Salesforce Einstein: Talk to Salesforce with Einstein Voice Assistant (Beta), Resolve Cases Faster with Article Recommendations, and Predict the Future Using Filters
Here are even more ways that Salesforce Einstein helps you work smarter.
Let users make updates to Salesforce—like logging events, creating contacts, and updating opportunities—all by voice. Einstein Voice Skills lets you build custom skills tailored to your users’ most common tasks, generating more high-quality Salesforce data to inform decision making.
The data requirements for several Sales Cloud Einstein features are now simpler, making it easier to get meaningful sales intelligence within Salesforce.
Ever wonder how we build your Einstein Behavior Scoring model? Now you can find out. A new dashboard in B2B Marketing Analytics shows how your custom model weights each engagement activity and which assets relate most strongly to future conversion.
Einstein Article Recommendations uses data from past cases to identify Knowledge articles that are most likely to help your agents address customer inquiries. Einstein Article Recommendations is generally available this release.
Einstein Reply Recommendations analyzes data from chat transcripts to create chat replies that address your customers’ inquiries. Agents select the most relevant chat reply from a list in the Lightning Service Console as they communicate with customers.
Let bots deliver personalized experiences to your customers with built-in logic to send customized messages based on conversation context: either through a pre-chat form, data inside the CRM, or earlier replies to the bot. With Conditional Messaging, bots provide more intelligent responses and can handle nuanced discussion — turning good conversations into great conversations.
Bots are better helpers when they have accurate information, and with Conversation Repair, bots are able to capture formatted text in a kind and graceful manner. Now, bots can suggest to the customer when an email or phone number is formatted incorrectly, and if the customer needs further assistance, the bot can route the customer to an agent.
Get your bot in shape with an improved dashboard that provides actionable recommendations to improve model quality. Recommendations such as identifying well-performing intents provide suggestions to improve your intent training. Better intents mean better bot performance with your customers.
We’re making the conversation logs more granular to give admins a deeper understanding of exactly where the bot needs improvement. This addition, which shows customer inputs and bot actions categorized by dialog, shows admins every move the bot makes.
Now bots can gracefully check whether agents are available for transfer and notify the customer over SMS and Chat. The Agent Availability Check that launched in the Winter ‘20 release is expanding to SMS to provide a multi-channel experience between bots and agents.
Keep customers engaged while the bot delivers the next message by using the chatbot typing indicator. These three dots mimic messaging apps that signify when a person is typing, which shows the customer that the bot is still actively processing their request.
Pack your bot’s bags, because Einstein Bots is now available in 180 countries on the world’s largest messaging app, WhatsApp. Bots can interface with WhatsApp to deliver messages to your customers, just like they do with SMS. Bots can also use automation to bring complex cases directly to agents, all on WhatsApp.
Learn about other changes to Einstein Bots.
Set up and build predictive models for various business segments, such as building and consulting services or enterprise and consumer divisions. You can also narrow the scope of a single model if you want to predict field values for only a subset of your business. Previously, Case Classification considered all your data. Within each segment, specify the closed cases that best reflect the completed fields and field values you’d like to use for case classification. This example data trains each model with data that represents best practices for your cases.
You can now create models that align with the way you organize your business or focus on a particular segment of your business. The Case Classification setup page redesign has a new section for multiple models.
Heard about Einstein Case Classification but aren’t sure it will work for your data? Enterprise, Performance, and Unlimited Edition customers can get to know it by creating a case classification model!
You can now include external data to better predict outcomes. Some predictive models require information that is not found in Salesforce, such as explanatory variables stored in outside data sources, or computed fields. Let's say you're building an opportunity time-to-close predictive model. You want to use counts of opportunity team members and qualified tasks associated with that opportunity record. Creating these aggregations using Einstein Analytics data prep is simple. Quickly build the predictive model as an Einstein Discovery story. When scoring individual records back in Salesforce, tell the predictive model how many tasks or team members are connected to the opportunity. Instead of writing code or modifying the data model, simply build a dataset that refreshes on a frequent basis where this aggregation is performed. Make sure to map the predictive model fields that are not natively found in Salesforce back to the dataset.
Get live prediction scores interactively, including top factors and actionable insights. You can now use What Could Happen insights to perform “What If” simulations on a row of data. What Could Happen insights replace the previous Predictions & Improvements insights with a streamlined, interactive interface.
