Every number, traceable to the cell it came from.
Consolidata reads documents arriving from departments and subsidiaries in any format, with no fixed template. It maps them to the organization’s common data model, validates them against accounting rules, and serves them as a single structured source.
Low-confidence values go to a person for approval before they reach a report.
Much of a company’s data lives outside its ERP
Organizations run standard processes such as purchasing, bookkeeping and financial consolidation on central ERP systems. Yet a large share of their data still sits in Excel sheets, PDF reports, Word documents, scanned papers and emails.
Today these documents are combined by hand: headers are matched one by one, units and formats are fixed, consistency is checked. The work is slow, depends on specific people and is prone to error. The origin of a figure is often lost along the way, which in financial reporting is a direct audit risk.
Ad-hoc report requests
When management asks several units for the same report, each one sends it in its own format, with its own headers and units.
Different system landscapes
Units on different software, or on no central system at all, send their data to headquarters as files.
Documents that never enter a system
Signed payment orders, bank statements, contracts and guarantee letters usually remain documents only.
From upload to report in four stages
The platform is built on a single processing pipeline that runs from uploading a document to presenting output to the user. Each stage passes its results forward together with their source.
-
Ingestion
Documents are tagged with their source unit, company and period. The original file is archived unchanged.
-
Extraction
Text, tables and page layout are extracted by document type. Digital and scanned documents are processed separately; tables, paragraphs and footnotes are told apart.
-
Mapping and validation
Line items are mapped to the common data model; currency, scale, date and number formats are normalized, and financial consistency rules are applied.
-
Output
The structured data powers consolidated reports, analysis views and, in a later phase, natural language queries.
Scope. The infrastructure is domain-independent; financial and accounting data is the first application area. Sales, inventory and HR can be added as separate modules on the same foundation. Consolidata does not replace the ERP. It works alongside it and structures the data the ERP does not cover.
What we do differently
OCR, layout analysis and LLM-based extraction are not new on their own. The novelty lies in combining them inside a validation and learning loop designed for financial reporting.
- Template-free, contextual schema mapping
- A line item’s meaning is inferred not only from its header but from neighboring columns, data types and sample values. Is “Total Revenue” net or gross?
- A mapping memory that learns
- Verified mappings are stored semantically. The language model runs only on the few ambiguous items the memory cannot resolve, so the system grows more consistent and cheaper with use.
- Validation with accounting rules
- Subtotals, balance sheet equality and scale consistency are checked automatically, catching both extraction errors and calculation errors in the source before they reach a consolidated report.
- Cell-level traceability
- Every value is stored with its source file, page or cell position and a confidence score. The original of any corrected value always stays accessible.
- Cross-document inconsistency detection
- Different names for the same concept (“Ciro”, “Net Satışlar”, “Revenue”), shifting scales and currencies, inconsistent period labels, duplicate records and missing fields are flagged before data is merged.
- Natural language queries
- Users type their question instead of writing queries or building filters, and get answers linked back to the source documents.
- On-premise operation
- Open-source local models process data without it leaving the organization, making the platform viable under data protection law (KVKK) and sector regulations.
Compared with existing solutions
| Solution type | Examples | What it provides | What is missing |
|---|---|---|---|
| Document extraction platforms | ABBYY, Rossum, Azure Document Intelligence, AWS Textract | Field extraction from fixed-type documents such as invoices and receipts | Does not map free-form reports to a common schema or interpret financial meaning |
| ERP consolidation modules | SAP Group Reporting, Oracle FCCS | Consolidation of structured data | Requires data to already be in the ERP or arrive in a fixed template |
| Financial reporting platforms | Workiva, OneStream | Reporting process management | Data entry still depends on templates and manual preparation |
| General-purpose language models | Cloud LLM services | Flexible document interpretation | Costly, inconsistent and unverifiable; cloud use carries privacy risk |
Same concept, different meaning by sector
The hard part is not reading a document but interpreting what a value means and mapping it correctly. The Turkish Uniform Chart of Accounts (Tek Düzen Hesap Planı) serves as a shared reference: sector-specific items are linked first to account groups, then to the organization’s reporting model.
| Sector | Typical documents and data | What makes mapping hard | Effect on cash flow |
|---|---|---|---|
| Construction and real estate | Progress payment reports, rent and property tables, subcontractor agreements | Revenue is recognized by percentage of completion; progress billing and collection are separate concepts | Advances and collection delays create large deviations |
| Energy | Production and sales reports, long-term contracts | Some amounts are FX-based; exchange rate and index clauses are buried in contract text | Future inflows depend on indices and exchange rates |
| Tourism | Occupancy and revenue reports, agency reconciliations | Strong seasonality; period definitions and prepayments vary | Cash inflows concentrate in certain months |
| Mining and manufacturing | Production, inventory and export documents | Physical units and monetary values are intertwined; exports are in foreign currency | Letter of credit and export payment terms are decisive |
| Logistics and trade | Invoice lists, account statements | The same customer appears under different trade names; record volume is high | Receivables aging is decisive |
| Finance (banking, factoring) | Credit and risk reports | Regulatory reporting formats; personal data restrictions | Data cannot leave the institution |
Mixed Turkish and English headers, abbreviations, the Turkish number format (1.234,56) and varying scale units are shared requirements across every sector.
Team
Bilkent University, Department of Computer Engineering, CS491/492 Senior Design Project.
- Ahmet Kenan Atamankenan.ataman@ug.bilkent.edu.tr
- Ece Şeşenece.sesen@ug.bilkent.edu.tr
- Mehmet Emre Şahinemresahin@ug.bilkent.edu.tr
- Nevzat Han Altınhan.altin@ug.bilkent.edu.tr
- Umut Çağın Özcancagin.ozcan@ug.bilkent.edu.tr
Supervisor: to be announced
Project documents
- Project Specification ReportComing soon
- Analysis and Requirement ReportComing soon
- Detailed Design ReportComing soon
- Final ReportComing soon