Prediction
A prediction is the scored output of a machine learning model that estimates a future outcome, a likelihood, or a classification.
Definition
A prediction is the scored output of a machine learning model that estimates a future outcome, a likelihood, or a classification. In Salesforce, predictions appear on records as a number (probability between 0 and 1), a percentage (1 to 99), or a categorical label, and they update as the underlying record changes. Predictions can come from built-in Einstein features (Lead Scoring, Opportunity Scoring, Case Classification), from custom Einstein Prediction Builder models, from Tableau-trained models surfaced into Salesforce, or from third-party models accessed through external connectors.
The unifying property is that a prediction is the output of a discriminative or probabilistic model rather than the output of a generative one. Predictions are deterministic: the same inputs produce the same output. They are interpretable: each prediction can be traced back to the features that influenced it. They are typed: a probability is a number between 0 and 1, a classification is one of a fixed set of values. These properties make predictions easier to consume in flows, reports, and routing rules than the free-text outputs of generative AI features.
Why a prediction is a different shape of AI than a generated response
Prediction shapes: probability, score, classification, regression
Probability predictions output a value between 0 and 1 representing how likely an outcome is (lead conversion probability, opportunity close-won probability). Score predictions output a normalized number, often 1 to 99 (Einstein Lead Score, Einstein Opportunity Score). Classification predictions output one of a fixed set of values (Case Type prediction returns "Billing" or "Technical" or "Other"). Regression predictions output a continuous numeric value (predicted deal amount, predicted resolution time). Each shape has different downstream consumption patterns; a probability fits into routing thresholds while a classification fits into branching logic.
Where predictions come from in Salesforce
Four common sources. Built-in Einstein features train Salesforce-managed models on the org's data and write the prediction to a standard or custom field (Lead Score, Opportunity Score). Einstein Prediction Builder lets admins build custom predictions through a no-code wizard. Einstein Discovery produces both predictions and a narrative explaining them. Tableau models trained outside Salesforce can be deployed into the org and write predictions back. Each source has different setup work and different ongoing maintenance, but the resulting prediction field behaves the same way downstream regardless of where the model lives.
How predictions integrate with the platform
Predictions surface as fields on records. List views sort by prediction. Reports filter on prediction. Flows branch on prediction. Apex reads prediction. Lightning components display prediction. Routing rules use prediction to assign records. The flexibility comes from the prediction being just another field; the platform does not need new constructs to consume it. This is intentional and is the largest reason Salesforce AI integrates so cleanly into existing org behavior; downstream automation does not need to know whether a field value came from a human, an integration, or a model.
Refresh cadence and the staleness question
Predictions stale over time as the underlying record changes and as the model drifts. Built-in Einstein predictions retrain on Salesforce's schedule (typically weekly or monthly). Per-record predictions update when the record updates, not on the retrain cadence. So a record edited today gets a prediction that reflects today's record state against last week's model. The combination means predictions are usually fresh enough for operational use but not necessarily for time-critical decisions. For decisions where seconds matter (real-time routing on inbound contact), pair the prediction with a current-record check rather than relying on the prediction field alone.
Top Predictors and the explainability surface
Every Einstein prediction surfaces a Top Predictors component that shows which fields most influenced the score on a specific record. The component is the explainability layer. It helps users trust the prediction ("this opportunity is scored 80 because the deal is large, the contact engagement is high, and the close date is near") and helps admins debug surprising scores. Without Top Predictors, predictions read as magic numbers users distrust. With it, predictions become a conversation starter for sales managers reviewing pipeline. Surfacing Top Predictors on the record page is a small layout change with outsized adoption impact.
Predictions vs forecasts vs recommendations
Predictions estimate a specific outcome for a specific record. Forecasts aggregate predictions into a rolled-up view (Einstein Forecasting predicts commit at the rep, team, and org level). Recommendations rank multiple options for a given context (Einstein Product Recommendations ranks products for a shopper). Each is a different shape: predictions are per-record, forecasts are per-aggregate, recommendations are per-context. Salesforce features blur the line, but the distinction matters when choosing the right tool. A request to "predict which leads to call first" is a prediction (score each lead) feeding a recommendation (rank by score).
