Predictive Use Case Framing Template

{ "Title": "Predictive Use Case Framing Template", "IntroductionHtml": "

Use this short, structured form to frame predictive analytics work before modeling begins. Good framing reduces wasted effort, prevents data leakage, clarifies who will act on predictions, and helps you choose the right evaluation metric and monitoring plan. Fill the required fields and attach this saved use case to experiment trackers, model registries, or deployment requests.

\n

How to use

\n

Be specific. Define the target variable precisely, describe the decision the prediction will support, translate model performance into business terms, and record how humans will review or override outputs. Two short examples are included at the end.

", "SubmitLabel": "Save use case", "SuccessMessage": "Use case saved. You can return to edit, export, or attach this template to modeling and deployment tasks.", "DataType": "PredictiveUseCase", "SchemaVersion": "1.0", "Fields": [ { "Key": "businessObjective", "FieldType": "textarea", "Label": "Business objective", "Required": true, "HelpText": "Describe the measurable business goal this predictive model should support (who benefits and how). Be concrete (e.g., reduce 30-day churn by X%, reduce stockouts, reduce false alarms)." }, { "Key": "primaryStakeholders", "FieldType": "text", "Label": "Primary stakeholders", "HelpText": "Names, teams, or roles that will use or be affected by the predictions (e.g., collections team, demand planning, safety operations)." }, { "Key": "decisionContext", "FieldType": "textarea", "Label": "Decision context", "HelpText": "What decision or action will the prediction inform? Include timing, frequency, and who acts on the prediction." }, { "Key": "targetDefinition", "FieldType": "textarea", "Label": "Target variable (explicit)", "Required": true, "HelpText": "Exactly define the target, units, prediction horizon, and how it's measured. Example: '30-day churn: customer unsubscribed within 30 days (binary)'. For regression targets include units (e.g., weekly units sold)." }, { "Key": "businessImpact", "FieldType": "textarea", "Label": "Business impact", "HelpText": "Describe benefits of correct predictions and harms of errors in business terms (revenue, safety, operational cost, customer experience). Quantify if possible." }, { "Key": "valueCurve", "FieldType": "textarea", "Label": "Value curve / utility function", "HelpText": "Describe how prediction scores map to business value or action (thresholds, graded utility, or a short table). For example: 'Top 5% receives manual review -> expected $500/save per TP; cost per FP = $20'." }, { "Key": "metricChoice", "FieldType": "select", "Label": "Preferred evaluation metric", "Options": [ { "Value": "precision_recall", "Label": "Precision & recall" }, { "Value": "auc", "Label": "AUC / ROC" }, { "Value": "mae_rmse", "Label": "MAE / RMSE (regression)" }, { "Value": "f1", "Label": "F1 score" }, { "Value": "custom_cost", "Label": "Custom cost matrix" } ], "HelpText": "Select the most relevant metric to evaluate model performance. Choose custom cost matrix when business costs drive decisions." }, { "Key": "acceptablePrecision", "FieldType": "number", "Label": "Acceptable precision (if applicable)", "HelpText": "Enter minimum acceptable precision as a decimal (0–1). Leave empty if not applicable." }, { "Key": "acceptableRecall", "FieldType": "number", "Label": "Acceptable recall (if applicable)", "HelpText": "Enter minimum acceptable recall as a decimal (0–1). Leave empty if not applicable." }, { "Key": "costMatrix", "FieldType": "textarea", "Label": "Cost matrix or error costs", "HelpText": "If using a custom cost approach, list costs or savings for TP, FP, FN, TN (example: TP=+100, FP=-10, FN=-500, TN=0)." }, { "Key": "humanDecisionPoints", "FieldType": "textarea", "Label": "Human decision points & override rules", "HelpText": "Where will people review or override predictions? Specify roles, steps, expected review rate, and allowed overrides." }, { "Key": "deploymentConstraints", "FieldType": "textarea", "Label": "Deployment constraints", "HelpText": "Performance, latency, integration, regulatory, budget, or infrastructure constraints that affect how the model can be used." }, { "Key": "monitoringNeeds", "FieldType": "textarea", "Label": "Monitoring needs & frequency", "HelpText": "Which metrics should be monitored in production and how often (e.g., daily precision@threshold, weekly data drift). Include alert thresholds if known." }, { "Key": "successCriteria", "FieldType": "textarea", "Label": "Success criteria & acceptance tests", "HelpText": "Concrete conditions and tests that must be met to consider the model successful (e.g., precision >= 0.6 on hold-out for 30 days, no increase in bias metrics)." }, { "Key": "dataRequirements", "FieldType": "textarea", "Label": "Data requirements & availability", "HelpText": "List input features, required sample size, frequency, freshness, retention windows, and any known gaps or collection plan." }, { "Key": "dataLeakageRisks", "FieldType": "textarea", "Label": "Potential data leakage or bias risks", "HelpText": "Note any fields or time windows that risk leaking future information into training, and known biases in labels or sampling." }, { "Key": "privacyEthics", "FieldType": "textarea", "Label": "Ethical, legal & privacy notes", "HelpText": "Any personally identifiable data, consent issues, fairness concerns, or compliance/regulatory constraints. Describe mitigation plans if required." }, { "Key": "exampleRiskScoring", "FieldType": "textarea", "Label": "Example: Risk scoring (example entry)", "HelpText": "Short example entry: Business objective: reduce costly defaults. Target: 90-day default (binary). Metric: precision@top10%. Value: manual review of top 5% reduces losses. Human review: credit team reviews top 5%. Monitoring: weekly precision and application rate." }, { "Key": "exampleDemand", "FieldType": "textarea", "Label": "Example: Product demand prediction (example entry)", "HelpText": "Short example entry: Business objective: reduce stockouts. Target: weekly demand per SKU (count). Metric: MAE; Acceptable MAE: 10 units. Deployment: nightly batch predictions; Monitoring: daily stockout rate and weekly MAE." }, { "Key": "notes", "FieldType": "textarea", "Label": "Additional notes", "HelpText": "Anything else worth capturing about this use case." } ] }