Why should you care?

Models need reproducible feature snapshots; apps need prediction audit trails.

See it live — copy this example

Run in SQL Server Management Studio (SSMS) or Azure Data Studio.

USE DataVerse;
IF OBJECT_ID(N'dbo.ModelPredictions', N'U') IS NOT NULL DROP TABLE dbo.ModelPredictions;
CREATE TABLE dbo.ModelPredictions (
    PredictionId BIGINT IDENTITY PRIMARY KEY,
    CustomerId INT NOT NULL,
    ModelName VARCHAR(50) NOT NULL,
    Score DECIMAL(9,6) NOT NULL,
    PredictedAt DATETIME2 NOT NULL DEFAULT SYSUTCDATETIME()
);
INSERT INTO dbo.ModelPredictions (CustomerId, ModelName, Score)
VALUES (1, 'churn_v3', 0.82);
SELECT ModelName, AVG(Score) AS AvgScore, COUNT(*) AS N
FROM dbo.ModelPredictions
GROUP BY ModelName;

What happened?

• ModelPredictions logs scores for monitoring drift and auditing decisions.
• Aggregates show average score per model version.

Practice next

1. Create ModelPredictions and insert samples.
2. Average scores by ModelName.
3. Add FeatureJson NVARCHAR(MAX) carefully if needed.
4. Filter Score >= 0.8 high-risk churn.
5. Add ExperimentId for A/B models.

Remember

Log predictions with model version. Keep OLTP lean. Monitor score distributions.

Churn scores in DataVerse

Marketing reads latest churn predictions per customer.

Outcome: Campaigns target high scores with an audit trail.