overall score for a given period. You can also review the full feedback results with an option to export them. Learn more.
The Kore.ai XO Platform now allows developers to create a survey by providing a Survey Name and pushing analytics data using a public API if a dialog is not created. Learn more.
Pre-built Integrations with Popular Business Applications
Jumpstart your VA development using the pre-built conversation flows and integrations with popular business systems.
The XO Platform now offers pre-built integrations with many popular business applications like Salesforce, Zendesk, ServiceNow, and others. The low-code integrations come with full conversation flows to automate the popular actions, including the entities, API integrations, and response mapping for presenting the messages to the users. Prebuilt integrations with many more business applications will be launched very soon.
You can enable the integrations relevant to your virtual assistant by providing the necessary authorization details. Explore the various actions associated with the application and install the ones you need. These actions get created as Dialog Tasks in your virtual assistant. Each of these tasks comes with a complete flow to capture the user’s required inputs using Entities and present responses to the users using Message nodes. Learn more.
Pre-built Integrations with Contact Center Systems for Handing-off the Conversations
INTEGRATIONS AGENT TRANSFER
Kore.ai’s pre-built Agent Transfer integrations allow you to seamlessly handoff the conversations to any of the popular Contact Center applications without the need to write any custom code using the BotKit. These integrations reduce dependency on skilled IT staff and empower business users to configure IVAs quickly.
Pre-built integrations with Drift, Genesys, HelpShift, Intercom, ServiceNow, Salesforce, Unblu, and Zendesk are available now. The platform hosts these integrations, and there is no need to host any custom BotKit.
You can enable the agent transfer integration of your choice by providing the required configurations. You can also enable multiple agent integrations and map them to specific channels if you have such a use case. When you enable multiple applications for Agent Transfer, you can also choose one of them as your default. The Agent Transfer node also supports the IVR Properties to hand off the conversations to human agents for Voice channels. Learn more
Expand VA Skills and Improve Intent Coverage
The new integrated Intent Discovery module is designed to help you auto-extract popular intents from previous user conversations. It helps automate the creation of intents and provides insights for the conversation design. Upload your existing information sources like previous chat transcripts, support tickets, and knowledge base to discover the popular intents. It reduces the time and effort to build a virtual assistant and leads to the success of your Conversational AI Journey. This is a beta feature and is available only for the English language. Learn more.
Sharing & Transfer of Data Tables and Views
DATA TABLE & DATA VIEWS
You can now transfer the ownership of data tables and views that you own to other users of your workspace. The complete list of all data tables and views is also available in the Admin Console so that your account or workspace administrators can manage the ownership. The platform automatically identifies any dependent tables or views and transfers all of them to the new owner.
The new Shared Tables menu shows all the tables with assignments to the bots that are shared with a user. Similarly, Shared Views are also now available. Shared tables or views can be accessed in the read-only view.
You can now see the complete definition of the table or view from the Service Node configurations. This helps to map the correct values while using the Data Service to view or update the details. Learn more.
Support for Large Language Models and Generative AI Technologies
This release also includes multiple new features that leverage the power of Large Language Models (LLMs) and Generative AI technologies. These features simplify the intent detection and answering from documents processes. We will launch many more features that intelligently enhance the virtual assistant experiences using these technologies.
ZERO-SHOT MODEL with OpenAI
Your virtual assistants can now identify the intents from user utterances without providing any training. Yes, no training is required! You can now integrate your VA’s NLU Engine with OpenAI for intent identification. Make sure to use intent names that correctly describe the user’s intent and ensure good intent coverage for the best outcomes. The intent names and user utterances (after applying PII redactions) are shared with OpenAI for intent identification. This is a beta feature and is available only for the English language. Learn more.
FEW-SHOT MODEL (Kore.ai Hosted Embeddings)
The Few-shot Model uses the power of large language models to identify the intents with no or minimal training. This model creates embeddings from LLMs hosted by Kore.ai and uses custom models for identifying the intents from the user utterances. Make sure to use intent names that correctly describe the user’s intent and ensure good intent coverage for the best outcomes. This is a beta feature and is available only for the English language. Learn more.
