Files
Excel, CSV, PDF and Word inputs, including multi-tab workbooks with merged cells, headers in unexpected places and provider-specific layouts.
AI Agents
Turn messy inputs into structured reporting data.
Reporting teams spend significant time turning inconsistent files, documents and system outputs into usable reporting data. Datox extraction agents read the inputs you already receive, produce structured records, and keep a reference back to exactly where each value came from.
Inputs
What extraction agents read
Extraction covers files, documents and live system data, so the same agents serve spreadsheet-heavy and system-heavy reporting alike.
Documents
Statements, reports, disclosures, invoices and factsheets where the reporting data exists only inside a document.
Systems
APIs, databases, SFTP drops and scheduled exports where column order or naming varies between systems and versions.
Evidence
No value without a source
An extracted number that cannot be traced is a liability. Datox stores the origin of each value at the moment of extraction, so later review and audit do not depend on rerunning anything.
Recorded per extracted value
- The input file, version and reporting period it belongs to.
- The precise location, sheet and cell, table and row, or document region.
- A confidence indicator reflecting how clear the read was.
- The extraction run and configuration that produced it.
Behaviour
Designed not to guess
The failure mode that matters in reporting is a confident wrong answer. Datox extraction agents are configured to surface uncertainty rather than resolve it themselves, because an escalated field costs minutes while an undetected error costs far more.
- Ambiguous or unreadable fields are flagged for review with the source shown.
- Layout changes in a familiar source are detected and raised.
- Missing expected sections or tabs are reported rather than silently skipped.
- Confidence thresholds are configurable per workflow and per field importance.