---
title: "AI Data Governance: A Framework for UK and EU Organisations"
description: "An AI data governance framework for UK and EU mid-market organisations: GDPR, data quality, bias detection, synthetic data and cross-border transfers."
image: https://helium42.com/hubfs/blog/images/v3-g-07-ai-data-governance-hero-hero-1.webp
---

[![Helium42](https://helium42.com/hubfs/raw_assets/public/Helium42%20Clean%20Pro%20Child%20Theme/images/logo.svg)](https://helium42.com/)

[Home](https://helium42.com)

AI Education ▾

- [Intro To AI Webinar](https://helium42.com/ai-education/intro-to-ai-webinar)
- [AI Marketing Workshop](https://helium42.com/ai-education/ai-marketing-workshop)
- [AI Sales Workshop](https://helium42.com/ai-education/ai-sales-workshop)

AI Implementation ▾

- [Salesman Sam](https://helium42.com/ai-implementation/pre-build-ai-solutions/salesman-sam)
- [Content System](https://helium42.com/ai-implementation/content-system)

AI Resources ▾

- [Blog](https://helium42.com/blog)
- [AI Policy Template](https://helium42.com/ai-usage-policy-guidelines-template)
- [ChatGPT For Business](https://helium42.com/gpt-business-bundle)
- [Complete Guide To Claude AI](https://helium42.com/complete-guide-claude-ai)
- [Free AI Marketing Strategy](https://helium42.com/ai-marketing-strategy)

[About Us](https://helium42.com/about-us) 

[Contact Us →](https://helium42.com/contact-us)

![AI data governance framework showing connections to GDPR, EU AI Act, and data quality metrics](https://helium42.com/hs-fs/hubfs/blog/images/v3-g-07-ai-data-governance-hero-hero-1.webp?width=1600&height=900&name=v3-g-07-ai-data-governance-hero-hero-1.webp)

Table of contents

Published by

![Peter Vogel](https://helium42.com/hs-fs/hubfs/Team/peter%20vogel%20round.png?width=54&height=54&name=peter%20vogel%20round.png)

Peter Vogel

Peter has led digital marketing transformation programmes for over 500 organisations across the UK and Europe, with particular expertise in marketing and sales team enablement. Together with Helium42 co-founder Clwyd ...

[AI Governance](https://helium42.com/blog/tag/ai-governance) 

Mar 24, 2026, 10:00:00 AM

# AI Data Governance: A Practical Framework for UK and EU Organisations

Organisations across the UK and EU are rapidly deploying artificial intelligence systems to improve efficiency, reduce costs, and enhance decision-making. Yet the very data these systems depend on remains largely unmanaged. In many organisations, documented AI data governance policies are still missing, despite regulatory scrutiny intensifying across GDPR compliance, the EU AI Act, and evolving UK data protection guidance.

The consequences are material. Poor data quality leads to AI model failures, retraining cycles, and productivity losses. More troubling, **many data governance failures in AI systems originate during the data preparation phase**—not in model architecture—suggesting that technical teams alone cannot solve this problem. It requires cross-functional governance: business leaders, data scientists, legal, and compliance working in concert.

This guide provides a practical framework for establishing AI data governance in organisations of 150–1,500 employees operating in regulated sectors. You will learn how to structure governance for minimal overhead, implement quality controls that prevent costly model failures, manage regulatory risk in a fractured adequacy landscape, and measure the maturity progression that delivers faster time-to-value and fewer retraining cycles.

![AI data governance quality framework showing interconnected data pipeline elements](https://helium42.com/hs-fs/hubfs/blog/images/v3-g-07-data-quality-framework-body-body-1.webp?width=1200&height=675&name=v3-g-07-data-quality-framework-body-body-1.webp)

## Why AI Data Governance Has Become a Board-Level Risk

Traditional data governance—cataloguing assets, enforcing quality standards, controlling access—emerged from IT risk management. AI governance extends this significantly. Where traditional governance asks "Is our database clean and secure?", AI governance must answer "Will our AI system produce fair, accurate, and legally defensible outcomes across diverse populations and changing market conditions?"

The amplification problem is unique to AI. A single data quality defect affecting 0.5 per cent of rows in a traditional data warehouse impacts 0.5 per cent of reports. That same defect in an AI training dataset becomes embedded in model weights and affects prediction quality across **thousands of transactions**. When a loan approval model trained on biased historical data systematically denies credit to applicants from particular postcodes, the organisation faces regulatory action, reputational damage, and legal liability.

**Regulatory momentum is accelerating.** The EU AI Act (Article 10) makes data governance and management practices a legal requirement for the training, validation and testing data of high-risk AI systems; for the high-risk uses listed in Annex III, these rules apply from 2 December 2027, according to the [European Commission](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai). In the UK, the Information Commissioner's Office states that you must determine and document your lawful basis before you start using personal information, and that retrospectively switching lawful basis is likely to be unfair ([ICO lawful basis guidance](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/lawful-basis/a-guide-to-lawful-basis/)). Its [guidance on AI and data protection](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/), last updated in March 2023, is under review following the Data (Use and Access) Act 2025. If you trained a system on personal data without a documented lawful basis, you face compliance exposure. For mid-market firms that have trained models on historical operational data, **this creates immediate risk.**

Beyond compliance, governance delivers business value: mature data governance can mean **fewer unplanned model retrainings and a faster path from pilot to production**—translating to reduced technical debt and faster payback on AI projects.

## The Three-Layer Data Quality Framework for AI

AI systems depend on data quality at three distinct points in the lifecycle. Each requires different governance mechanisms and oversight structures.

**Layer 1: Training Data Quality.** Before any model is trained, organisations must establish data quality baselines. Illustrative starting thresholds, to be calibrated to each use case, cover completeness (\>98 per cent non-null values for classification tasks, \>99 per cent for time-series), accuracy (\>95 per cent matching source truth), consistency (100 per cent standardised formats for categorical fields), representativeness (distribution matching production population, validated via chi-square testing), timeliness (\<90 days old for economic or behavioural data), and uniqueness (\<2 per cent duplicate records). A single data quality failure—such as 5 per cent of training records missing critical features—propagates through model development and creates technical debt. Models trained on static historical data drift as conditions change, creating a hidden retraining cost that often emerges months after deployment.

**Layer 2: Input Data Quality (Production).** Once a model is live, input data quality directly determines prediction reliability. Schema validation ensures incoming data matches training data structure. Out-of-range detection flags values outside the training distribution (e.g., age \<18 or \>120). Temporal anomaly detection surfaces gaps or sudden distribution shifts indicating upstream pipeline failures. These checks must be automated, so that quality failures are caught before they reach production predictions.

**Layer 3: Output Validation and Bias Detection.** Model predictions themselves must be continuously validated. Accuracy decay—the rate at which prediction quality degrades over time—should remain below 2 per cent monthly decline. Bias emergence must be monitored: outcomes for protected classes that vary more than 15 per cent from baseline trigger automated alerts. Organisations must generate human-readable explanations for high-stakes decisions (loan approvals, recruitment scoring, insurance pricing) to satisfy regulatory requirements and support internal audit. In the UK, significant decisions based solely on automated processing are subject to the safeguards in [Articles 22A–22D of the UK GDPR](https://www.legislation.gov.uk/eur/2016/679/article/22A), in force since 5 February 2026 ([SI 2026/82](https://www.legislation.gov.uk/uksi/2026/82/made)).

Many mid-market organisations operate at "Level 2" maturity today: quality standards are documented but not enforced, input data is validated manually, and output audits occur monthly rather than continuously. The progression to "Level 3" (standards enforced via MLOps pipelines, automated input validation, continuous output monitoring) requires technical tooling and operational discipline but catches more quality failures before they affect models.

## GDPR Compliance and AI Training Data: The Retroactive Risk

![Synthetic data governance transformation process with privacy controls](https://helium42.com/hs-fs/hubfs/blog/images/v3-g-07-synthetic-data-governance-img-1.webp?width=600&height=600&name=v3-g-07-synthetic-data-governance-img-1.webp)

The intersection of GDPR and AI training data creates unique compliance exposure for mid-market organisations. GDPR requires documented lawful basis for processing personal data. For AI training data, this means: you must show that individuals consented to their data being used for model training, or that another legal basis (contract, legitimate interest, legal obligation) applies.

Many organisations have trained AI systems on personal data collected for operational purposes (CRM records, transaction histories, customer service logs) without documenting a lawful basis for AI use. The UK Information Commissioner's Office states that you must determine and document your lawful basis before processing begins, and that **retrospectively switching lawful basis is likely to be unfair** ([ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/lawful-basis/a-guide-to-lawful-basis/)). Organisations face potential enforcement action if they cannot document a lawful basis for the personal data used to train their AI systems.

A governance response: conduct a "Data Provenance Audit" for every AI system currently in production. Document the source of training data (which operational systems, which date range, which consent/legal basis). For data lacking documented consent, you have three options: (1) seek retroactive consent from individuals (expensive and often impractical), (2) assess whether another lawful basis genuinely applied (legitimate interests, for example, but only if you have conducted a documented balancing test, and bearing in mind the ICO's warning against switching lawful basis), or (3) exclude that data from future model retraining cycles.

The third option has created unintended consequences: organisations are systematically excluding years of historical data from model training, which reduces model representativeness and triggers more frequent retraining. This is where synthetic data governance emerges as a mitigation strategy. **Organisations increasingly use synthetic data to reduce GDPR risk**, but the governance frameworks are often immature, and few can demonstrate structured validation and lineage tracking for synthetic datasets. A governance gap is opening: synthetic data is proliferating without adequate oversight.

[Read the ICO's full guidance on AI and data protection](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/) to understand the specific compliance obligations in your jurisdiction. For a comprehensive overview of GDPR principles and personal data processing, consult [GDPR-Info.eu](https://gdpr-info.eu/).

## Bias Detection and Mitigation at the Data Layer

Model bias is a data governance problem before it is an algorithmic problem. **Much detectable bias originates during data preparation**, not in model architecture. This is a critical insight because it means that technical solutions alone—fairness constraints, algorithmic re-weighting, explainability layers—cannot fix bias that is baked into training data.

Six bias types require specific data governance interventions: **selection bias** (training data not representative of production population, mitigated through stratified sampling and continuous monitoring of input data demographics); **measurement bias** (data collection methods biasing outcomes by group, mitigated through audit of collection protocols and testing for differential accuracy across groups); **aggregation bias** (pooling diverse groups obscuring group-level patterns, mitigated through disaggregated analysis and separate models if justified); **proxy variable bias** (features correlated with protected class, such as postcode correlating with ethnicity, mitigated through feature audits and removal if correlation \>0.3); **temporal bias** (historical data encoding past discrimination, mitigated through recency filtering and annual retraining); and **label bias** (training labels reflecting past discrimination, mitigated through audit of label generation and exclusion of labelled data from discriminatory periods).

A structured pre-training data audit prevents most bias: calculate the percentage of each protected class in training data and compare to population baseline (should match within ±5 per cent or be documented with business justification). Test whether model accuracy differs meaningfully across demographic groups (\>5 per cent difference warrants investigation). Validate that no single demographic cohort represents \>15 per cent of minority training examples (which can cause overfitting to that group). For organisations in healthcare, financial services, or public sector roles, these checks should be mandatory before any model training begins.

The governance structure: assign a "Data Quality Steward" (often a business analyst or data scientist) to each major AI project. This role owns the bias audit, maintains a decisions log documenting every data exclusion or transformation, and presents findings to legal/compliance before model training. For mid-market firms with 3–5 AI projects, a single part-time steward can manage this across projects.

## Synthetic Data Governance: Managing Risk Without Limiting Innovation

Synthetic data—artificially generated data that preserves statistical patterns of real data without containing actual personal information—offers a compliance shortcut for GDPR-constrained organisations. Rather than securing retroactive consent for training data, organisations generate synthetic datasets from aggregated, anonymised patterns and train models on synthetic examples instead.

The problem: synthetic data introduces its own governance risks. Synthetic data can encode or amplify bias if the underlying real data was biased. Synthetic data can introduce distributional artifacts that models treat as real patterns, leading to overfitting. Synthetic data lacks the explainability of natural data: if a model trained on synthetic data makes a poor prediction, tracing the failure back to a specific synthetic training example is difficult.