Now you can revisit and work with previous story versions at any time. Each time a user creates a story with updated settings, Einstein saves a snapshot of the previous version under the same name. You can open and work with previous versions of a story.
Einstein Discovery now adds a third modeling approach known as XGBoost, which implements gradient boosting learning algorithms. Previously, Einstein Discovery relied on two types of models to predict outcomes: regression and GBM models. Now when you create a story, Einstein Discovery generates predictions using all three types of models. Einstein then shows the results of the one model that performed best. You get the best of all approaches.
When you create a story with a binary outcome, Einstein Discovery now generates it faster by automatically selecting just the features and variables required to build the model. The story creation process is streamlined and quicker.
Before you create a story for the first time, you can now quickly decide ahead of time which fields to include. With manual story setup, Einstein now shows you correlations between dataset columns and the outcome variable. Previously, correlations were visible only after initial story creation.
With increased Einstein Discovery limits, you can now create stories from Einstein Analytics datasets containing up to 100 million rows of data. Previously, the maximum was 20 million rows.
In certain cases, model deployment no longer requires selecting a Salesforce object. This exception applies only if you intend to use the model just for programmatic predictions via the Einstein Prediction Service, or for dataset scoring via the prediction component in Einstein Analytics dataflows. In all other cases, selecting a Salesforce object during deployment is required.
For insights associated with text fields, you can easily drill down into the underlying details by clicking the Explore button. Einstein displays a Lens view so that you can interactively explore the data for this insight.
We retire Einstein Discovery Classic in Spring ’20. Einstein Discovery Classic will be replaced with the new experience for all orgs starting in Spring ’20.
With Einstein Voice Assistant, your users can get more done while they’re on the go by speaking in to their mobile phone. From the new Salesforce mobile app, they can ask Einstein to make changes to Salesforce, like logging events, creating contacts, and updating opportunities.
Now it’s easier for your managers to evaluate referrals. Einstein Referral Scoring helps you build a prediction that scores referrals in Salesforce org. Your users can view referral scores on the Einstein Top Referrals component.
For Salesforce orgs in the European Union (EU), Einstein Object Detection data is now stored locally.
You can now choose which users in your Salesforce org can access Einstein Object Detection.
You can disable object detection models that you are in the process of building. You can also disable underperforming models and models that aren't used. You can always enable a model later.
When you build a prediction, you no longer need a field that answers your prediction question. As long as the records on the object that you base your prediction on have the data, you can use filters instead.
Now you and your users can gain insights on which predictors are most influential at the record level, not just for the entire model. You can show top predictors on records to see which field values have the biggest impact on a specific record’s score.
Turn on Einstein Prediction Builder without spending a dime to see how your business can benefit from custom predictions. Decide later whether to upgrade.
Learn as you build with the Trusted AI sidebar help. Trusted AI helps you identify and mitigate potential bias in your data, making your predictions less error prone.
Now you can track the aggregate usage of Next Best Action in a Salesforce org. Analyze metrics at the strategy and the recommendation level. Compare the relative performance of two strategies. For a given calendar month, see the total number of recommendations an org’s strategies served. Find out how many served recommendations agents accepted and rejected.
Declaratively create recommendations from the records of any Salesforce object, either standard or custom. Serve recommendations from accounts, contacts, opportunities, and products. Load and filter the records of any Salesforce object, and convert them into recommendations at the end of the strategy. Previously, when you created recommendations declaratively, you were limited to using only Recommendation objects. You could programmatically convert the records of other objects into recommendations, but you used Apex code with the Generate element.
Now you can launch a flow when a user rejects a recommendation, which gives you more flexibility. For example, a flow can run an automated process, write to another system, or create a reminder email when a recommendation is rejected. Previously, Next Best Action launched a flow only when a user accepted a recommendation.
To address HIPAA requirements, Next Best Action has made it possible to encrypt recommendation description information using Shield Platform Encryption.
Security, Privacy, and Identity
To address HIPAA requirements, Next Best Action has made it possible to encrypt recommendation description information using Shield Platform Encryption. When you use probabilistic encryption, you can’t use encrypted fields like Recommendation Description when you specify conditions to load recommendations. When you use deterministic encryption, you can use encrypted fields in load conditions only with the equals or not equals operator.
Calls to the Einstein Object Detection training APIs now return more descriptive error messages.
Calls to the Einstein Language APIs that return model metrics information now contain the macroF1 field, the precision array, and the recall array.
Build and run predictions in a sandbox org to experiment with different settings.