AI is accelerating decisions. Finance must preserve accountability.

Artificial Intelligence is transforming financial decision-making at a speed few organisations have experienced before. For FP&A leaders, AI expands analytical capacity: faster forecasts, broader scenarios, new insights, and a strengthened strategic role for Finance.

But a question is becoming unavoidable: when a financial decision is influenced by Artificial Intelligence, can the organisation still explain, reconstruct, and defend how that decision was made?

The challenge is no longer only analytical capability. It is the integrity of the decision.

Data → Analysis → Decision

Data → Analysis → Professional Judgement → Accountability → Defensibility

A financial decision can no longer be assessed solely by its outcome. It must also be assessed by the organisation's ability to reconstruct what information was available, what assumptions shaped the analysis, where professional judgement intervened, who approved the decision, and under what governance conditions it was taken.

This is the problem I have been exploring through my work in Decision Architecture.

1. Finance Enters a New Era of Accountability

Historically, financial governance focused mainly on controlling outcomes: Was the forecast correct? Was the investment profitable? Was the budget met?

The era of Artificial Intelligence introduces a different question: How was the decision produced?

A model can recommend a scenario, identify a pattern, or generate a projection. But technology does not assume institutional accountability. The organisation remains responsible for the decision.

This creates a new requirement for Finance: not only better analysis, but better decision processes.

2. A Pattern Emerging Across Different Fields

What has caught my attention is that similar concerns are now emerging across very different fields.

The OECD, the Financial Stability Board (FSB), the EU AI Act, and the German judiciary approach these issues from different perspectives, but converge around questions of governance, explainability, human oversight, data governance, third-party dependency and attribution of responsibility.

Rather than a list of independent risks, these documents reflect a transformation in the nature of governance. The challenge is no longer solely about developing more accurate or efficient Artificial Intelligence systems, but about preserving the conditions that allow a decision to be understood, reconstructed, and defended when subjected to scrutiny.

The same pattern is also visible in the evolving European approach to anti-corruption and financial integrity. Directive (EU) 2026/1021 on combatting corruption reinforces the importance of effective prevention, detection, investigation and accountability mechanisms. For financial decision-makers, this adds another dimension to traceability: it is not enough to reconstruct how a decision was produced; institutions must also be able to demonstrate that the conditions surrounding the decision supported integrity and accountability.

Table 1. OECD-identified risks and their institutional anchor

These risks are not new. In 2021, together with Professor Yaohuai Lu, I examined how privacy, ethics, and financial compliance under Big Data required governance structures capable of preserving institutional accountability (Lu & Aravena, 2021).

Five years later, that same principle re-emerged independently. While reviewing a presentation for Frankfurt in April 2026 together with Charles Levine — a US engineer specialising in critical aerospace systems (AS9100) — we arrived at the same conclusion:

Traceability and structural control must be designed into the decision process from the outset, not added after a failure occurs.

That same month, in Germany, judicial authorities published the Grundlagenpapier on the use of Artificial Intelligence in the courts, formally establishing an equivalent principle: AI-assisted decisions must remain explainable, auditable, and subject to institutional accountability.

What I find significant is the convergence. Research on financial compliance, critical-systems engineering, the judiciary and emerging regulatory frameworks are approaching the problem from very different directions, yet they point towards the same underlying requirement: accountability cannot depend solely on the outcome. The conditions that produced the decision must also remain visible and reconstructable.

This is the structural gap I have been exploring through Decision Architecture: how to design decision conditions so that responsibility remains traceable and a critical decision remains defensible as systems, technologies and circumstances change.

3. Why FP&A Needs Decision Architecture

FP&A increasingly occupies a central place in strategic decisions. Finance teams are involved in capital allocation, investment decisions, scenario planning, risk assessment, and strategic projections.

With AI, the speed of analysis increases. But so does the need for governance.

A decision that cannot be reconstructed becomes vulnerable — not necessarily because it is wrong, but because the organisation cannot demonstrate how it got there.

In an environment of growing regulatory scrutiny, board oversight, and stakeholder expectations, financial leaders must be able to demonstrate not only what decision was made, but how that decision was reached. This is where Decision Architecture becomes essential.

4. From AI Explainability to Organisational Defensibility

AI systems are becoming increasingly capable of explaining their outputs. But organisations need something additional: understanding how an output became an institutional decision.

In my work, I distinguish three different layers:

• Explainability: the technical logic behind the recommendation.

• Traceability: the connection between data, analysis, professional judgement, approval, and execution.

• Defensibility: the ability to demonstrate that the decision was accountable, governed, and properly authorised.