ANSWER FROM DOCUMENTS
Your virtual assistants can now intelligently extract knowledge from documents and automatically answer user queries without curating questions or responses. All you have to do is just upload the PDF documents that contain your knowledge base. This feature leverages LLMs hosted by Kore.ai to identify the right content for a user query and Generative AI models from OpenAI to dynamically curate the answer. The content from the documents and user utterances (after applying PII redactions) are shared with OpenAI for curating the answer. This is a beta feature and is available only for the English language. Learn more.
NLU Accuracy and Performance
NLU VERSION 3
The new version of the NLU Engine (Version 3) allows you to explore the latest features like Zero-shot Model, Few-shot Model, Intent Discovery and also improves accuracy and performance. The new VAs will automatically use the new version by default.
We recommend migrating all your existing VAs to the new version to experience the new features and improvements. Make sure to back up your virtual assistant before upgrading, validate the NLU performance using the Batch Testing module, and publish the changes.
The previous version (Version 2) will be discontinued in the upcoming product releases (Q2’ 2023). On-Premise customers can run both versions in parallel or switch directly to Version 3.
Connect to External NLU Engines
EXTERNAL NLU INTEGRATIONS
The Platform now allows you to connect to the external NLU engines, especially if you wish to migrate from another platform to the XO Platform. With external NLU integration, you can continue to have the NLU training on an external system and build the conversation flows on the XO platform. The integration is supported during utterance testing, batch testing, and end-user interactions and is useful for Intent and Entity detection. Conversation management and all other functionalities are managed on the Kore.ai XO Platform. Integration with Dialog Flow ES is now available, and more will be available soon.
If the external NLU engine does not identify any intent, then the XO platform’s NLU engine will come into play; it works as a fallback engine. Learn more.
Automate Regression Flow Testing
With the improved Conversation Testing framework, you can now confidently roll out new updates to your assistants by automating the conversation flow testing using dynamic text, flow, and context assertions.
You can create test suites either by recording the conversation or by directly uploading the conversations. After you upload or record test cases, the platform validates the conversation and assigns text and flow assertions for every step.
Text assertions compare the text of the expected output with the actual output. You can also mark specific parts of the text assertion as dynamic so that the test cases are marked as a success, even if the value of that specific part is different during the execution. For example, a user’s balance can be marked as dynamic as it can change every time the suite is run. Flow assertions refer to the tasks, nodes, FAQs, or standard responses triggered during the flow execution. It ensures that the bot has traversed the same path as you expected. You can also use context assertions to verify if the context object contains specific parameters that you may require for successful flow execution.
For every test case, you can now see the complete test coverage information across the dialog, FAQ, and small talk intents. For every execution, you can see the overall execution statistics and test results of every test case, along with failure reasons, if any. You can also visually compare the expected and actual results. Learn more.
Analyze and Improve Flow Coverage and Accuracy
HEALTH & MONITORING
In addition to the NLU Health, the Health & Monitoring dashboard now includes Flow Health to present the overall flow coverage and accuracy.
You can view the analytics for Dialog Tasks, FAQs, and Small Talk. You can also view the intent level summaries and test coverage details to easily identify the nodes or intents to be covered. You can also drill down to the test case results to view the execution analytics. Learn more.
Improved Messaging Channel Support
The Platform now supports the following additional messaging channels:
Sunshine Conversations channel now supports in-built live agent transfer to Zendesk via Switchboard. Please note that you need to reach out to your Zendesk account manager to enable this integration. Learn more.
Webhook v2 now supports Sync Mode without the need for implementing the polling framework. All consecutive responses are delivered to the incoming request and do not require you to poll. It also supports setting user language and the end-of-conversation status. Learn more.
You can now configure multiple phone numbers for the WhatsApp channel with Infobip as the partner. The platform will automatically respond to the ‘from number’ in the incoming request. It empowers you to enable the same virtual assistant for multiple geographies or business lines with different contact numbers. Learn more.
The Platform now secures the WhatsApp channel integration for GupShup by authenticating the Callback/Webhook URL. Learn more.
The Platform now supports Templates and Free-form texts for WhatsApp business communications on the Infobip Channel. Learn more.
New and Improved Analytics Capabilities
This release includes significant updates to analytics capabilities on the platform, making it easy for you to review and continuously improve the NLU and flow performance.
The new Conversation Insights analytics module provides a visual map of the user utterances to easily discover false positives and false negatives and identify opportunities to train new intents.
