Velten Advisors

Transparency & Governance

AI Governance

Technology is now a permanent feature of the tax landscape: inside the tax function and outside it. Inside the tax function, AI and automation are reshaping research, drafting, modelling, review and reporting. The opportunity is significant; so is the governance challenge. Velten Advisors advises on the controls, data architecture, verification discipline and operating model that make AI-assisted tax work durable. Outside the tax function, tax authorities are increasingly using analytics-enabled risk selection, anomaly detection and cross-jurisdiction reconciliation. We apply a dual lens: governing how AI is used within the tax function while testing whether the resulting positions and files can withstand technology-enabled scrutiny from outside it.

AI Governance
Tool-agnostic. Independent. Designed for environments where individuals sign opinions or approve positions under personal accountability.
AI does not reduce professional responsibility. It concentrates it.

What We Do

We structure AI governance around four connected dimensions, bound together by the continuous Scepticism Loop:
  • 01. Discover: Where AI Touches Tax Map every place AI already touches the tax function, including shadow AI. Identify intersections with opinion- signing, filings and regulator-facing positions. Establish the jagged frontier between reliable and fragile performance.
  • 02. Develop: Verification & Documentation Discipline Build verification and documentation discipline. Check outputs against primary-source law; record material decisions with timestamps and source trails; require a qualified professional to take responsibility for conclusions rather than merely reviewing prose.
  • 03. Deploy: Embedding Governance into Workflows Embed governance into workflow through policy, RACI maps, vendor oversight, data-privacy alignment, human-in-the- loop controls and incident escalation. Governance that exists on paper but not in the workflow is not governance.
  • 04. Deepen: Preserving Judgment & Apprenticeship Protect the human capabilities AI cannot replace: judgment, mentorship and durable professional formation. Unmanaged automation erodes skill; designed co-work deepens it.
The Scepticism Loop
Continuous primary-source verification through every workflow stage. Every AI-assisted output remains provisional until verified against primary-source law, recorded in a reconstructable form, and confirmed by a qualified professional who takes personal responsibility for it.
Specialist Advisory Modules
  • Twin-Lens Controversy Preparation: Reconstructing AI workflows ahead of authority audit or enquiry.
  • Position Integrity Review: Testing historical AI-assisted filings and technical files.
  • AI & PI Exposure Review: Aligning professional indemnity coverage with AI workflows.
  • Data-Coherence Audit: Cross-border data flows and confidentiality boundary protection.
  • Fractional AI Governance Steward: Senior independent advisory for heads of tax and governance committees.

How We Work

The Trust Stack Hierarchy Governance responsibility is distributed across four distinct layers:
L1

Professional Judgment & Capability

AI fluency, scepticism, discipline, verification habit, earnings durability. Capability erosion degrades review quality, weakens control robustness and diminishes succession depth.

L2

Governed AI for Tax Functions

Governance frameworks, audit trails, human-in-the-loop workflow design.

L3

Leadership & Board Advisory

Board-level oversight, regulatory readiness, incident response.

L4

AI Assurance & Trust Validation

Independent confirmation that AI-assisted work is trustworthy.

  • Practitioners with Live AI Operation Experience Designing, governing and operating a purpose-built AI tax research platform for our own internal use surfaced the critical failure modes (hallucinated authority, silent contextual drift, unverifiable outputs) in our own environment before they reached client work. We are not advising on AI governance from the outside.
  • Structural Independence Tool-agnostic by design. No software vendor relationships. No reseller commissions. No bundled technology sales. Designed specifically for environments where individuals sign opinions under personal liability. Independence is structural, not a policy.
  • Not the Right Fit If You Need AI tool selection, software vendor procurement, technical tax advice detached from governance, or off-the-shelf template slide decks.

Who We Act For

Heads of Tax

Preparing for board or regulatory scrutiny on AI use across active advisory and filing positions.

In-House Tax Teams

Scaling AI across tax and finance teams with eight or more professionals.

MAS-Regulated Institutions

Where tax AI must sit strictly within FEAT, TRM and enterprise AI governance frameworks.

Boards & Audit Committees

Seeking independent, external assurance and verification before an incident occurs.

CFOs & Executive Committees

Quantifying AI operational risk and ensuring durable governance return on investment.

Professional Advisory Firms

Where AI outputs touch signed opinions, deliverables, and partner-level PI exposure.

Why It Matters Now

  • The Predicament Tax teams are already using AI in research, drafting, reconciliations and scenario analysis. In a profession where mostly correct is unacceptable, unstructured AI use is unmanaged risk. Most AI failures begin at the desk level and scale quietly through systems; by the time issues surface, positions are often already filed, relied upon, or disclosed.
  • It Has Already Happened (Tribunal Precedents)
  • UK First-tier Tribunal: A taxpayer lost a £13,000 claim in full when AI-drafted arguments cited non-existent authorities; the tribunal criticized reliance on AI "short-cuts."
  • HMRC AI Disclosure Order: A tribunal ordered HMRC to disclose whether generative AI was used in its R&D claim reviews. Both sides of the table now carry AI risk.
  • Dual-Lens Authority Assessment Tax authorities across Asia-Pacific are increasingly AI-enabled. The ATO runs AI-enabled risk engines. Korea's NTS holds ISO/IEC 42001 certification for its AI systems. Japan's NTA uses machine learning for automated audit target selection. IRAS has publicly indicated the use of analytics in compliance and risk selection.
"The question is not whether you will be reviewed by AI: but whether your file holds up when you are."
  • Local Standards & Local Defensibility
    • Singapore: IMDA Model AI Governance Framework, AI Verify, MAS FEAT and Veritas Toolkit, MAS TRM Guidelines, PDPA, IRAS Tax Governance Framework.
    • Hong Kong: PDPO, PCPD Model Personal Data Protection Framework for AI, HKMA Principles on AI, SFC guidance.
    • Malaysia: PDPA 2010, National Guidelines on AI Governance and Ethics, Bank Negara Malaysia RMiT.
    • Australia: AI Ethics Principles, National Framework for Assurance of AI in Government, APRA CPS 230, ASIC expectations.
  • Governance as a Financial Decision Governance protects against rework, QA failures, and late penalties, delivering durability across 5 core pillars: Productivity, Capacity, Readiness, Quality, and Risk Reduction.

Start with an AI Readiness & Risk Assessment.

A structured diagnostic conducted over 3 to 4 weeks with minimal client-team time commitment (6 to 10 hours). Deliverables include a complete AI use inventory, risk heatmap, dual-lens authority assessment, and a concise, board-ready summary deck outlining a prioritized governance roadmap.

Arrange a Confidential Discussion