Tax management hubTax management hub

Problem

Tax credit reviews usually live in hand-built spreadsheets, with rules that change by regime, period and court ruling. Checking who calculated what, on which base, takes as long as the calculation itself.

What we built

Each project follows four steps (configure, import, validate, compute), guided by five gated stages: parameters, import, payroll item classification, integrity check, calculation.

• A private flow (companies) and a public flow (municipalities, with the gradual payroll tax relief of Law 14.973/2024, mandatory approval, and a block when data coverage is low or sources disagree).
• It imports SPED EFD-Contribuições and SPED Fiscal.
• Engines cover social security on indemnity payments (employer, RAT×FAP, third parties, capped at what was actually paid), PIS/COFINS excluding ICMS from the base (RE 574.706: regime, rates, statute of limitations and modulation, two calculation routes that must converge, critical validations that block the run) and ICMS-ST in the PIS/COFINS base.
• Amounts are updated by the SELIC rate pulled from the Central Bank; reference tables include TIPI, CEST, SAT/CNAE and DARF.
• The engine is checked automatically against a test suite before any calculation is released.
• The system also has an internal library of legal theses by segment, NCM analysis, rule-based audit, and AI that cross-checks imported data against the rules to point to possible credits.

How it works in practice

In day-to-day reviews, calculation stays deterministic and auditable, every number tied to the rule and table that produced it. AI helps decide where to look but never writes the number. Coverage and divergence blocks surface problems before the calculation, not after.