Capital markets leaders are in the early stages of implementing comprehensive artificial intelligence governance frameworks as they begin to realise the challenges as well as the opportunities offered by the technology.
As the adoption of AI accelerates it’s becoming apparent that it needs its own set of rules on how it can be effectively and safely rolled out and used within organisations, a recent A-Team Live webinar discussed. The nascent development of these, however, was underlined by a poll of viewers during the event; it showed about a third of respondents had “moderately established” frameworks while another third was still “thinking about” drafting guidelines.The webinar, entitled “Hearing from the Experts: AI Governance Best Practices”, featured experts Gaston Hummel, EMEA Lead Strategic Services at Precisely; Nicole Hansen, Compliance and Conduct Lead at NatWest Group; and Jovita Tam, Business-focussed Data/AI Adviser and Attorney (England & NY).
Moderated by Data Management Insight editor Mark McCord, the webinar established early that foundational to any governance code was good quality data. Panellists consistently underscored its criticality, highlighting that without high-quality data, AI models risk generating erroneous or misleading outputs, known as “hallucinations”.
What is Good?
One panellist suggested that “good” was too vague a determinant for data quality and instead urged organisations to strive for “fit for purpose data quality”, defining precisely what quality attributes – completeness, accuracy, integrity, provenance – are essential for specific AI use cases.
The prevalence of weak data governance foundations had led many firms to try “leapfrogging” to AI without adequately addressing unstructured, siloed or inconsistent data. Critically, it was noted that data often was not collected primarily for its eventual use in AI applications and thus introduced biases or privacy concerns from the outset. This necessitates an end-to-end measurement of bias and representativeness across the entire model pipeline, not merely at the training data stage.
Strategic Approach
Developing an AI governance framework requires a strategic approach, the panel agreed, noting key considerations including transparency, ensuring that AI initiatives are not “skunkworks projects” but are understood across the organisation and by external stakeholders.
Operationalising policies early was another crucial piece of advice, as was guidance that frameworks should be designed for continuous improvement, acknowledging that initial implementations will not be perfect.
The regulatory environment for AI is rapidly taking shape, the webinar heard, with the EU AI Act emerging as a pivotal piece of legislation, demanding continuous monitoring, incident reporting, transparency, explainability and third-party oversight of AI models by August 2026. While the Act may set a de facto standard, the global regulatory landscape remains fragmented, the webinar heard.
Shared Responsibility
Another poll revealed that AI governance responsibility is predominantly shared across multiple departments. The experts responded by saying that executive leadership teams must set the overarching AI agendas, providing authority and direction to both a data governance teams and a cross-functional “AI excellence steering team”.
Measuring the effectiveness of an AI governance policy involves evaluating regulatory adherence, incident frequencies, risk management, transparency, stakeholder trust and the tangible value realised from AI initiatives, the webinar heard.
For organisations embarking on their AI governance journey, the panellists suggested getting started as soon as possible while ensuring any framework aligns with the organisation’s current capabilities and awareness. Investing in workforce development is paramount, the experts agreed, as was aligning AI use cases with corporate objectives from the outset.
Subscribe to our newsletter
