Understand Predictions and Improvements
Learning Objectives
- Understand predictions and improvements
- Navigate to a story’s predictions and improvements and explore them.
- Identify the best predicted future outcome for your scenario.
- Identify the top factors behind a prediction.
Note
The instructions in this unit assume that you have successfully created an Einstein Discovery story. Refer to the instructions in "Create a Story," in the first unit in this Trailhead module.
About Predictions and Improvements
While other story insights show historical data from past activities, predictions and improvements calculate statistically probable outcomes of future events. Einstein Discovery conducts predictive analysis to predict future outcomes, and prescriptive analysis to suggest ways in which to improve those predicted outcomes. Predictions and improvements come from the kind of regression and machine-learning analysis that data scientists conduct using advanced analytics and AI tools. Use predictions and improvements to interactively perform “what if” analyses on your data.
When you create a story with Insights & Predictions selected, Einstein Discovery also generates a model that it uses to produce predictions and improvements. You can think of a model as a sophisticated algorithm that accepts inputs (the values of explanatory variables in your data) and produces outputs (predictions and improvements). Therefore, a prediction is a derived value, produced by an Einstein Discovery model, that represents a possible future outcome. An improvement is a suggested action, based on prescriptive analytics, that a user can take to improve the likelihood of a desired outcome.
Note
What Could Happen insights allow you to get predictions and improvements using the data in your story’s dataset. Einstein Discovery also enables you to use predictions and improvements in a production environment. For instructions on how to use predictive models in your org, see Predict Outcomes and Take Actions.
Get Predictions
Let's start by seeing how Einstein Discovery can help predict CLV by industry.
In the left panel, you see a list of columns in your dataset. The Model Feature label simply means that the column is represented in the predictive model that Einstein Discovery uses to generate predictions. These are your explanatory variables (sometimes called predictor variables). The columns are sorted by the correlation percentage, which indicates the strength of the association between the explanatory variable and the outcome variable. As we saw earlier in the Variables panel, Division correlates most highly with CLV, followed by Type and Rating.Each column has a list of values to choose from. To see a prediction, click the dropdown and select a value from the list.
2. For Division, select Naval.
Einstein shows the predicted CLV, when Division is Naval, to be 20995.67. It also shows us that this selection has a negative effect on the average CLV (lowers it by 267).
Note
3. Try selecting other divisions until you find the one that contributes to the highest predicted CLV. When Division is Raw Materials, the predicted CLV is 24926.15, a positive factor that’s 721.6 higher than the average.
That’s 3930.48 higher than when Division is Naval. Quite a difference!
4. To learn what’s behind this prediction, scroll down to the waterfall chart.
This chart shows you the baseline CLV, the impact when Division is Raw Materials, the expected impact of other fields, and the overall predicted outcome. To see statistical details, hover over the bars in the chart.
You can select combinations of field values to see how the interaction among these factors affects the predicted CLV. Let’s see what happens.
5. Try selecting values for Type until you find the one that contributes to the highest predicted CLV when Division is Raw Materials.
Get Improvements
Einstein Discovery can recommend actions you can take to change a predicted outcome. Actionable variables drive improvements in Einstein Discovery. An actionable variable is a variable that people can control or influence, such as deciding which marketing campaign to use for a particular customer, or which shipping method to recommend to a customer.
If you click the Actionable button next to a column, Einstein displays actions you can take, if any, to improve the outcome in the Top Improvements box.
Resources
- Salesforce Help: Explore Predictions and Improvements
Conclusion
In this module, you learned how diagnostic insights give you a deeper understanding of the complex relationships in your existing data. You saw how related and unrelated factors affected the observed outcome, then used predictions and improvements to make better decisions about future business actions.