When organisations rush from data to execution without governance, they fall into a passive shortcut:

Data → Analysis → Error Inheritance → Inherited Approval → Vulnerable Decision

To prevent this systemic breakdown, Decision Architecture restores the missing human and governance conditions, turning a linear shortcut into an Architecture of Responsibility:

Data → Analysis → Professional Judgement → Accountability → Institutional Accountability → Private Legacy

The final dimension is often overlooked: the personal and institutional legacy of those who carry responsibility for the decision. Decisions may need to be reconstructed years after the original people, assumptions or circumstances have changed.

The Jurisprudential Standard: Support, Not Substitute

To understand how traceability operates in practice, finance leaders can look to high-stakes decision-making frameworks. In May 2026, the Presidents of Germany's Higher Regional Courts, the Kammergericht, the Bayerisches Oberstes Landesgericht and the Federal Court of Justice (Bundesgerichtshof) adopted an updated Grundlagenpapier on the use of AI and algorithmic systems in the judiciary. One of its core principles is that AI and algorithmic systems may support, structure and standardise preparatory work, but do not replace judicial decisions. The principle can be summarised simply: support, not substitute.

In FP&A, this distinction is critical. AI can process vast volumes of data, model scenarios, and generate recommendations at unprecedented speeds. However, the model does not carry fiduciary duty.

The Silent Systemic Risks: The Double-Risk Dynamic

In my work, I see two risks emerging together when AI is integrated into decision workflows:

1. Error Inheritance. A minor hallucination, data drift, or misinterpretation at Stage A is silently passed as “factual input” to downstream models — the error amplifies at every step.

2. Inherited Approval. Because the AI tool was authorised for initial reporting, the organisation assumes downstream outputs are equally authorised — the amplified error is blindly approved.

By the time an executive reviews the final output, the organisation has effectively authorised an amplified error without explicit human review.

This is why traceability must evolve from a passive logging exercise into an active Architecture of Responsibility. It requires establishing three non-negotiable operational conditions:

• Bounded: system authority is strictly defined within pre-established limits.

• Interruptible: a human with real authority can stop the automated execution in real time.

• Reconstructible: the organisation can audit.

  • What data was used?

  • What assumptions intervened?

  • What actions were triggered?

  • Where did professional human judgement resume?

Traceability becomes the bridge between AI's analytical capability and institutional financial accountability.

5. Three Capabilities FP&A Leaders Will Need

The future of FP&A will not depend solely on technical knowledge of AI. It will depend on three governance capabilities:

• Defining decision conditions before analysis: establishing boundaries before execution.

• Preserving traceability over time: protecting against Inherited Approval and Error Inheritance.

• Balancing analytical speed with professional judgement: ensuring AI remains a support, never an unexamined substitute.

None of the three is optional: together, they are what distinguishes a financial leader capable of sustaining decisions under scrutiny.

Conclusion: Finance's Next Competitive Advantage

Artificial Intelligence is not only changing how Finance analyses information. It is changing the very meaning of executive accountability.

The strongest organisations will not be those that eliminate uncertainty. They will be those capable of making decisions that remain understandable and defensible even under institutional scrutiny.

A mature financial decision is not one that will never be questioned. It is one that can withstand being questioned — protecting both the institution's integrity and the private legacy of the leaders who bear fiduciary responsibility.

Because resilience under pressure is not negotiated — it is designed in advance.

References

1. OECD. (2026). Artificial intelligence and open finance: Synergies, trade-offs and policy implications. OECD Artificial Intelligence Papers, No. 61. OECD Publishing, Paris. DOI: 10.1787/eb511c3c-en.

2. Financial Stability Board. (2026). Sound Practices for Responsible Adoption of Artificial Intelligence (AI): Consultation report. 10 June 2026.

3. European Parliament and Council of the European Union. (2024). Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L 2024/1689.

4. European Parliament and Council of the European Union. (2026). Directive (EU) 2026/1021 of 29 April 2026 on combatting corruption, replacing Council Framework Decision 2003/568/JHA and the Convention on the fight against corruption involving officials of the European Communities or officials of Member States of the European Union and amending Directive (EU) 2017/1371. Official Journal of the European Union, L 2026/1021.

5. Arbeitsgruppe „Einsatz von KI und algorithmischen Systemen in der Justiz“. (2026). Grundlagenpapier 2026 zum Einsatz von KI und algorithmischen Systemen in der Justiz. Beschluss der Präsidentinnen und Präsidenten der Oberlandesgerichte, des Kammergerichts, des Bayerischen Obersten Landesgerichts und des Bundesgerichtshofs, May 2026.

6. Lu, Y., & Aravena, P. A. (2021). On Privacy Issues and Ethical Countermeasures in Financial Compliance Management Under the Background of Big Data. Financial Theory and Practice (财经理论与实践).