The module aims to provide a structured approach for analysts to regularly review and improve the NLU performance. Utterance clusters are formed using deep neural models, and each cluster is assigned a name that is representative of the underlying utterances. These clusters are presented using a visual heat map. You can also switch to the tabular view to see the list of all intents and the corresponding clusters under these intents.
You can easily identify the clusters that need further analysis. Some common practices include looking at clusters that identify multiple intents or clusters where the name does not seem relevant for the corresponding intent, etc.
You can also use this module to create new intents for the utterances that do not belong to any of your existing intents. Learn more.
You can now create filter presets to focus on specific conversations and easily review them using the contextual annotations of key conversation events. In addition to the built-in presets, you can create custom presets to focus on conversations that match specific criteria, for example, all conversations with two or more intent failures, conversations that went beyond 15 mins, and more. Learn more.
The conversation transcripts are now enriched with key events like intent identification, sentiment-based triggers, entity retries, etc. These events could be system-triggered or any meta tags you may have configured to auto-emit during conversation runtime. Learn more.
The new Custom Dashboard filters provide insights about specific customer groups or business lines. These filters are at the dashboard level and are automatically applied to all the widgets in the dashboard. Adding these filters to the individual widgets is not required. Learn more.
Designing custom widget dashboards widgets is now made much simpler. The new query builder provides type-ahead suggestions to write the criteria for Select, Filter By, Group By, Having, and Sort By clauses. The platform provides suggestions of the fields present within the selected dataset or message tags, session tags, and user tags that are added to the bot. Learn more.
We have enhanced several other features in this release:
- You can mark Environment Variables as secured to encrypt the values and make them read-only once. It’s useful when co-developing the bots or using confidential keys as environment variables. Learn more.
- The new Ambiguous Intents Identified event allows you to customize the conversation flow when ambiguous intents are identified. You can associate the event with a dialog task, access the complete NLP information from the context object, and define the experience. Learn more.
- The platform now displays the popular intents identified during a conversation and their count on the Popular Intents Widget of the Performance Dashboard.
- You can now selectively disable the Global Grammar used for IVR channels at specific entities. For IVR channel conversations, the platform now provides additional information from the ASR engines, like input mode and confidence level in the context, allowing you to customize the conversation flow. Learn more.
- The release includes many improvements to intent, entity, and trait identification, especially for Polish and Arabic languages. Learn more.
- The Platform now supports additional Yes/No Synonyms for the Traditional Chinese Language. Learn more.
- The NLU validation ‘Incorrect Patterns’ of the FM Engine is now supported for all languages. The KG Engine also now supports Incorrect Patterns validation. Learn more.
- The Platform now supports Bot Substitutions, Aliases, and Homophones during NLP training on voice channels that implement STT transcriptions. Learn more.
- You can now filter the NLU recommendations to locate the intents that need to be fixed easily. Learn more.
- The definition of ‘Drop-off’ conversations can be customized for specific tasks. For example, you can mark the conversations abandoned during utility tasks like ‘follow ups’, ‘feedback flows’, ‘is there anything else’ etc., as ‘self-service’. Learn more.
- The Get Analytics API includes the details of the unhandled utterances. The Billing Sessions API is enhanced to provide more granular details of the billing units consumed, with the ability to view individual units or aggregated values.
- You can now customize the path behavior of buttons during Storyboard design by removing, reassigning, or rearranging them from/to their default assigned path. Learn more.
- Admin users can enable Two-Factor Authentication (2FA) at the workspace level for bot designers from the Admin Console. When accessing a 2FA-enabled workspace, the user is prompted to complete the login using the verification code sent to the registered email address. Learn more.
- This release also includes revised subscription plans for Standard accounts/workspace. Every Standard Account is credited with $500 Free Credits that any of the VAs in the workspace can use. As always, you can add Paid Credits to specific VAs for production use cases. Note that there is no change to the Enterprise Accounts. Learn more.
- Mongo database, RabbitMQ, and NodeJS modules are updated.
- The Platform now supports the migration and import of a Google DialogFlow project including the intents, entities, and the ML training utterances to an existing/new Bot Builder project. Learn more.
- The platform now notifies the bot designers if the size of the context object exceeds 1 MB in size. We recommend you review these notifications and ensure that the data added to the context does not exceed 1 MB. In the upcoming releases, the platform will discard any conversations if the context exceeds 1 MB in size.
The following are the compatible platform extensions for Version 10.0 of the XO Platform: