Case study · State revenue

Data-driven tax revenue. Nine workstreams in production

A state revenue service's data program turned auditing into continuous supervision, with relationship analysis at billion-record scale, continuous near-real-time transaction monitoring and machine learning applied to inheritance and gift tax filings.

Client: State Finance Department Solution: Analytics

Billion-record scale, decisions in seconds

30,4M People and companies mapped (tax IDs)
215M Relationships between entities
R$ 5B Tax risk identified
5min Alerts after the taxable event
Workstreams

Nine workstreams in production

From the integrated dossier to machine learning for inheritance and gift tax, by way of continuous transaction monitoring and relationship analysis at state scale.

Relationship analysis

30.4 million people, 215 million relationships and 264 million addresses processed (−37.66% → 164.5 million unique). Sources: motor vehicle department, electronic tax documents (NF-e, CT-e, MDF-e) and Federal Revenue Service (RFB) records.

Integrated dossier

16 sources integrated, 30 dashboards, 36 views per dossier, 11.4 million company and 17.3 million individual tax IDs consolidated — replacing manual collection.

Continuous monitoring · Rice sector

173 taxpayers, 13,000 operations and R$ 465 million monitored; R$ 110,000 in state VAT assessed in the first month, on a 150-second near-real-time cycle.

Continuous monitoring · Import invoices

9,361 import declarations monitored, 3,351 flagged (36%) and R$ 82 million in discrepancies identified, out of ~R$ 80 billion a year in imports.

Tax filing risk analyzer (ML)

~90,000 filings a year; model trained on 320,000 declarations — 97% true negatives and 85% true positives.

Continuous monitoring · Micro-entrepreneurs

800,000+ micro-entrepreneurs monitored and 1,300+ suspensions — oversight impossible to do by hand, with alerts in minutes.

Electronic tax documents (DF-e) — fraud alerts

An average of 1.75 million electronic invoices processed per day; 497,000 involved in alert scenarios and 5,341 alert events.

Data governance

R$ 5 billion in tax risk identified (R$ 2.5 billion from unreported outbound sales) and ~12,700 data assets cataloged.

Near-real-time architecture

60,000 electronic invoices per hour and 270,000 consumer invoices per hour, with alerts in under 5 minutes via an API integrated with the audit team's channels.

Data Quality

Data quality, end to end

  1. Standardization and validation of taxpayer addresses
  2. Registry deduplication into a single record (golden record)
  3. Data enrichment with external data and public sources
  4. Address geocoding for territorial intelligence
  5. Data governance with a metadata catalog and business glossary
  6. End-to-end data lineage, from source to dashboards
  7. Continuous monitoring with automated quality indicators
In practice

A single query

Information that used to take several queries across several systems now comes from a single query — an auditor's account of the Integrated Dossier.

Does your revenue service want to move beyond sampling?

See the path a state revenue service took — relationship analysis, integrated dossier, continuous monitoring and governance.

Talk to Vert View all case studies