A governance framework for synthetic data requires: (1) **documented provenance**—maintain records of which real data fed the synthetic data generation algorithm, what transformation rules were applied, and which parameters governed the generation process; (2) **validation against holdout real data**—before using synthetic data for production model training, test whether a model trained on synthetic data achieves similar accuracy on real holdout data as a model trained on real training data; (3) **bias audit on synthetic data**—run the same demographic representation checks on synthetic data that you would run on real data, to ensure synthesis did not introduce unexpected distributional changes; (4) **lineage tracking**—document which version of synthetic data was used to train which model version, so you can reproduce or audit any model decision.

For organisations piloting synthetic data, formalising this governance framework will be the difference between a compliant AI deployment and a high-risk programme vulnerable to audit challenge.

## Cross-Border Data Transfers and the Adequacy Uncertainty

![AI data governance maturity progression from reactive to continuous](https://helium42.com/hs-fs/hubfs/blog/images/v3-g-07-governance-maturity-img-1.webp?width=600&height=600&name=v3-g-07-governance-maturity-img-1.webp)

UK and EU organisations using cloud-based AI training pipelines face ongoing uncertainty about international data transfer adequacy. Adequacy coverage differs between the EU and UK regimes and can change, creating compliance friction for organisations whose AI training pipelines route data through US, Canadian, or APAC cloud providers. Between the EU and the UK, the European Commission renewed its UK adequacy decisions in December 2025, until 27 December 2031 ([European Commission](https://ec.europa.eu/commission/presscorner/detail/en/ip_25_3059)).

Where adequacy decisions are not in place, organisations must implement contractual safeguards: Standard Contractual Clauses (SCCs) or Binding Corporate Rules (BCRs) under the EU GDPR; for UK transfers, the International Data Transfer Agreement (IDTA), the UK Addendum or UK BCRs ([ICO](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/international-transfers/a-brief-guide-to-international-transfers/)). Both regimes require a documented assessment: a transfer impact assessment under the EU GDPR, showing that the destination country's surveillance or data access regimes do not undermine GDPR protections, and a UK transfer risk assessment, showing that protection is not materially lower after the transfer. For cloud AI training (where data may be processed across multiple geographies invisibly), maintaining this documentation is operationally complex.

A governance response: (1) audit your AI training architecture to identify which cloud regions your training data touches; (2) for each region, determine whether an adequacy decision exists (check the [European Commission's official adequacy list](https://ec.europa.eu/info/law/law-topic/data-protection/international-dimension-data-protection/adequacy-decisions_en) or equivalent UK guidance); (3) for non-adequate regions, document your SCCs or BCRs and conduct a transfer impact assessment; (4) consider data residency controls: if your cloud provider allows EU/UK-only data processing, configure your training pipelines to use only EU/UK regions, even if this incurs modest cost penalties.

Adequacy decisions can be renewed, amended or challenged, so mid-market organisations should not wait for certainty. Documenting your current transfers and identifying governance gaps now prevents compliance surprises and avoids costly re-architecting of AI training pipelines.

## Establishing the Governance Structure for Mid-Market Organisations

AI data governance does not require a dedicated Data Governance Office. Mid-market organisations (150–1,500 employees) typically lack the headcount for centralised governance. Instead, a distributed model works: identify 1–3 cross-departmental Data Stewards who own governance for their domain, supported by self-service tooling and quarterly governance reviews at leadership level.

**Role: Data Steward (1 FTE per 2–3 AI projects).** The Data Steward is typically a senior analyst or mid-level data scientist who owns the data lifecycle for a specific AI application. Responsibilities include: conducting pre-training data audits, maintaining data provenance documentation, monitoring input and output data quality in production, leading bias detection activities, and escalating governance risks to the Data Governance Board. This is not a full-time governance specialist; it is a domain expert (e.g., sales operations analyst) given governance accountability alongside their existing role.

**Role: Data Governance Board (monthly, 1 hour).** Quarterly executive-level oversight. Attendees include: Chief Technology Officer or equivalent, Head of Compliance/Legal, Chief Financial Officer or equivalent, and heads of major business functions using AI (sales, marketing, operations). The board reviews: new AI projects entering governance, data quality incidents and root cause analyses, upcoming regulatory changes affecting data governance, and maturity progression metrics (percentage of production models with continuous monitoring, time-to-remediation for data quality issues, model retraining frequency). This role ensures governance decisions are aligned with business strategy and regulatory risk appetite.

**Self-Service Tools: Governance Automation.** Without tooling, governance becomes a manual bureaucratic burden that stifles AI adoption. Invest in: (1) a data catalog identifying all datasets used in AI training (can be as simple as a shared spreadsheet with ownership and retention metadata); (2) automated data quality pipeline checks (most modern cloud data platforms include these); (3) production model monitoring dashboards showing accuracy decay, distributional drift, and bias metrics for each deployed model; (4) a decisions register logging all governance decisions (data exclusions, bias mitigations, synthetic data approvals). These tools can be implemented with existing platforms (Databricks, Snowflake, Vertex AI, SageMaker) without requiring new vendor contracts.

The governance structure is intentionally lightweight to minimise overhead whilst maintaining accountability. A mid-market organisation with 5 AI projects can operate this structure with 2–3 FTE dedicated to governance across the year, supported by executive oversight occurring 4 times annually.

## Maturity Progression: From Reactive to Continuous Governance

AI data governance maturity exists on a spectrum. Understanding your current level and the path to advancement is essential for prioritising investments and measuring governance progress.

**Level 1: Unaware.** Data quality standards are either non-existent or undocumented. Input validation is manual or absent. Model performance is audited ad-hoc (monthly or quarterly) rather than continuously. Risk profile: high. A single undetected data quality issue can degrade model performance for weeks before discovery.

**Level 2: Documented.** Data quality standards are defined in writing (data dictionary, quality KPIs) but enforcement is inconsistent. Input validation occurs manually (a data analyst checks data before feeding it to models). Output audits happen monthly. Risk profile: moderate. Standards exist but gaps in enforcement create compliance exposure.

**Level 3: Partially Automated.** Quality standards are enforced via automated data pipeline checks for training data. Input validation is partially automated (schema validation, range checks, null detection) with manual review gates. Output monitoring is weekly or bi-weekly. Risk profile: low-to-moderate. Most quality issues are caught before they affect models, but high-impact failures can still occur.

**Level 4: Continuous Governance.** All three data quality layers are monitored continuously (or near-continuously) via automated pipelines. Bias detection is embedded in production model serving. Data lineage is tracked end-to-end. Model retraining is triggered automatically when accuracy decay exceeds thresholds. Risk profile: low.

The pathway from Level 2 to Level 3 typically requires 3–4 months of engineering effort and a scope-dependent investment (production-stage AI work runs from about £8,000 to £250,000+, by scope, according to [our UK AI Consultancy Price Index](https://helium42.com/reports/uk-ai-consultancy-price-index-2026)), with returns coming through reduced retraining costs and faster model deployment. Organisations should target Level 3+ as a maturity baseline for regulated industries (financial services, healthcare, legal, insurance, public sector).

## The EU AI Act and High-Risk Classification: Governance as a Legal Mandate

The EU AI Act entered into force on 1 August 2024 and created a new legal framework for AI governance, particularly for "high-risk" systems, including those listed in Annex III. High-risk systems—those used in recruitment, credit decisions, biometric identification, or law enforcement—must include documented data governance systems as a legal requirement before they are placed on the EU market; for Annex III systems, these obligations apply from 2 December 2027 ([European Commission](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai)).

The Act specifies that high-risk AI systems must include: (1) documented training data governance addressing representativeness, quality, and bias; (2) records of training, validation, and testing data used for model development; (3) data quality assurance procedures; (4) bias monitoring and mitigation strategies. Failure to document and maintain these systems exposes organisations to regulatory action, potential product delisting from the EU market, and fines of up to €15 million or 3 per cent of worldwide annual turnover (whichever is higher) for breaches of high-risk obligations, with the top tier of €35 million or 7 per cent reserved for prohibited practices ([Article 99](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng)).

The UK has not adopted equivalent legislation, and no AI bill was included in the [King’s Speech of 13 May 2026](https://www.gov.uk/government/speeches/the-kings-speech-2026). In the UK, AI data governance is regulated through existing law, chiefly the UK GDPR as amended by the Data (Use and Access) Act 2025. Organisations operating across UK and EU markets should apply the EU Act's data governance requirements wherever they place AI systems on the EU market or their outputs are used in the EU.

For organisations building or deploying AI systems in high-risk domains, governance is no longer optional or a best practice. It is a regulatory requirement. The governance investments described in this guide—data stewardship, audit trails, quality monitoring, bias detection—are the legal baseline for operating high-risk AI systems lawfully in the EU.

For a detailed reading of the EU AI Act and its data governance mandates, refer to the [official text of the EU AI Act (Regulation (EU) 2024/1689)](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng).

## Practical Audit: Assessing Your Current Data Governance Maturity

Before designing a governance programme, establish a baseline. Use this structured self-assessment to identify your current maturity level and priority gaps.

**Training Data:** (1) Do you maintain a documented record of all datasets used to train production AI models? (2) Can you produce evidence that each training dataset meets quality standards (completeness, accuracy, consistency, representativeness) before model training? (3) Do you conduct demographic representation tests to ensure protected classes are adequately represented in training data? (4) Is there a formal approval gate before training data is used for model development?

**Input Data Quality:** (1) Are schema validation and out-of-range checks automated for production model inputs? (2) Is null data flagged and monitored? (3) Are temporal anomalies (data gaps, distribution shifts) detected automatically? (4) Is there a formal escalation process when data quality thresholds are exceeded?

**Output Validation:** (1) Is model accuracy monitored in production? (2) Is bias emergence monitored (disparate impact by demographic group)? (3) Are model explanations generated automatically for high-stakes decisions? (4) Is there a formal retraining trigger when accuracy decay exceeds thresholds?

**Governance Infrastructure:** (1) Is there a named Data Steward for each major AI system? (2) Do governance decisions (data exclusions, bias mitigations, synthetic data approvals) get logged in a decisions register? (3) Is data lineage tracked (which training data fed which model version)? (4) Does executive leadership review data governance metrics at least quarterly?

Score 1 point per "yes" answer. A score of 8–12 indicates Level 2–3 maturity (governance is formalised but partially automated). A score of 4–7 indicates Level 1–2 maturity (governance is emerging but inconsistently applied). A score of 0–3 indicates Level 1 maturity (governance is largely absent).

## Building Internal Capability: Governance Skills and Training

AI data governance is a new discipline. Most organisations do not have existing staff with formal training in AI governance frameworks, regulatory requirements, or data quality monitoring for AI systems. Building internal capability requires targeted investment in three areas.

**Technical Foundations (for data scientists and engineers):** Data scientists need training in bias detection methodologies, data audit frameworks, and reproducibility requirements. Engineers need training in automated data quality testing, data lineage tracking, and model monitoring pipelines. A typical programme (2–3 days of workshops) is priced by the day, with specialist day rates of £900–£1,500 according to [our UK AI Consultancy Price Index](https://helium42.com/reports/uk-ai-consultancy-price-index-2026), and can be tailored to your technology stack.

**Governance Frameworks (for stewards and leaders):** Data Stewards and governance board members need training in AI governance maturity models, regulatory requirements (GDPR, EU AI Act, UK DPA guidance), and governance decision-making frameworks. A typical programme runs for 1–2 days and is priced on the same day-rate basis.

**Cross-Functional Alignment (leadership):** For governance to stick, business leaders (sales, marketing, operations, legal, compliance) must understand why data governance matters to their function and how governance decisions affect project timelines and risk. Targeted workshops (half-day) for each business function, conducted annually, support governance adoption.

[The NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) and the [OECD AI Principles](https://oecd.ai/en/ai-principles) (free, publicly available) provide a standards-based foundation for governance training and can be customised for your industry and organisation size.

## Governance Metrics and Measurement: Demonstrating ROI

AI data governance often competes with other investments for limited budget. Demonstrating governance ROI requires metrics that business leaders understand: cost avoidance, time savings, and risk reduction.

