Central banks are adopting AI cautiously but keenly, and enthusiasm for AI-driven supervision is growing. The Irving Fisher Committee’s October 2026 survey finds central banks moving beyond experimentation towards AI readiness, preparing both their organisations and their data to use the technology (Schmidt et al., 2026).One question deserves closer scrutiny: how much of this value could be delivered more efficiently, transparently and reliably through conventional automation? The question arises  as   much of banking supervision (rule-based calculations, regulatory reporting, compliance checks, risk scoring and administrative workflows) is deterministic: identical inputs, processed under identical rules, should produce identical outputs. Here, conventional automation is often not merely sufficient but preferable.

1. Regulatory Compliance: Deterministic Rules, Deterministic Solutions

Capital adequacy, liquidity coverage, large exposure and concentration limits involve prescribed calculations against defined thresholds. Conventional software can:

  • calculate ratios and compare them with regulatory minima;
  • identify breaches and generate alerts;
  • preserve a complete audit trail: source data, applicable provision, methodology and resulting flag.

Generative AI adds uncertainty here without improving accuracy. The Basel Core Principles call for effective supervisory techniques, reliable reporting and timely intervention, not any particular technology (BCBS, 2024). Where requirements are explicit and machine-executable, conventional automation should be the first choice.

2. Risk-Based Supervision: Transparent Scoring Before Algorithmic Complexity

Structured assessments of capital, asset quality, earnings, liquidity, governance and operational risk can be fully automated: indicators calculated, predetermined weights applied, scores aggregated and ratings generated. The advantages are:

  • consistency and reproducibility;
  • transparent scoring that can be explained and challenged;
  • straightforward recalibration.

AI may assist with qualitative judgements or complex relationships, but its added value should be demonstrated against a conventional benchmark. A rating that cannot be explained or challenged may be less useful than a simpler, transparent one. BIS research likewise stresses moving suptech from experimentation to tools that demonstrably improve supervision (Prenio, 2024).

3. Early Warning Systems: Thresholds Before Predictions

Statistical rules, tolerances and trend analysis can automatically flag:

  • sustained deterioration in capital or liquidity ratios;
  • rising non-performing loan concentrations;
  • repeated regulatory reporting deficiencies;
  • excessive single-counterparty exposures;
  • delays in implementing supervisory corrective actions.

AI may improve prediction where relationships are non-linear or involve many interacting variables, but any gain must be weighed against false positives, missed warnings, model instability and explainability. A complex algorithm is not inherently superior to a well-calibrated threshold system.

4. Inspection Management: Workflow Automation Before Intelligent Agents

Scheduling, document requisitions, checklists, workpapers, observation classification, approvals and corrective-action tracking all run well on conventional database and workflow technology. Historical observations can be retrieved through structured taxonomies, regulatory references, risk categories and keyword search.

AI semantic search helps where observations use different wording, but good indexing and classification should be exploited first, because inspection management needs traceability, version control and accountability, not just retrieval. The suptech opportunities in data collection and analysis identified by Broeders and Prenio (2018) do not necessarily depend on AI.

5. Supervisory Dashboards: Data Engineering Before AI

A dashboard that consolidates returns, computes indicators, tracks trends and produces exception reports needs:

  • accurate source data and consistent definitions;
  • reliable interfaces and validation;
  • sound data governance.

Conventional pipelines and business-intelligence tools deliver these. A poor data architecture cannot be rescued by an AI interface. BIS work on the suptech generations shows that data collection, integration and reporting are foundational capabilities (di Castri et al., 2019).

6. Why Conventional Automation May Be Superior

The preference reflects the nature of supervisory decision-making, not technological conservatism:

  • Predictability — the same inputs and rules always give the same outputs.
  • Explainability — every alert traces to a precise calculation, threshold or rule.
  • Auditability — each step can be documented, independently tested and reproduced.
  • Regulatory defensibility — conclusions rest on explicit provisions and verified calculations.
  • Cost efficiency — stable, narrowly defined tasks need fewer specialist resources.
  • Operational control — changes to rules, thresholds and permissions are explicitly authorised and tested.

These are potential advantages, not guarantees: poorly designed automation can also err, apply outdated rules or mask data weaknesses. AI adds further concerns around model behaviour, explainability, third-party dependence and cybersecurity (FSB, 2024, 2026). The EBA (2020) similarly treats explainability, traceability and auditability as core elements of trust in advanced analytics.

7. Where AI Is Genuinely Useful

AI earns its place where tasks demand contextual interpretation or unstructured information, for example:

  • reviewing complex governance reports;
  • finding recurring weaknesses across differently worded inspection observations;
  • interpreting supervisory correspondence;
  • spotting emerging risks that fall outside established indicators.

The FSI’s stocktake groups gen AI applications in supervision into document processing, knowledge management and document review, and identifies inaccurate outputs and user acceptance as the main challenges (Prenio, 2025).

The right question is not whether AI can perform a task, but whether it performs it materially better than a simpler alternative. Even where AI does the interpreting, deterministic tools should still perform the regulatory calculations and verify numerical conclusions.

8. An Automation-First SupTech Framework

  • Standardise: clear supervisory processes, taxonomies, data definitions and decision rules.
  • Automate: conventional software for calculations, compliance monitoring, reporting, workflow and structured retrieval.
  • Augment: AI only where complexity, volume or unstructured information defeat conventional methods.
  • Validate: test AI against conventional alternatives on accuracy, processing time, false-positive rates, explainability, cost and supervisory usefulness.

Industry evidence supports this sequencing: UK financial firms cite integrating AI into existing processes and legacy infrastructure as their leading non-regulatory constraints (Bank of England & FCA, 2026). The approach matters most for smaller supervisory authorities, where investment must be proportionate to resources.

Conclusion: Automation Before Artificial Intelligence

Many supervisory functions are rules-based, repetitive and data-intensive; their effectiveness depends more on accurate data, clear methodologies and disciplined implementation than on sophisticated algorithms. AI broadens what supervisors can analyse, but its adoption should rest on demonstrable incremental value, not enthusiasm. The goal is better supervision, not greater complexity. An authority that automates conventional processes well today will be far better prepared for AI than one deploying advanced models on fragmented data and ill-defined workflows.

 

References

Bank of England & Financial Conduct Authority. (2026, October 26). Artificial intelligence in UK financial services – 2026. https://www.bankofengland.co.uk/report/2026/artificial-intelligence-in-uk-financial-services-2026

Basel Committee on Banking Supervision. (2024, April 25). Core principles for effective banking supervision. Bank for International Settlements. https://www.bis.org/publications/202404-standards-core-principles-effective-banking-supervision

Broeders, D., & Prenio, J. (2018, July 16). Innovative technology in financial supervision (suptech) – The experience of early users (FSI Insights No. 9). Bank for International Settlements. https://www.bis.org/publications/fsi-insight-9-innovative-technology-financial-supervision-suptech-experience-early-users

di Castri, S., Hohl, S., Kulenkampff, A., & Prenio, J. (2019, October 17). The suptech generations (FSI Insights No. 19). Bank for International Settlements. https://www.bis.org/publications/fsi-insight-19-suptech-generations

European Banking Authority. (2020, January). Report on big data and advanced analytics (EBA/REP/2020/01). https://www.eba.europa.eu/sites/default/files/document_library/Final%20Report%20on%20Big%20Data%20and%20Advanced%20Analytics.pdf

Financial Stability Board. (2024, November 14). The financial stability implications of artificial intelligence. https://www.fsb.org/2024/11/the-financial-stability-implications-of-artificial-intelligence/

Financial Stability Board. (2026, October 15). Sound practices for responsible adoption of artificial intelligence (AI): Final report. https://www.fsb.org/2026/10/sound-practices-for-responsible-adoption-of-artificial-intelligence-ai-final-report/

Prenio, J. (2024, June 14). Peering through the hype – Assessing suptech tools’ transition from experimentation to supervision (FSI Insights No. 58). Bank for International Settlements. https://www.bis.org/publications/fsi-insight-58-peering-through-hype-assessing-suptech-tools-transition-experimentation-supervision

Prenio, J. (2025, June 12). Starting with the basics: A stocktake of gen AI applications in supervision (FSI Briefs No. 26). Bank for International Settlements. https://www.bis.org/publications/fsi-brief-26-starting-basics-stocktake-gen-ai-applications-supervision

Schmidt, R., Sirello, O., Tissot, B., & Cova, J. P. (2026, October 28). AI readiness in central banks (IFC Report No. 19). Irving Fisher Committee on Central Bank Statistics, Bank for International Settlements. https://www.bis.org/publications/ifc-report-19-ai-readiness-central-banks


Discover more from SUNANDO ROY – On Banking, Finance and Society

Subscribe to get the latest posts sent to your email.

Leave a Reply