Best Practice: When to use single or multiple NLU Models in your Custom Scoped Applications and how many intents / utterances are supported in a single NLU Model.DescriptionIn a Service Portal for general usage, best practice is to use a single NLU Model for all intents and if there is any overlap of intent predictions, you should review your utterances for the conflicting intents and rectify these overlaps to fine tune your NLU Model. It is easier to fine tune a single NLU Model than multiple NLU Models, as your fine tuning will span across multiple models, but you can only test an utterance on a single NLU Model via the NLU Workbench. NLU Batch Testing will enable you to test utterances across multiple NLU Models at the same time, an essential tool for NLU testing to see how your NLU Models perform as a whole. You can use multiple NLU Models in a single Scoped Application, but these NLU Models are not aware of each other and so it becomes more difficult to manage and debug multiple NLU Models, whether in a single Scoped Application or multiple Scoped Applications and so we recommend using a single NLU Model in a single Scoped Application where possible, as the Virtual Agent response times will also increase with each NLU Model it needs to do a prediction on. In addition, with a single NLU Model, you only need to maintain one set of Vocabulary Items for your acronyms and business terms that require a synonym. NLU Models once trained and published will be made available in Virtual Agent Designer to any VA Topic within any Scoped Application, irrespective of the Scope Application the NLU Model is in. Note: We now support up to 750 intents or 20,000 utterances in a single NLU Model. However, models with over 300 intents or 4500 utterances, whichever comes first, will take longer to train, test, and publish, as it will use asynchronous training, instead of synchronous training.InstructionsVirtual Agent (VA) is the application and NLU is the service and the ServiceNow Platform is designed that you control the VA Topics displayed to the end-user in the VA Topic Properties tab, where you can control who uses the topic and what channels it can run in. You cannot control NLU in this way, as there is no option to add such properties on the NLU Intent in your NLU Models and so restrictions are done at the application layer. Predictions on the NLU Model are invoked for any active/published VA Topics that are mapped to an NLU Model/Intent in the VA Topic Properties tab. Note: If the NLU Intent is not mapped to a Virtual Agent Topic, then you should disable the NLU Intent in the NLU Workbench, to exclude it from the NLU Model without having to delete it, and retrain the NLU Model in order to prevent the utterances in this unused NLU Intent from influencing predictions on your active NLU Intents that are mapped to a Virtual Agent Topic. However, if using a Service Portal, that is application specific, such as an HR Portal or ITSM Portal, where you have similar VA Topics and intents, you should use a different NLU Model for each Service Portal contained within a single Scoped Application for each application. In the Advanced properties (optional) section on the VA Topic Properties tab, the Who can access this topic section enables you to control who uses the topic and what channels it can run in. You can also set a Context with a condition where the expression that must return true or false in order to display the topic to the user. Our ServiceNow application teams have their own NLU Models, hence out-of-box, we have a NLU Common Models application for general usage, with multi-lingual support and each application team provide their own Scoped Application, such as ITSM NLU Model for Virtual Agent Conversations, which contains their own single NLU Model and is application specific. These out-of-box NLU Models are designed to be used as templates only, as both NLU Intents and Virtual Agent Topics are read-only. You should import the out-of-box NLU Intent into your single custom NLU Model and duplicate the associated Virtual Agent Topics, so that they can all be customised to fit your business needs.