Artificial Intelligence and Spain’s Tax Agency: Tax Revolution or Technological Mirage?

AI and taxation in Spain: what is really changing (and what is hype)

You do not need AI just because it is fashionable; you need clean data, processes and criteria. The Spanish Tax Agency (AEAT) is already automating and using advanced analytics. This will raise expectations for data quality and traceability from businesses and the self-employed. The useful question is not “AI or no AI,” but what it is for, how it is governed and who bears the risk.

1. Where tax AI and automation stand in Spain

• The 2025 Annual Tax Control Plan (BOE) sets prevention and control priorities, emphasizing large-scale data use, new tools and better assistance, continuing the 2024–2027 Strategic Plan.
• The AEAT’s 2025 guidelines consolidate digital assistance, fraud prevention and control, and development of technology tools.
• Digital Assistance (ADI) and virtual assistants cover IRPF, censuses, VAT, SII and collection. They offer guided natural-language questionnaires, transaction locators, property classifiers and help with Forms 303 and 390. They are not magic, but they change how taxpayers interact with the AEAT.
• VERI*FACTU / SIF is a verifiable invoicing system intended to support tamper-resistant records and real-time analysis. Looking toward 2026, it is a major driver of standardized, traceable data—and therefore more effective analytics and AI.
• Professional commentary describes the use of big data, machine learning and deep learning to prioritize audits and detect discrepancies.

Interim conclusion: The AEAT is accelerating digitization and data use. Whether this is called AI or intelligent automation matters less than the effect: earlier detection of inconsistencies and greater expectations for traceability.

2. The adviser’s perspective: where AI adds value—and where it does not

It adds value when it:

1. Normalizes and validates data before filing with the AEAT (accounting, SII and VERI*FACTU invoicing): cleaning, deduplication, cross-checks of dates, VAT and counterparties, and matching against bank statements.
2. Documents criteria for intra-Community VAT, place of supply and property matters using internal assistants that replicate the firm’s decision tree and preserve reasoning logs and versions.
3. Prioritizes risk by scoring files according to common errors, materiality thresholds, related-party transactions and sensitive deductions, aligned with public warning signs in the SII, Forms 303/390 and information returns.
4. Produces evidence through pre-checks that simulate what an AEAT analyst might see, helping prepare for reviews and avoid repeated information requests.

It is not useful—or is hype—when:

• You expect a large language model (LLM) to decide tax treatment without rules or supervision.
• You use a “copilot” that is not grounded in your criteria, binding rulings and case law; it will eventually hallucinate.
• You skip data quality and traceability. Without them, no AI can rescue Forms 303, 390 or 200.

My view: An adviser’s competitive edge is not “using AI,” but turning criteria into repeatable processes and measuring whether they reduce requests, reassessments and penalties. The rest is marketing.

3. The client’s perspective: what businesses and self-employed people should ask for

Ask your adviser or ERP provider for:

• VERI*FACTU readiness and an immutable log for each invoice, including issuance, changes and submissions to the AEAT.
• Automatic checks for tax IDs/VAT numbers, transaction codes, VAT rates, the equivalence surcharge, reverse charge and pro rata rules.
• Assisted bank reconciliation combining rules with learning from your own patterns.
• Traceable decisions explaining why a transaction received a particular treatment, with the relevant law, ruling or analogous case.
• Early alerts for inconsistencies between Forms 303/390 and SII, tax losses, transactions with non-EU countries and errors in statutory ledgers.

Watch out for:

• “Copilots” that cannot justify their criteria by reference to law, rulings or case law.
• Systems that fail to retain reproducible records for responding to an information request.
• Prompt dependence without governance: staff changes can change the tax position and create legal uncertainty.

