Kore.ai XO Platform v10.0 Release Notes

The feedback module is a use case and channel agnostic. The core of the feedback module is the ability to generate analytics, but to generate analytics, the platform has to start capturing the feedback. When a feedback survey is created on the platform, all messages are in text format. The text from the platform gets converted into a voice using the ASR and TTS engines. You can leverage channel-specific configuration to modify the feedback prompts.

The beauty of this feature is that these feedback surveys get created as dialogue tasks with a series of entities, service calls, and message notes. You will have complete control of modifying the out-of-the-box messages, adding additional messages to specific channels, such as voice channels, and capturing the feedback on voice.

That said, yes, it is available for all the channels, but by default, it only shows text messages. But you can always add additional channel-specific prompts to collect feedback from voice channels.

Yes, Feedback flows are fully customizable. Users can use the automatic feedback template that the platform generates or build their flows and submit the feedback to the platform feedback service.

Alternatively, it could be a hybrid approach where the feedback flow can be in work from some other flow. All of those are possible.

Yes. While the XO Platform provides built-in feedback analytics, you can export the data.

The XO Platform lets you configure feedback in two ways. One, you could use the built-in flows to collect the feedback and use the new service type called feedback service to submit the feedback to the bots platform. Or you might collect feedback from other sources that can be pushed to the platform as part of the conversation execution.

The built-in dashboard provides all of the trends and scores, and you can export the collected individual feedback into CSV format.

Yes, we can export the independent feedback data captured by the user.

The Kore.ai XO Platform v10.0 Release provides 70+ integrations with 15+ applications that include Salesforce, ServiceNow HubSpot, Zendesk, Freshdesk, Freshservice and others. You can find the full list here. All of the details to enable these integrations will be updated in our documentation.

Zero Short: It does have a lot of advantages, but it also has some disadvantages, especially when it comes to enterprise data sharing data with the external systems now getting additional API keys, and there is obviously, there’s no way to fine-tune the model if something is going wrong.

Few Short: This will be our recommended intent identification model going forward. It works beautifully. Even this model does not require you to train for most of your use cases. However, if there are some specific variations of our lenses that the model is not able to handle, you just need to add those few utters.

Currently, the Kore.ai Zero Shot ML Model supports the English language. For other languages, it will be available in the next two to three months.

The R&D team at Kore.ai is working to extend the Zero Shot ML model support for other languages; it will be ready by the end of February. The latest model will support all 100+ languages that the platform currently supports.

The Kore.ai Zero-shot ML model is trained for various domains and data sources. We did extensive testing, and in our test scenarios, we found no challenges in meeting specific domains like finance, banking, manufacturing, and others.

But in the Zero-shot model, we don’t provide training; for this reason, it entirely relies on the intent’s name. It uses the intent name to map or identify the intent name’s similarity with the user input, and that’s how the prediction works. So, the intents have to be defined very well.

For example, in the banking use case, instead of saying transfer funds, it has to be more descriptive within a subject, object, and nouns, i.e., it could be want to transfer funds. It improves intent identification to a great extent. Instead of using simple words like check balance, use descriptive terms like I want to know my balance. It significantly improves the intent identification rate.

The XO Platform and Open AI integration support answering from PDF documents. The users have to upload a PDF, and that’s it. They can ask any question based on the content available in the pdf, and the model can answer. The platform sends data to Open AI as the platform leverages Open AI APIs to process the data. To overcome this, the platform developed a model where the content in the PDF is first split into different pages and sections. The platform uses embeddings to represent the page content as vectors. These are then passed to the Open AI model.

The Kore.ai NLU version 3.0 allows you to move to the latest versions of the underlying libraries. With that, you get the advantage of faster training times and improved entity and intent detection.

The latest model doesn’t apply by default for any of your existing IVAs. All your existing IVAs will continue to be as they are, and there will be no backward compatibility issues.

However, we strongly recommend you migrate to the new version because of the performance and accuracy improvements. But there is an option in the bot builder for you to upgrade to the latest version. All the new bots created after the release of XO Platform v10.0 will automatically start using NLU v3.0.

The Kore.ai NLU v3.0 offers multiple advantages over the current version –

• Performance: The latest version provides better identification rates, entity identification rates, and traits. Especially on the entities, the DNN model performs better than the current CRF approach.
• Training Time: The new NLU model reduces the training time by 20%.
• Backward Compatibility: It ensures none of the existing IVAs will auto-upgrade to the latest NLU version. It’s always a user initiative journey to move from one version to another.

But we strongly recommend doing that because you get better prediction rates and faster training times. You can also verify it by running batch testing on both versions. It is advised to take a backup of the current version before you upgrade to the new version.

The flow health dashboard is one of the critical features we are launching in the v10.0 release.
We recommend that every IVA designer actively start using the conversation testing feature to create the regression test to identify the flow issues. The Flow Health dashboard will show you the summary for dialogues, FAQs, small talk and intents within the dialogues, and the transitions’ coverage. It identifies all possible transitions between two subsequent nodes.

The dashboard provides a top-level summary; you can tap on the coverage or accuracy numbers to see a detailed view. So this gives you a comprehensive framework for managing regression testing of the flow from coverage and test success parameters.

1. @18:39 – Explain the latest NLU (v3.0) model
2. @19:57 – Can I export data from the feedback dashboard?
3. @20:45 – What are the integrations supported in the latest release?
4. @21:24 – Which all languages do the latest Zero Shot ML model support?
5. @22:20 – How does the custom dashboard help?
6. @24:00 – Explain the upgraded version of the custom dashboard.
7. @27:25 – Does the feedback survey support voice channels?
8. @29:15 – Can we alter the conversation flow? Can we download Feedback data?
9. @30:31 – Does the Kore.ai zero-shot ML model support domain-specific terms?
10. @33:00 – What are the advantages of the NLU 3.0 upgrade?
11. @35:30 – Can we create custom message templates within the development environment?
12. @37:03 – What is the Flow health framework?
13. @39:45 – Where do I see the list of new integrations? Can I request an integration?
14. @40:49 – When can I start using the XO Platform v10.0 Release?

Custom dashboards allow users to build dashboards by using out-of-the-box data sets or configuring custom tags, like user tags, session tags, and message tags – generating business-specific KPIs.

The latest update allows business users to create widgets without writing queries. The low code interface enables you to build a custom widget by selecting the required intents, entities, and other parameters from a dropdown menu.
For example: If the user journey includes multiple channels, you can build a custom dashboard to track each of the journeys that the user goes through and create a funnel to see how users interact on each channel.