**Cost Avoidance Metrics:** Track the financial impact of data quality issues prevented by governance. Compare retraining and model failure costs before and after governance controls are introduced, including retraining labour, cloud compute, and downtime costs.

**Time-to-Value Metrics:** Measure the time from AI project initiation to production deployment. Mature governance can shorten this by reducing the time spent on data preparation and validation, enabling faster project delivery.

**Risk Metrics:** Track the number of data quality issues detected and remediated before they affect production models. A baseline target for Level 3 organisations: \>95 per cent of critical data quality issues detected within 24 hours. Track regulatory compliance metrics: percentage of AI systems with documented data governance, percentage of training data with documented lawful basis, percentage of production models with continuous bias monitoring.

Report these metrics to the Data Governance Board quarterly and to the executive team semi-annually. Demonstrating clear, measurable returns on governance investment builds stakeholder support and justifies continued investment in governance infrastructure.

## How Helium42 Supports AI Data Governance Implementation

Establishing AI data governance requires expertise across multiple domains: AI strategy, regulatory compliance, data engineering, and organisational change. Most mid-market organisations do not have the internal capability to build governance programmes independently. Peter Vogel, our AI Strategy Lead, has led digital marketing transformation programmes for over 500 organisations. Governance is increasingly central to our AI implementation work, and it is the focus of our [AI governance consulting support](https://helium42.com/ai-governance-consulting).

[Read our complete AI governance guide](https://helium42.com/blog/ai-governance-guide) for more context on why governance matters across your entire AI programme.

Helium42 works with regulatory-minded managing partners and scale-focused COOs to design and implement AI data governance frameworks tailored to your sector, risk appetite, and current maturity level. Our approach combines:

**Diagnostic Assessment:** We audit your current AI systems, training data, production pipelines, and governance infrastructure against established maturity models. The output: a detailed report identifying priority governance gaps, estimated remediation costs, and a phased roadmap for advancing maturity.

**Governance Framework Design:** We design role structures (Data Stewards, Governance Board), governance policies (data quality standards, bias audit procedures, synthetic data validation frameworks), and technology architecture (data catalogs, quality monitoring dashboards, model monitoring systems) tailored to your organisation size and regulatory context.

**Implementation and Capability Building:** We work alongside your data science, engineering, and compliance teams to implement governance in production, deploy monitoring infrastructure, and train staff on governance responsibilities and decision frameworks. The goal is not to hand-off a consultant deliverable; it is to embed governance practices so your team can operate independently long-term.

**Compliance Readiness:** For organisations operating in regulated sectors (financial services, healthcare, legal, insurance), we conduct regulatory gap analyses and prepare documentation for internal audit, external counsel, and potential regulatory review. This can include preparation for ICO data protection impact assessments for AI systems, EU AI Act compliance readiness reviews, and sector-specific expectations (e.g., those of the FCA, which [has said it will rely on existing frameworks](https://www.fca.org.uk/firms/innovation/ai-approach) rather than introduce extra regulations for AI).

If your organisation is deploying AI systems in regulated domains, lacking documented governance, or facing upcoming compliance deadlines, [get in touch with Helium42](https://helium42.com/get-in-touch) to discuss how we can accelerate your governance maturity and reduce compliance risk. We provide a free 30-minute governance assessment to help you understand your starting point and next steps.

## Frequently Asked Questions on AI Data Governance

**Q: Do we need AI data governance if we use pre-trained models (GPT-4, Claude, etc.) rather than building our own?**

A: Yes. Even when using pre-trained models, you must govern the data you provide for fine-tuning, the data the model processes in production, and the outputs the model generates. Governance requirements do not disappear because you are not training from scratch; they shift focus to input validation, output monitoring, and bias detection in production. Additionally, if your organisation uses a pre-trained model to process personal data (customer records, employee information), you must document lawful basis for that processing under GDPR—a governance responsibility distinct from the model vendor's responsibilities.

**Q: How much does it cost to implement AI data governance at Level 3 maturity?**

A: Implementation costs vary significantly by organisation size, technology stack, and current maturity baseline. For a mid-market organisation (150–1,500 employees) starting from Level 1, advancing to Level 3 typically requires: governance training for the staff involved (typically priced by the day, with specialist day rates of £900–£1,500 according to our UK AI Consultancy Price Index); technology tooling (data catalog, quality monitoring, model monitoring—often building on existing data platforms rather than new vendors); and implementation labour (internal team effort or external consulting) costing from about £8,000 to £250,000 or more, depending on scope. Total cost is scope-dependent, with the work typically spanning 3–4 months. The returns come through avoided retraining costs and faster project delivery.

**Q: Which regulatory bodies enforce AI data governance requirements?**

A: Enforcement varies by jurisdiction and industry. The EU AI Act is enforced through national market surveillance authorities; its prohibitions have applied since February 2025, and its obligations for Annex III high-risk systems apply from 2 December 2027. The UK Information Commissioner's Office enforces the UK GDPR and the Data Protection Act 2018, including where AI systems process personal data. Financial regulators (the FCA in the UK; national supervisors in the EU, guided by the EBA) apply their existing rules to AI used in lending, credit scoring, and investment advisory. Healthcare regulators, insurance regulators, and public sector oversight bodies all have AI governance expectations emerging in 2025–2026. For most mid-market organisations, the immediate enforcement risk comes from GDPR compliance (ICO) and sector-specific regulators, not the broader EU AI Act (unless you place AI systems on the EU market or their outputs are used in the EU).

**Q: How often should we audit data governance and model performance?**

A: Audit frequency depends on maturity level and risk. At Level 1–2, quarterly audits of training data and monthly output audits are typical. At Level 3, input data quality is monitored continuously (automated checks run hourly or daily) and output performance is monitored weekly. At Level 4, all three data layers are monitored near-continuously via automated pipelines with human review triggered only when anomalies cross thresholds. For high-risk systems (financial decisions, healthcare, legal), continuous or daily monitoring is recommended regardless of maturity level.

