In our previous posts of the Winter ’22 Release Series we had a look at some of the amazing new features from Salesforce Winter ’22 release in areas of Salesforce Platform and Flows . In this post we will be having a look at the exciting Einstein features that will be delivered as part of Winter ‘22 release.
1. Teach Your Bot to Speak Multiple Languages (Beta)
Monolingual bots are so 2020 ! introducing multi-language support in Einstein bots ! You can now add multiple secondary language translations for each dialog to the same bot. This will help deliver consistent experience across multiple geographies with minimal overhead for the Salesforce Admins.
What’s great is that you will also get individual scoring for each language intent model, which can be used to train and improve the chat experience for all customers.
Reference URL: https://help.salesforce.com/s/articleView?id=release-notes.rn_automate_bots_multiple_languages.htm&type=5&language=en_US
2. Gather Insights Directly from Snowflake Schema (Pilot)
Currently to create a discovery story you need to connect to Snowflake, design a dataflow to ingest the data and then create a dataset as base for the story creation. With the new live connection to Snowflake, you will be able access the data directly from Snowflake and create insight stories faster without the need to replicate the Snowflake data as a dataset in data manager. This will also prevent duplication of data in Tableau CRM dataset and also save your Tableau CRM storage space.
Reference URL: https://help.salesforce.com/s/articleView?id=release-notes.rn_bi_edd_story_direct_live.htm&type=5&release=234
3. Predict Outcomes for Multiple Categories with Multiclass Models (Pilot)
Currently Einstein Discovery supports numeric and binary classification models. Now with the introduction of multiclass model you can predict outcome of up to 10 classes. Although maximizing and minimizing the outcome is not supported for the multiclass models they can still be very useful where you need to predict the classifications. For example, The model can help predict which product package a customer is more likely to purchase.
Reference URL: https://help.salesforce.com/s/articleView?id=release-notes.rn_bi_edd_multiclass_classification_pilot.htm&type=5&release=234
4. Train and Retrain a Dataset with Less Data
The earlier object-detection-v1 model requires at least 200 bounding box annotations per object for training. With the introduction of object-detection-v2 this requirement of the minimum bounding boxes is reduced down to 50 annotations, just a quarter of what was needed earlier ! While we can still expect higher accuracy with the same amount of data, Einstein Vision’s updates are making significant improvements to the product and making it easier for developers to build better apps with lesser effort.
Reference URL: https://help.salesforce.com/s/articleView?id=release-notes.rn_einstein_vision_bounding_box_increase.htm&type=5&release=234
5. Operationalize Externally Built Models in Einstein Discovery (Generally Available)
Love tinkering with hyperparameters? Well, get your inner data scientist back to work! Create your own custom models with the tools you know and like, bundle the files, deploy to Salesforce and you are good to go! The ability to use external models in Einstein Discovery opens up a whole gamut of possibilities, you can now have faster or more powerful externally trained models which can be plugged in with Einstein Discovery to give an altogether different meaning to AI based insights.
Reference URL: https://help.salesforce.com/s/articleView?id=release-notes.rn_bi_edd_model_import.htm&type=5&release=234
While Salesforce Einstein gets some cool new features, the best part about this release has been the availability of capabilities that can help make Salesforce Einstein simpler, scalable and extendable. If you have any other Einstein favourites from Winter ’22 then do share it with us as comments.