4. The AEAT’s perspective: what it is automating and where it is heading

• Assistance and control: More virtual assistants, locators and VAT/IRPF calculators, linked to ADI specialists through chat. This reduces friction and standardizes answers.
• Verifiable invoicing (VERI*FACTU): It may have an “SII 2.0” effect: richer, more uniform records, more automatic checks, less room for manipulation and real-time anomaly analysis.
• Intelligent case prioritization: Big-data and machine-learning screening to detect discrepancies, as described in professional literature and aligned with the control guidelines.
• 2024–2027 framework: The strategic roadmap is to deepen assistance, prevention and complex control with technology.

In practical terms: more standardized data can produce better risk scores and more feedback loops. If your organization does not measure its own consistency—what it files against what it records—others will do it for you.

5. You may not need AI yet: you need data governance and executable criteria

Autonomous and predictive AI is real, but there are no miracles without data governance. Before investing, measure:

1. Rules coverage: How many of your tax scenarios have codified criteria?
2. Traceability: Can you reconstruct exactly why a transaction was treated as it was?
3. Data quality: What proportion of invoices passes validation without human intervention?
4. Effectiveness: Have information requests or penalties fallen since automation began?
5. Reuse: Is a learned criterion applied automatically throughout the firm or business?

If the answer is no, you do not need AI; you need a method.

6. Practical recommendations by profile

For tax firms

• Map risks by tax and form (VAT, Corporation Tax, IRPF, IRNR, Forms 349/390, SII and ledgers), including thresholds and warning signs.
• Build a catalogue of criteria (law, rulings, Supreme Court and TEAC decisions) and a version-controlled internal rules engine.
• Use VERI*FACTU/SII pre-validators and bank reconciliation combining rules with supervised learning.
• Keep an evidence repository of contracts, invoices, logs and AEAT-assistant records, indexed by file.
• Run quarterly audits of data quality and defensive KPIs, such as avoided requests, reassessments and surcharges.

For businesses and self-employed people

• Adopt VERI*FACTU early and train key users.
• Choose an ERP with embedded tax controls and immutable logs.
• Document every exception, such as a VAT place-of-supply decision, with its justification.
• Run “as if you were the AEAT” simulations before filing.

For the administration

• Provide a minimum level of algorithmic transparency by publishing warning-sign categories and feedback on common errors to support voluntary compliance.
• Extend assistants to complex scenarios (pro-rata calculations, triangular chains and property transactions) and offer a validation API for third-party pre-checks.

7. Ethics, legality and security: three non-negotiables

1. Traceable decisions: Every AI recommendation should explain why, citing the statute, ruling or case law.
2. Data protection: Models and prompts containing personal or tax data need safeguards equivalent to those in your ERP.
3. Do not delegate judgment: AI assists; the law supports the tax position. The guidelines and Strategic Plan emphasize assistance and prevention, not arbitrariness.

8. What you need before buying AI

• Do you have a VERI*FACTU plan, pilot and KPIs?
• Are your tax rules codified and version-controlled?
• Do your data pass automatic checks for VAT, SII and Forms 303/390?
• Do you generate evidence for defending a position, such as logs, AEAT-assistant records and calculation support?
• Do you measure avoided requests and closing times?
• Have you defined roles and controls for prompts and models?

If two or more answers are no, invest first in data and processes.

9. What is useful—and what is hype—in 2025

Useful now	Hype now
Automatic tax validators before Forms 303/390 and SII	“Copilots” without rules or legal citations
VERI*FACTU readiness with immutable logs	Promises of “zero information requests” thanks to AI
Internal assistants grounded in your criteria and evidence	LLMs without safeguards or data governance
AEAT-view simulations and risk scoring	Attractive dashboards without defensive metrics

You do not need AI—you need to be auditable

The AEAT will continue moving toward verifiable data (VERI*FACTU), standardized digital assistance and intelligent screening. The defense for advisers and businesses is to be auditable by design: codified criteria, clean data and complete traceability. Truly useful AI will strengthen that triangle, not cover it with promises.

Read the original article in Spanish