**Q: Can we use the same governance framework across all our AI systems, or does each system need bespoke governance?**

A: Governance frameworks should be standardised where possible—e.g., all systems should have a named Data Steward, all training data should undergo bias audit, all production systems should have continuous monitoring. However, governance policies must be tailored to risk level. A low-risk system (e.g., an internal productivity tool) might have lighter audit requirements than a high-risk system affecting customer decisions (e.g., loan approval, recruitment scoring). A tiered governance approach—light governance for low-risk systems, comprehensive governance for high-risk systems—balances standardisation with proportionate risk management.

**Q: Where should data governance oversight sit organisationally—with IT, Data, Compliance, or a dedicated office?**

A: For mid-market organisations, governance should be sponsored at the executive level (CTO or Chief Data Officer, with Compliance as a co-sponsor) but implemented as a distributed responsibility. Data Stewards sit within business functions (data scientists in engineering, business analysts in operations). A Governance Board meets quarterly with representation from IT, Compliance, business leadership, and Legal. This model avoids creating centralised bottleneck roles whilst ensuring accountability and executive visibility. A dedicated Data Governance Office is rarely justified in mid-market organisations unless you have \>20 production AI systems.

## Next Steps: Implementing AI Data Governance in Your Organisation

The regulatory and business case for AI data governance is clear. The challenge for most mid-market organisations is execution: translating governance requirements into practical programmes that do not stifle AI innovation. The framework outlined in this guide is designed for your constraints—organisations with limited governance resources, multiple business functions deploying AI, and regulators expecting documented compliance.

Start with a diagnostic assessment: score your current maturity using the self-assessment framework provided earlier. Identify your top three governance gaps. Design a 12-month roadmap advancing from your current level by one maturity level. That single increment—from Level 1 to Level 2, from Level 2 to Level 3—delivers measurable business value whilst remaining achievable with existing resources and moderate investment.

If you want expert guidance tailoring this framework to your sector, technology, and risk context, Helium42 offers a free 30-minute governance assessment for organisations exploring AI data governance for the first time. We tailor the framework to the requirements of your sector, whether financial services, healthcare, legal, retail, or manufacturing.

For more on how to integrate governance into your broader AI strategy, [read our guide on AI governance best practices](https://helium42.com/blog/ai-governance-best-practices), or explore [compliance considerations for regulated industries](https://helium42.com/blog/ai-compliance-regulated-industries).

**Related reading:** [What is AI governance](https://helium42.com/blog/what-is-ai-governance), [AI governance framework](https://helium42.com/blog/ai-governance-framework), [AI policy template](https://helium42.com/blog/ai-policy-template), [EU AI Act and UK implications](https://helium42.com/blog/eu-ai-act-uk), [AI governance, risk, and compliance](https://helium42.com/blog/ai-governance-risk-compliance).

[AI governance readiness](https://helium42.com/reports/ai-governance-readiness-benchmark-2026)

Continue reading

## More on AI Governance.

### [![Law firm supervision chain for AI-assisted drafting, from fee-earner verification through to partner sign-off](https://helium42.com/hs-fs/hubfs/blog/images/legal-ai-supervision-chain-hero-1.webp?width=600&height=450&name=legal-ai-supervision-chain-hero-1.webp) AI Legal Drafting Implementation: A Rollout Guide for UK Firms AI legal drafting implementation is the operational work of embedding AI-assisted drafting into a law firm's ... Aug 9, 2026, 4:00:00 AM](https://helium42.com/blog/ai-legal-drafting-implementation-guide)

### [![A marketing leader reviewing an AI-readiness dashboard across data, skills, governance and process](https://helium42.com/hs-fs/hubfs/blog/images/ai-readiness-md-dashboard-20260722-224137.webp?width=600&height=450&name=ai-readiness-md-dashboard-20260722-224137.webp) AI Strategy AI Readiness for Marketing Leaders: The 2026 Diagnostic Guide AI readiness for a marketing function is the combination of data, skills, governance, process, tooling and ... Jul 22, 2026, 4:00:00 AM](https://helium42.com/blog/ai-readiness-for-marketing-leaders)

### [![A UK law firm protecting confidential client data while using AI, with a secure vault and legal case files](https://helium42.com/hs-fs/hubfs/blog/images/private-legal-ai-confidentiality-20260722-230142.webp?width=600&height=450&name=private-legal-ai-confidentiality-20260722-230142.webp) AI Implementation Private and Secure AI Deployment for UK Law Firms (2026) Deploying AI privately in a law firm means choosing a model — public tool, enterprise service with a ... Jul 22, 2026, 4:00:00 AM](https://helium42.com/blog/private-legal-ai-deployment)

AI transparency

## How AI shows up in this article.

- **Drafted with AI assistance.** Research and draft prepared via frontier large language models, then human-edited by the named author.
- **Sources checked.** Statistics and citations are checked against their cited sources, and we are re-verifying older articles during 2026. If you spot an error, email [info@helium42.com](mailto:info@helium42.com) and we will correct it and date the correction.
- **Compliance posture.** EU AI Act Article 50 transparency obligations (applicable from 2 August 2026) and the UK GDPR transparency principle.

AI Newsletter

## Weekly AI insights for B2B leaders.

Practical use-cases, real client wins, and the tools we run in production. One email a week. No drip sequences, no upsells.

- **Founders write it.** Not a content team, not an AI summary — the same people delivering Helium42 engagements.
- **One email a week.** Friday morning, three to five practical items.
- **Cancel any time.** Unsubscribe link in every issue.

Want the methodology?

## The system that produced this article.

Every post on the Helium42 blog is produced through The Content System — our productised, 9-phase AI content methodology with quality gates between each phase.

[See The Content System →](https://helium42.com/ai-implementation/content-system) [Book a discovery call →](https://helium42.com/contact-us)

![Helium42](https://helium42.com/hubfs/raw_assets/public/Helium42%20Clean%20Pro%20Child%20Theme/images/logo.svg)

Practical AI transformation for European mid-market leaders. London & Hamminkeln.

