Process at a glance | Experimental custom models | Historical-only and future-required drivers | Deterministic Authoring Mechanics | Automatic Authorization | Example Dates and Retries | Source Repair Diagnostics | Manual Review | Reviewed Hierarchy Rules | Custom models in Agent iterations and updates | Future-forecast checks | Supported historical growth | Accuracy allowance | Least-issues fallback | Standard forecasts and saved averages | Reconciliation | Iterative Agent forecasts | Updated Agent forecasts | Updating an earlier update | Changing hierarchy membership | Evidence and limitations | Historical actuals | Backtest weighted MAPE

Univariate vs Multivariate Models | Adaptive Daily ARIMA | Global vs Local Models | Ensemble Models | Multistep Horizon Models | Leveraging the Tidymodels Framework

Why an AI agent in finnts? | What the agent produces | Approved Custom Models | Updating Approved Choices | Prerequisites | End-to-end: first run with the AI agent | 1) Create a project | 2) Bring data | 3) Define the LLM | Alternative: GitHub Copilot CLI | 4) Create the agent run | 5) Let the agent iterate to a best run | 6) Retrieve results | Ask questions about your forecast results | How it works | Example questions | What data sources are available | Tips for effective questions | Updating with new data (production loop) | Hierarchies (optional) | External regressors (xregs) | Parallelism knobs | Reading artifacts directly (optional)

Selecting a Hierarchy | Standard Hierarchy | Grouped Hierarchy | Reviewed Custom Rules | External Regressors | Spark Parallel Processing | Corner Cases in Forecast Reconciliation

Local Machine | Parallel AI Agent Workflows | Within Azure using Spark

Chronos2 | Authentication and Setup | Using Chronos2 within finnts | Data Size Requirements and Automatic Padding | Global Model Support | Chronos Bolt Base | Using Chronos Bolt Base within finnts | Chronos Bolt Tiny | Using Chronos Bolt Tiny within finnts | TimesFM | Using TimesFM within finnts | TimeGPT | Azure-hosted TimeGPT | Option A: Environment variables | Option B: Using nixtlar helper | Nixtla API | Option A: Environment variable | Quick start with nixtlar (standalone) | Using TimeGPT within finnts | Long-Horizon Forecasting | Hyperparameter Tuning

1. Bring Data | 2. Set Up Agent | 3. Iterate Forecast | 4. Analyze Forecast Output | 5. Update Forecast

Optional Dependency | Feature Selection Techniques | Target Correlation | Leave One Feature Out (lofo) | Boruta | Model Variable Importance | Voting Process

Missing Data and Outliers | Understanding "_original" Columns | Box-Cox | Differencing | Date Features | Lags, Rolling Windows, and Polynomial Transformations | Custom Approaches | Model Specific Preprocessing

Overview | Key Capabilities | Getting Started | Installation | Basic Workflow | 1. Import Your Glint Export | 2. Get a Question-Level Summary | 3. Focus on Specific Questions | 4. Explore Response Distributions | Multi-Cycle Trend Analysis | Manager-Level Analysis | Roll Up to Manager Level | Demographic Analysis | Segment by Employee Attributes | Pre-Joining Attributes for Multiple Analyses | Attrition Risk Analysis | Segment Attrition by Demographics | Correlation and Factor Analysis | Understand Which Items Move Together | Factor Analysis | Working with Comments | Full-Text Comment Search | Convert to Long Format for NLP | Separating Quantitative and Qualitative Data | Privacy and Data Handling | Minimum Group Sizes | Local Processing | Additional Resources

Get Data and Set Run Info | Prep the Data | Train Individual Models | Train Ensemble Models | Final Models | Get Forecast Results

Historical Values Only | Historical and Future Values

Manage Dependencies | Azure ML Pipelines