When a prediction is the right tool and when it is not
Predictions are excellent when the outcome is discrete and historically labeled (will close vs lost, will convert vs not, will churn vs stay), the input data is consistent, and the business decision can act on a probability or category. Predictions are bad when the outcome is open-ended (what should I write to this customer), the input data is sparse (new product launch with no history), or the decision needs explanation in plain language. For open-ended outcomes use a generative AI feature instead; for sparse data, use rules until enough data accumulates to train.
How to build a prediction that earns trust on day one
The prediction that gets adopted has three properties: it predicts something the user cares about, it is accurate enough to bet on (above 70 percent), and it shows its work through Top Predictors. The prediction that gets ignored fails on at least one. Plan for all three from day one.
- Pick a target the user has a decision around
The prediction must inform a specific decision (which lead to call first, which case to escalate, which customer to renew). A prediction the user does not act on is academic and gets ignored.
- Pick the right source for the model
Built-in Einstein if the target is one of the canned ones (Lead Scoring, Opportunity Scoring). Prediction Builder for custom targets. Einstein Discovery if you also need narrative explanations. Tableau if a data team will own model maintenance.
- Audit training data for leakage and labels
Exclude any feature filled in after the outcome was decided. Confirm labels are consistent. Both audits matter; both are skipped more often than they should be.
- Train and review accuracy on a holdout set
Above 70 percent accuracy is a deployable threshold for most decisions. Below 60 percent is rarely worth shipping. Between 60 and 70 is a judgment call; pilot with one user team to see if the lift is real.
- Add the prediction field to list views and record pages
Without the field on relevant list views, users cannot sort by prediction. Surface it on the record page near where the decision is made.
- Add Top Predictors to the record page
Top Predictors is the explainability surface. A prediction without it reads as a magic number users distrust; with it, the prediction becomes a conversation about the underlying signals.
- Schedule monitoring and a retraining cadence
Weekly accuracy review for new predictions, monthly for stable ones. Retrain when accuracy drops more than 5 percentage points from baseline. Without monitoring, drift goes unnoticed for quarters.
Probability, score, classification, or regression. Drives downstream consumption pattern (threshold, branching, arithmetic).
Built-in Einstein, Prediction Builder, Einstein Discovery, Tableau, or external connector. Drives setup work and maintenance burden.
The explainability component. Surface on the record page; without it, predictions read as magic.
How often the model retrains. Weekly or monthly typical; depends on how fast the underlying business shifts.
Production-quality threshold per use case. 70 percent is a reasonable floor for most decisions; safety-critical decisions need higher.
- Data leakage in training features makes a model look brilliant in evaluation and fail in production. Audit features for "was this knowable at prediction time" before evaluating accuracy.
- Predictions without Top Predictors read as magic. Users distrust magic; the small layout change to surface Top Predictors has outsized adoption impact.
- Predictions stale as the model drifts and the underlying world changes. Monitoring catches drift; without monitoring, the model degrades quietly.
- A prediction with 80 percent accuracy is wrong 20 percent of the time. Downstream automation that treats predictions as certain creates incidents.
- Picking the wrong AI shape (using an LLM to score, or a predictive model to draft text) is slower, more expensive, and less accurate than picking the right tool.
Trust & references
Cross-checked against the following references.
- Salesforce Einstein overviewSalesforce
- Salesforce AI strategySalesforce
Straight from the source - Salesforce's reference material on Prediction.
- Einstein Prediction BuilderSalesforce Help
- Einstein Platform OverviewSalesforce Help
- Einstein DiscoverySalesforce Help
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. What is a Prediction in Einstein?
Q2. How can admins create custom predictions?
Q3. Why validate predictions?
Discussion
Loading discussion…