#### Solutions

- [The Content System](https://helium42.com/#content-system)
- [Salesman Sam](https://helium42.com/ai-implementation/pre-build-ai-solutions/salesman-sam)
- [AI Training](https://helium42.com/ai-training)
- [Strategic Consultancy](https://helium42.com/ai-consultancy)

#### Company

- [About](https://helium42.com/about-us)
- [Methodology](https://helium42.com/#approach)
- [FAQ](https://helium42.com/#faq)
- [Contact](https://helium42.com/contact-us)
- [Blog](https://helium42.com/blog)

#### Contact

- [info@helium42.com](mailto:info@helium42.com)
- [+44 (0) 208 834 4795](tel:+442088344795)
- 80A Uxbridge Road, W12 8LR
- London · Hamminkeln

© 2026 Helium42. All rights reserved.

[Privacy](https://helium42.com/privacy-policy) [Terms](https://helium42.com/terms-of-service) [Disclaimer](https://helium42.com/disclaimer) [Anti-Spam](https://helium42.com/anti-spam-policy)

Helium42 is a trading name of White Hat SEO Limited, registered in England and Wales, company number 07802734. Registered office: 75 Aston Road, Shifnal, England, TF11 8DU.

```json
{
  "@context" : "https://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Peter Vogel",
    "url" : "https://helium42.com/blog/author/peter-vogel"
  },
  "datePublished" : "2026-03-24T14:00:00.000Z",
  "headline" : "AI Data Governance: A Framework for UK and EU Organisations",
  "image" : [ "https://helium42.com/hubfs/blog/images/v3-g-07-ai-data-governance-hero-hero-1.webp" ],
  "mainEntityOfPage" : {
    "@id" : "https://helium42.com/blog/ai-data-governance",
    "@type" : "WebPage"
  },
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "url" : "https://helium42.com/hubfs/Favicon-1.png"
    }
  }
}
```

```json
{
  "@context" : "https://schema.org",
  "@type" : "Article",
  "author" : {
    "@type" : "Person",
    "name" : "Peter Vogel",
    "url" : "https://helium42.com/about-us"
  },
  "dateModified" : "2026-11-13T00:00:00Z",
  "datePublished" : "2026-03-24",
  "description" : "A comprehensive guide to establishing AI data governance frameworks for mid-market organisations navigating GDPR compliance, regulatory requirements, and the business imperative to deploy AI systems responsibly.",
  "headline" : "AI Data Governance: A Practical Framework for UK and EU Organisations",
  "image" : "https://145656140.fs1.hubspotusercontent-eu1.net/hubfs/145656140/blog/images/v3-g-07-ai-data-governance-hero-hero-1.webp",
  "mainEntityOfPage" : "https://helium42.com/blog/ai-data-governance",
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "url" : "https://helium42.com/hubfs/raw_assets/public/Helium42%20Clean%20Pro%20Child%20Theme/images/logo.svg"
    },
    "name" : "Helium42",
    "url" : "https://helium42.com"
  }
}
```

```json
{
  "@context" : "https://schema.org",
  "@type" : "FAQPage",
  "mainEntity" : [ {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Yes. Even when using pre-trained models, you must govern the data you provide for fine-tuning, the data the model processes in production, and the outputs the model generates. Governance requirements do not disappear because you are not training from scratch; they shift focus to input validation, output monitoring, and bias detection in production. Additionally, if your organisation uses a pre-trained model to process personal data (customer records, employee information), you must document lawful basis for that processing under GDPR—a governance responsibility distinct from the model vendor's responsibilities."
    },
    "name" : "Do we need AI data governance if we use pre-trained models (GPT-4, Claude, etc.) rather than building our own?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Implementation costs vary significantly by organisation size, technology stack, and current maturity baseline. For a mid-market organisation (150–1,500 employees) starting from Level 1, advancing to Level 3 typically requires: governance training for the staff involved (typically priced by the day, with specialist day rates of £900–£1,500 according to our UK AI Consultancy Price Index); technology tooling (data catalog, quality monitoring, model monitoring—often building on existing data platforms rather than new vendors); and implementation labour (internal team effort or external consulting) costing from about £8,000 to £250,000 or more, depending on scope. Total cost is scope-dependent, with the work typically spanning 3–4 months. The returns come through avoided retraining costs and faster project delivery."
    },
    "name" : "How much does it cost to implement AI data governance at Level 3 maturity?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Enforcement varies by jurisdiction and industry. The EU AI Act is enforced through national market surveillance authorities; its prohibitions have applied since February 2025, and its obligations for Annex III high-risk systems apply from 2 December 2027. The UK Information Commissioner's Office enforces the UK GDPR and the Data Protection Act 2018, including where AI systems process personal data. Financial regulators (the FCA in the UK; national supervisors in the EU, guided by the EBA) apply their existing rules to AI used in lending, credit scoring, and investment advisory. Healthcare regulators, insurance regulators, and public sector oversight bodies all have AI governance expectations emerging in 2025–2026. For most mid-market organisations, the immediate enforcement risk comes from GDPR compliance (ICO) and sector-specific regulators, not the broader EU AI Act (unless you place AI systems on the EU market or their outputs are used in the EU)."
    },
    "name" : "Which regulatory bodies enforce AI data governance requirements?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Audit frequency depends on maturity level and risk. At Level 1–2, quarterly audits of training data and monthly output audits are typical. At Level 3, input data quality is monitored continuously (automated checks run hourly or daily) and output performance is monitored weekly. At Level 4, all three data layers are monitored near-continuously via automated pipelines with human review triggered only when anomalies cross thresholds. For high-risk systems (financial decisions, healthcare, legal), continuous or daily monitoring is recommended regardless of maturity level."
    },
