All case studies
FinTech / Financial Advisory AI

Qontica

Deterministic financial analysis, with zero hallucinated numbers
2025

The Problem

Financial advisory firms drown in low-value work: manually re-keying trial balances from client ERPs, reclassifying accounts by hand, building the same budget spreadsheets over and over, formatting board reports. The judgment work clients actually pay for (interpretation, planning, advice) gets squeezed into whatever time is left.

The Product

Qontica is the AI platform for corporate finance advisory. Upload a trial balance PDF from any ERP (TeamSystem, Zucchetti, Passpartout, anything) and the platform extracts accounts, codes, and amounts automatically, maps the chart of accounts, reclassifies income statement and balance sheet, computes KPIs with sector benchmarking, builds monthly budgets and industrial plans, and generates board-ready reports, including Italy's statutory early-warning frameworks (art. 2086 c.c., 6-month DSCR, debt stress tests).

The defining design decision: AI only reads and consults data, like a financial analyst would. Every calculation (ratios, reclassifications, projections) runs through deterministic, inspectable, modifiable formulas. No hallucinated numbers. Every figure traceable to the balance line or assumption it came from.

The Engineering

We engineered the platform's hybrid architecture: LLMs for what they're good at (reading messy PDFs, mapping inconsistent charts of accounts), deterministic calculation engines for everything with a number in it, because a hallucinated figure in a board report is a fired advisor. This LLM-for-language, code-for-math split is the pattern we now recommend for every financial AI system.

Multi-tenant isolation keeps each firm's client data in its own environment, with EU data residency and GDPR compliance throughout.

How the analysis pipeline actually works

A trial balance PDF from any ERP flows through the same pipeline every time: LLMs read the messy source document, then a deterministic engine does all the math, so every number in the final report is traceable, never hallucinated.

Ingest

Trial balance PDFs from any ERP (TeamSystem, Zucchetti, Passepartout) uploaded into the platform.

Extract

PyMuPDF reads text per page with OCR fallback; a vision LLM pulls every account code, name, and amount.

Map accounts

The LLM maps each inconsistent client line onto the standard chart of accounts.

Reconcile

Deterministic checks verify assets = liabilities + equity; suspect rows are auto-corrected until the books balance.

Deterministic calculation engine — no LLM, every number traceable
Reclassify

Income statement and balance sheet reclassified through inspectable formulas.

KPIs & benchmarks

Ratios computed and benchmarked against sector data.

Budgets & plans

Monthly budgets, industrial plans, and DSCR / crisis-alert stress tests.

Board-ready report

Every figure traceable to the balance line or assumption it came from — zero hallucinated numbers.

Outcomes

  • Trial balance PDFs from any ERP parsed automatically, eliminating hours of data entry
  • Every figure deterministic and traceable to its source, with zero hallucinated numbers
  • Automated reclassification, KPIs, and sector benchmarking
  • Monthly budgets, industrial plans, and statutory crisis-alert dashboards (art. 2086 c.c., DSCR)
  • Multi-tenant isolation with EU data residency, GDPR compliant