Forecasting and financial planning with AI

Day 1: Theory and basics

Introduction to artificial intelligence in finance

• Current Product Portfolio: Claude Opus 4.7, Sonnet 4.7, Haiku 4.5, and Their Roles in Finance.
• Trends for 2026: From specialized models to foundation forecasters, from chat to agent-based systems.
• Regulatory framework: EU AI Act (AI Literacy effective February 2, 2025; GPAI effective August 2, 2025; High-Risk Annex III effective December 2, 2027) and the GDPR.

An Overview of Traditional Methods

• ARIMA, SARIMA, Holt-Winters, Regression: When Are They Still Useful?
• Limitations of traditional models when dealing with volatile, multivariate financial data.

Time-Series Foundation Models 2026

• Multivariate Zero-Shot Forecasting (Amazon Chronos-2).
• Architectures and Use Cases (Google TimesFM and Salesforce Moirai).
• Use Cases with Nixtla TimeGPT-2 (Mini, Standard, Pro).
• Comparison of accuracy, costs, and implementation effort.

Claude as a Co-Pilot in Forecasting Processes

• Realistic role: Claude is not a self-forecaster, but rather a facilitator and validator.
• Claude Code: Generate and execute Python and R snippets for data preparation, model training, and evaluation.
• Claude Projects: Keep datasets, assumptions, and prompts permanently within the project context.
• Claude Artifacts: Build interactive forecast charts and dashboards directly in your browser.
• Claude Connectors and MCP: Integration with internal data sources and ERP systems.

Day 2: Practice and application

Data Preparation and Feature Engineering with Claude

• Frequency harmonization, missing values, outliers, seasonality.
• Reproducible Pipelines: Prompts, code snippets, and validation checks in the project.

Hands-on: Creating Forecasts

• Workshop A: Foundation Model vs. classical ARIMA with Claude's help, same dataset, documented comparison.
• Workshop B: Cash Flow and Burning Rate Forecast with Claude Code.
• Workshop C: Scenario Logic (Base, Optimistic, Pessimistic) as an Artifact.

Forecasting in Excel with Claude

• Cowork and the Claude in Excel Add-in: Align forecast assumptions directly in the workbook.
• Audit-ready documentation of the assumptions and model parameters.

Implementation within the Company

• Governance: Prompt versioning, KPI catalog, audit trail.
• GDPR-compliant options: Anthropic API with DPA, Claude via AWS Bedrock Frankfurt, EU Data Residency.
• Credit Checks as a High-Risk Application (Annex III, Deadline: December 2, 2027): Requirements and Preparation.
• AI Literacy in the Team: Concrete Measures for Finance Departments.
• Ethical and Legal Framework, Outlook, and Future Prospects.