    "name" : "How often should we audit data governance and model performance?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "Governance frameworks should be standardised where possible—e.g., all systems should have a named Data Steward, all training data should undergo bias audit, all production systems should have continuous monitoring. However, governance policies must be tailored to risk level. A low-risk system (e.g., an internal productivity tool) might have lighter audit requirements than a high-risk system affecting customer decisions (e.g., loan approval, recruitment scoring). A tiered governance approach—light governance for low-risk systems, comprehensive governance for high-risk systems—balances standardisation with proportionate risk management."
    },
    "name" : "Can we use the same governance framework across all our AI systems, or does each system need bespoke governance?"
  }, {
    "@type" : "Question",
    "acceptedAnswer" : {
      "@type" : "Answer",
      "text" : "For mid-market organisations, governance should be sponsored at the executive level (CTO or Chief Data Officer, with Compliance as a co-sponsor) but implemented as a distributed responsibility. Data Stewards sit within business functions (data scientists in engineering, business analysts in operations). A Governance Board meets quarterly with representation from IT, Compliance, business leadership, and Legal. This model avoids creating centralised bottleneck roles whilst ensuring accountability and executive visibility. A dedicated Data Governance Office is rarely justified in mid-market organisations unless you have >20 production AI systems."
    },
    "name" : "Where should data governance oversight sit organisationally—with IT, Data, Compliance, or a dedicated office?"
  } ]
}
```

```json
{
  "@context" : "https://schema.org",
  "@id" : "https://helium42.com/#organization",
  "@type" : "Organization",
  "areaServed" : [ {
    "@type" : "Country",
    "name" : "United Kingdom"
  }, {
    "@type" : "Country",
    "name" : "Germany"
  }, {
    "@type" : "Place",
    "name" : "Europe"
  } ],
  "contactPoint" : [ {
    "@type" : "ContactPoint",
    "areaServed" : [ "GB", "DE", "EU" ],
    "availableLanguage" : [ "en", "de" ],
    "contactType" : "customer support",
    "email" : "info@helium42.com",
    "telephone" : "+44 208 834 4795"
  } ],
  "description" : "Helium42 is an AI consultancy based in London and Hamminkeln. We deliver practical AI transformation for European mid-market leaders through an AI-produced content system, the Salesman Sam sales agent, hands-on education, and strategic consultancy.",
  "foundingLocation" : {
    "@type" : "Place",
    "name" : "London, United Kingdom"
  },
  "knowsAbout" : [ "Artificial Intelligence", "AI Consultancy", "AI Implementation", "AI Education", "Content Automation", "Answer Engine Optimization", "Sales Automation", "HubSpot CMS", "Business Transformation" ],
  "legalName" : "White Hat SEO Limited",
  "logo" : {
    "@type" : "ImageObject",
    "height" : 342,
    "url" : "https://145656140.fs1.hubspotusercontent-eu1.net/hubfs/145656140/brand/helium42-logo-schema.png",
    "width" : 1008
  },
  "name" : "Helium42",
  "sameAs" : [ "https://www.linkedin.com/company/helium42" ],
  "url" : "https://helium42.com"
}
```

```json
{
  "@context" : "https://schema.org",
  "@id" : "https://helium42.com/#office-london",
  "@type" : "LocalBusiness",
  "address" : {
    "@type" : "PostalAddress",
    "addressCountry" : "GB",
    "addressLocality" : "London",
    "postalCode" : "W12 8LR",
    "streetAddress" : "80A Uxbridge Road"
  },
  "email" : "info@helium42.com",
  "geo" : {
    "@type" : "GeoCoordinates",
    "latitude" : 51.5074,
    "longitude" : -0.1278
  },
  "name" : "Helium42 London",
  "parentOrganization" : {
    "@id" : "https://helium42.com/#organization"
  },
  "telephone" : "+44 208 834 4795",
  "url" : "https://helium42.com"
}
```

```json
{
  "@context" : "https://schema.org",
  "@id" : "https://helium42.com/#office-hamminkeln",
  "@type" : "LocalBusiness",
  "address" : {
    "@type" : "PostalAddress",
    "addressCountry" : "DE",
    "addressLocality" : "Hamminkeln"
  },
  "email" : "info@helium42.com",
  "name" : "Helium42 Hamminkeln",
  "parentOrganization" : {
    "@id" : "https://helium42.com/#organization"
  },
  "url" : "https://helium42.com"
}
```

```json
{
  "@context" : "https://schema.org",
  "@id" : "https://helium42.com/#website",
  "@type" : "WebSite",
  "inLanguage" : "en-GB",
  "name" : "Helium42",
  "potentialAction" : {
    "@type" : "SearchAction",
    "query-input" : "required name=search_term_string",
    "target" : {
      "@type" : "EntryPoint",
      "urlTemplate" : "https://helium42.com/?s={search_term_string}"
    }
  },
  "publisher" : {
    "@id" : "https://helium42.com/#organization"
  },
  "url" : "https://helium42.com"
}
```

```json
{
  "@context" : "https://schema.org",
  "@id" : "https://helium42.com/blog/ai-data-governance#post",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Peter Vogel"
  },
  "dateModified" : "2026-11-13T21:57:07+00:00",
  "datePublished" : "2026-03-24T14:00:00+00:00",
  "description" : "Organisations across the UK and EU are rapidly deploying artificial intelligence systems to improve efficiency, reduce costs, and enhance decision-making. Yet the very data these systems depend on remains largely ...",
  "headline" : "AI Data Governance: A Practical Framework for UK and EU Organisations",
  "image" : "https://145656140.fs1.hubspotusercontent-eu1.net/hubfs/145656140/blog/images/v3-g-07-ai-data-governance-hero-hero-1.webp",
  "inLanguage" : "en-GB",
  "isPartOf" : {
    "@id" : "https://helium42.com/blog#blog"
  },
  "keywords" : "AI Governance",
  "mainEntityOfPage" : "https://helium42.com/blog/ai-data-governance",
  "publisher" : {
    "@id" : "https://helium42.com/#organization"
  },
  "url" : "https://helium42.com/blog/ai-data-governance"
}
```

```json
{
  "@context" : "https://schema.org",
  "@type" : "BreadcrumbList",
  "itemListElement" : [ {
    "@type" : "ListItem",
    "item" : "https://helium42.com/",
    "name" : "Home",
    "position" : 1
  }, {
    "@type" : "ListItem",
    "item" : "https://helium42.com/blog",
    "name" : "Blog",
    "position" : 2
  }, {
    "@type" : "ListItem",
    "item" : "https://helium42.com/blog/tag/ai-governance",
    "name" : "AI Governance",
    "position" : 3
  }, {
    "@type" : "ListItem",
    "item" : "https://helium42.com/blog/ai-data-governance",
    "name" : "AI Data Governance: A Practical Framework for UK and EU ...",
    "position" : 4
  } ]
}
```