The Problem
April 30 is Mexico's annual tax filing deadline. It comes every year. And every year, business owners scramble to figure out if their fiscal house is in order — or hope their accountant caught everything.
This client had a more complex situation than most. Three separate fiscal entities. Three distinct income streams. Three sets of SAT declarations, CFDIs, bank statements, and third-party payment records. All of it needed to be cross-referenced before the deadline.
The question on the table: Are we compliant? Where are the gaps? What do we ask the accountant?
With manual review, this kind of multi-entity audit takes weeks — if it gets done at all. Most owners skip it and hope for the best. The risk stays hidden until SAT surfaces it.
This client didn't want to hope. They wanted facts.
The Approach
Rather than hiring an additional accountant or spending weeks in spreadsheets, we built a custom AI audit tool from scratch. The goal was clear: ingest every available fiscal document, cross-reference it automatically, surface every mismatch, and generate a structured set of questions ready to bring to the accountant.
This is what AI agents do well — high-volume, structured data processing that would exhaust a human analyst in days.
The source data included:
- Six CFDI export files from the SAT portal
- Monthly and annual tax declarations for all three entities
- Third-party payment records (47 invoices plus transaction CSVs)
- Wire transfer history across multiple periods
- Corporate registration documents
The Technical Solution
We built a Node.js script — cfdi-audit.mjs — that handled the full ingestion and analysis pipeline in a single run.
Step 1: Parse the CFDIs. The SAT exports CFDI data in Excel format with inconsistent column structures across different entity types and regimes. The tool normalized all six files into a unified schema, handling CFDI 4.0 format requirements — UUID, fiscal period, issuer RFC, receiver RFC, concept, and amount.
Step 2: Map income to entities. Mexico's tax system treats each fiscal regime differently. A person filing under PFAE has different obligations than someone under RESICO, which differs again from a Sociedad Civil under Régimen General. The tool applied regime-specific logic to flag what income should have corresponding CFDIs issued — and what income appeared to have none.
Step 3: Cross-reference third-party records. Payment processor invoices and wire transfer records were matched against emitted CFDIs by amount and period. Exact matches were confirmed. Gaps were flagged.
Step 4: Generate the audit report. Every mismatch, unexplained period, or structural inconsistency was logged as a numbered red flag with source references. The output included a structured list of 20 specific questions organized by entity — ready to paste into an accountant meeting agenda.
This is the difference between AI-driven fiscal compliance and hoping your accountant caught it.
The Results
- 1,650 CFDIs processed across 3 fiscal entities
- 6 CFDI export files normalized into a single audit-ready dataset
- 19 red flags identified — ranging from missing invoice coverage to structural questions about entity objeto social
- 20 accountant questions generated — organized by entity, specific, and sourced
- 48 hours from zero to complete audit report
- Manual equivalent: 2–4 weeks, minimum
The client walked into their accountant meeting with a structured report instead of a pile of downloads. Their accountant had everything needed to make decisions immediately.
Key Takeaways
1. Complexity is not a reason to skip the audit — it's a reason to automate it.
2. The SAT has your data. You should have it too. Every CFDI you've ever emitted or received is on the SAT portal. A custom tool can process all of it faster than you can open the files.
3. CFDI 4.0 compliance is not optional. Retroactive audits look at whether your invoices meet current requirements.
4. The accountant's job is strategy — not data archaeology. Give them a structured report. Let them focus on remediation, not discovery.
5. This is not a one-time tool. The script runs again next year. The infrastructure cost is a one-time build. The value compounds.
That's what real business automation looks like.
If your business has multiple fiscal entities, mixed income regimes, or international payment channels — start the conversation at heynext.ai/contact.
