← Models · How we model
From annual report to live model
A practitioner's account of building a fully AI-orchestrated 3-statement modelling pipeline: filing ingestion, structured extraction, terminology normalisation, and human-gated forecast approval.
Financial modelling remains constrained by workflows designed for spreadsheets rather than financial knowledge. Annual reports must be repeatedly searched, interpreted, reconciled, and re-entered into models despite representing structured financial statements. As a result, experienced analysts spend most of their modelling effort reconstructing information that already exists instead of evaluating the business itself.
No single AI model performs financial modelling end-to-end.
Dorrsum combines specialised AI capabilities into a structured workflow, where each stage verifies or prepares the next. Extraction feeds verification. Verification enables terminology normalisation. Normalised financial data populates the model, which then supports forecasting through explicit analyst approval gates.
The differentiation is not any individual AI model. As foundation models continue to improve and become increasingly commoditised, the value lies in how they are combined into a reliable financial modelling workflow that is transparent, auditable, and resilient.
Every filing is transformed into verified financial knowledge before any data reaches the model. The spreadsheet becomes the destination, not the starting point.
Why it matters
Historical actuals, forecasts, ratios, and future updates all work from the same trusted source of truth.
The pipeline targets financial concepts rather than spreadsheet coordinates. Financial labels are standardised before population, allowing the workflow to remain resilient even when model layouts or filing terminology change.
Why it matters
Automation follows financial meaning instead of Excel positions, making the model easier to evolve and maintain.
| Filing label (as seen in 10-K) | Company / context | Model standard label |
|---|---|---|
| Total net sales | Apple FY2025 | Gross Revenue |
| Cost of sales — Products / Cost of sales — Services | Apple FY2025 | Cost of Sales (COGS) |
| Income before provision for income taxes | Apple FY2025 | Pretax Profit (EBT) |
| Provision for income taxes | Apple FY2025 | Tax Expense |
AI prepares historical data and proposes forecast assumptions, but every projected value requires explicit analyst approval before entering the model.
Why it matters
Judgment remains visible: every accepted, modified, or rejected assumption is logged against the analyst who made the call.
