Accounting / Portugal
AI workflow design for e-Fatura reconciliation in Portuguese accounting
Use deterministic matching for amounts, dates, identifiers, and record state. Use AI for messy exception explanation, document classification, and reviewer notes, with accountant review before any compliance-sensitive action.
Why this workflow matters
The useful design starts with the local records and review steps: e-Fatura data, ERP rows, supplier identifiers, exception types, and accountant signoff boundaries.
Inputs and outputs
Inputs
- e-Fatura records
- ERP purchase ledger exports
- Supplier identifiers
- Invoice images or PDFs
- Prior accountant decisions
Outputs
- Matched invoices
- Unmatched or duplicate exceptions
- Suggested supplier corrections
- Reviewer notes
- Audit trail for accountant signoff
Current manual workflow
Start by modeling the work as it happens now.
- Export or collect the relevant e-Fatura records and internal ledger rows for the review period.
- Normalize supplier names, tax identifiers, dates, amounts, currency, and document references.
- Run deterministic matching first for exact and high-confidence matches.
- Route unmatched, duplicate, conflicting, or ambiguous records to an accountant review queue.
- Record the reviewer decision and link it back to the input records for auditability.
Where AI helps
Use models around the exception work.
- Classify invoice documents and supplier names when source data is messy.
- Cluster recurring exception patterns by supplier, ledger account, or missing field.
- Draft reviewer notes explaining why a record did not match.
- Suggest likely duplicate or related records for review.
- Prepare a clean exception summary for the accountant or controller.
System pattern
Keep deterministic checks in charge of the hard boundaries.
Architecture
- Ingest e-Fatura and ERP records into a normalized reconciliation table.
- Apply deterministic matching rules before asking a model to explain exceptions.
- Use AI only after the system has structured the mismatch type and candidate records.
- Present exception clusters, source records, and draft notes in a reviewer queue.
- Store every reviewer decision, reason, and linked source record for later audit review.
Keep deterministic
- Tax identifier comparison.
- Amount, VAT, date, and currency normalization.
- Exact-match and tolerance rules.
- Duplicate detection thresholds.
- Final status update and audit log creation.
Do not fully automate
- Final tax treatment decisions.
- Filing or portal actions without accountant signoff.
- Changing supplier master data without review.
- Handling ambiguous cases that require local professional judgment.
Evaluation and controls
A useful workflow design explains how to check the work.
Match precision
High-confidence matches should rarely be overturned by accountants.
Exception recall
Known duplicate, unmatched, and conflicting records appear in the queue.
Reviewer time per exception
Accountants spend less time reconstructing why a record was flagged.
Audit completeness
Each final decision links to source records and reviewer notes.
Accountant or controller
Professional review
Compliance-sensitive outcomes require a named reviewer decision.
Finance systems
Rule-first matching
Structured comparisons run before AI explanation or note drafting.
Accounting operations
Source traceability
Every exception links back to e-Fatura and ERP source records.
Controller
Change audit
Supplier or ledger updates are logged with reason and reviewer.
Pilot checklist
Test the workflow before widening automation.
- Select one review period and one company entity.
- Collect a sample of e-Fatura records, ERP ledger rows, and invoice documents.
- Label 100-300 historical matches and exceptions if available.
- Define exact-match, tolerance, duplicate, and human-review rules.
- Run the queue in shadow mode before using reviewer decisions operationally.
Synthetic example
A supplier name differs between the ledger and e-Fatura records while the tax identifier and amount match. The system marks it as a likely supplier-name mismatch, drafts a reviewer note, and asks the accountant to confirm before any supplier-master correction.
Sources and review notes
Source context matters when the workflow touches risk.
This is not Portuguese tax, accounting, or legal advice. Any workflow touching e-Fatura records, VAT treatment, filing, or compliance-sensitive decisions should be reviewed by qualified local professionals.
Portal das Financas
Official e-Fatura information from the Portuguese tax authority portal.
Related playbooks
Adjacent workflows to compare.
Model evaluation change control
A practical AI workflow for reviewing model, prompt, retrieval, and tool changes before they reach production.
Human review required
HR policy RAG evaluation
A source-grounded evaluation workflow for internal HR policy assistants where freshness, permissions, and escalation matter.
Human review required
Workflow review
Have a similar workflow that needs controls and evals?
Share the role, market, source systems, work item, and current failure modes. The useful first step is usually a small eval or shadow review before any automation is trusted.