No matches found
Try choosing different filters or resetting your filter selections.
Einstein Discovery: Custom Improvements Text, Predict Outcomes in Process Automation Formulas, High Cardinality
Customize user prompts for improvements. Use the new Predict function to get predictions in process automation formulas. Determine whether your data has a normal distribution with the new QQ plot.
Rights of ALBERT EINSTEIN are used with permission of The Hebrew University of
Jerusalem. Represented exclusively by Greenlight.
-
Guide User Actions with Custom Improvement Text
Einstein Discovery improvements are suggested actions that users can take to improve predicted outcomes. Previously, these suggestions used generic, impersonal text. Now you can customize this text to provide targeted, more specific instructions. You can communicate your organization’s recommended practices within the context of your business operations using a tone that reflects your organization’s culture. Direct your users to the best possible outcomes. -
Predict Outcomes in Process Automation Formulas
Embed predictive intelligence in your process automation formulas with the Einstein Discovery PREDICT function. Now you can make decisions in your automation logic based on predicted outcomes from Einstein Discovery predictions available in your Salesforce org. For example, in an approval process, a formula can determine whether a predicted outcome meets a threshold required for automatic approval. The PREDICT function is available when defining formulas for Next Best Action, validation rules, flows (screen, headless, and invocable), processes (in Process Builder), workflow rules, approval processes, predefined field values, field update actions, and default values. -
Enable High Cardinality for One Variable in a Story
High cardinality variables can be hard to interpret and visualize due to their high number of unique values. For this reason, Einstein Discovery ignored unique values above 100 in these variables or grouped them into a reserve category. Now you can add high cardinality variables to a story. Einstein Discovery automatically alerts you when it detects variables containing more than 100 unique values. You can either let Einstein Discovery handle these variables as before, or control how the variables are used. -
Visually Determine Whether Your Data Is Normally Distributed
For regression models, one of the key assumptions is that the residual errors for the outcome variable are normally distributed. Use the new QQ (quantile-quantile) plot to quickly check this assumption and determine whether and how residual errors depart from normality. If the QQ plot shows your residual errors to be approximately linear, then you can be confident that your model satisfies the normal distribution assumption. -
Update Your Models Regularly with Automatic Refresh
Over time, a deployed model's performance can drift. The model becomes less accurate in predicting outcomes due to changes in the business environment, data, or requirements. To combat drift, refresh your model occasionally by adjusting story settings, retraining it on newer data, and redeploying it. With automatic refresh, you can now schedule weekly or monthly model updates. Unattended refreshes give you peace of mind knowing that your models are regularly updated to reflect the latest, best version. -
Stay Informed with Dataset Change Notifications and Applied Filters
We've made it easier to get more accurate insights with notifications of changes to the underlying dataset, and at-a-glance visibility of variable filters. When a story is updated, dataset changes that impact your story triggers a notification banner. The alert prompts you to review your settings and consider creating a new story version based on the latest available data. Filters applied to story variables are now more easily seen within story settings, so you can easily focus or expand your data analysis -
Embed Predictions in Your Dataset During Data Prep (Pilot)
Give your predictions direct-to-dashboard visibility. Use the new Discovery Predict transformation for Data Prep to calculate and store predictions in your dataset. You can even store descriptions of top predictors and improvements. When you run a recipe with an Discovery Predict node, Einstein estimates and saves predicted outcomes on a row-by-row basis. Populate your datasets with predictive and prescriptive intelligence to quickly evaluate predictions across a large set of data, assess multiple models before deploying them into production, and aggregate this information in a dashboard. -
Minimize Disparate Impact in Your Stories with Sensitive Variables
In Einstein Discovery, you can identify and use variables that have the potential of bias in your story. It’s important to monitor these sensitive variables to ensure that they don’t negatively influence story insights. If recent story updates are affecting sensitive variables, Einstein Discovery may warn you of disparate impact, meaning other variables are being treated unequally in your model. A model trained on data with disparate impact can produce biased predictions. -
Jump-Start Story Creation Using Templates
Not sure where to begin with identifying winning deals or shortening deal cycles? Let the templates build the app, prep and load data, then create a story version using the best practices to meet your goals. Review completed story findings, and then customize the story to meet your needs. -
Get Better Tuned XGBoost and GBM Models
We improved the accuracy of models that are based on the XGBoost and GBM algorithms. We tuned internal settings for these models, so this change is automatic. If you compare model metrics between a model created in this release and a model created in a previous release, the metrics may differ. -
Implement External Models in Your Salesforce Org (Pilot)
Augment your predictive powers with externally built models that you can quickly start using in your production environment. In addition to the amazing models that Einstein Discovery builds from your stories, you can now upload and deploy models that are created outside of Salesforce. Your data scientists can use their favorite modeling tools to design, build, test, and tune TensorFlow models. Then, simply upload and deploy these carefully crafted models into your Salesforce org. Your users are up and running quickly, and using the model’s predictions to achieve better outcomes. -
Build Accurate Models Using Random Forest Algorithms (Pilot)
Einstein Discovery now adds a fourth type of model that is based on a modeling algorithm known as random forest. Einstein Discovery uses this supervised learning algorithm to create highly accurate models via multiple decision trees, randomization, and other optimization techniques. You can compare a random forest model with other types of models to determine whether this algorithm provides better accuracy for your story. -
Annotate Story Versions with Descriptions
Ever forget the exact changes that were made in a story version? Maybe why a change was made or who made the change? Now you can add informative and detailed descriptions that provide clarity, accountability, while avoiding redundant work. -
Cancel Story Creation During Analysis
Ever wanted to include a last-minute change before clicking Create Story, but had to wait for it to finish just to delete it? You now have the option of interrupting a story after submitting it. No more searching for the story then manually deleting it. -
Load Einstein Discovery Stories Faster
Stories with a lot of insights can take longer to load. Einstein Discovery now opens with a story’s top 15 insights, then loads more insights as you search for them. Top insights include all first order insights followed by the second order to complete the list. -
Focus Your Predictions in Einstein Prediction Service
The default behavior has changed for predict API calls in Einstein Prediction Service. Previously, the response included predictive factors and improvements by default. Starting in this release, the predict function returns just a single prediction value. If you want predictive factors and improvements, you must ask for them in the request body of your API call.