Data & AI Governance Unification for Data Professionals with Mario Cantin

Executive Summary

This webinar discusses the critical challenges and solutions involved in unifying data, Artificial Intelligence (AI), and records management. Howard Diesel and Mario Cantin highlight the importance of anomaly detection and data quality within AI governance, as well as the complexities and regulatory considerations associated with AI initiatives, including chatbot development. Mario underscores the significance of governance frameworks in AI deployment, balancing compliance and risk, and fostering evidence-based trust in AI systems, with a focus on implementing Data Governance using platforms such as Prodago. Lastly, the webinar emphasises the integration of agentic frameworks, project-based activities, and ethical considerations, all aimed at enhancing decision-making processes and ensuring robust data stewardship in AI governance.

Webinar Details

Title: Data & AI Governance Unification for Data Professionals with Mario Cantin
Date: 2025-08-28
Presenter: Howard Diesel & Mario Cantin
Meetup Group: African Data Management Community
Write-up Author: Howard Diesel

Challenges and Solutions in Unifying Data, AI, and Records Management

Howard Diesel opened the webinar and shared his feelings on a recent webinar on Data Governance within the Organisation, identifying the need to unify various aspects, including Data Governance, AI governance, and records management. As AI systems become more integrated, particularly with chatbots and human-in-the-loop processes that collect sensitive records, it is crucial to address these governance requirements collectively.

For instance, Howard noted that it is crucial to consider the potential implications of providing potentially incorrect advice through chatbots, as this could lead to liability claims from users. Another relevant case that was highlighted was that of an airline facing legal consequences for failing to honour promised bereavement fees, emphasising the importance of robust governance frameworks to avoid similar issues.

The management of diverse content data—ranging from unstructured and semi-structured data to tacit knowledge—necessitates a cohesive approach, particularly within the insurance claim process automation. A key aspect of this is the First Notice of Loss (FNOL), where chatbots can play a crucial role. These chatbots help collect essential information from claimants, guiding them to provide accurate data while also identifying gaps or unclear responses. As discussed in Mario’s upcoming presentation, integrating AI and chatbots into this process is complex due to various challenges and regulatory considerations.

Figure 1 Anomaly Detection

Figure 2 AI Governance in the Insurance Claim Process

Anomaly Detection and Data Quality in AI Governance

Advancements in claims processing, particularly in anomaly detection and triage procedures, are revolutionising the industry. Howard highlighted the necessity of normalising data and the critical role of human insight in interpreting anomalies, as not all flagged claims can be immediately deemed fraudulent. Data stewards play an essential role in this process, as they provide context and understanding behind algorithmic decisions, thereby enhancing data quality and ensuring that every decision regarding anomalous claims is well-explained.

Mario Cantin expressed gratitude to Howard and introduced his discussion on AI governance frameworks, emphasising the impact of specific elements such as First Notice of Loss (FNOL) and chatbots on governance decisions. He outlined his approach by starting with foundational aspects of the framework before delving into more detailed analyses, indicating a structured exploration of the topic.

Figure 3 Data Normalisation

Figure 4 Anomaly Detection Techniques and Normalisation

Challenges and Components of AI Governance

Organisations looking to leverage AI, particularly in sectors such as insurance, with use cases like automated claims processing, face numerous challenges despite the potential for significant business value. A recent MIT report indicates that 95% of organisations experience no return on investment from generative AI, highlighting the complexity involved.

When developing components like FNOL capabilities or chatbots, companies must address critical facets of AI governance, including quality assurance, explainability, data sovereignty, and risk mitigation. As organisations pursue AI initiatives, understanding these multifaceted governance requirements is essential for successful implementation.

AI governance is structured around six key pillars: AI frameworks, data operations, risk management, cybersecurity, ethics, and privacy systems. A fundamental aspect of successful AI governance is data operation, as effective Data Management and Governance are prerequisites for AI implementation.
Organisations often struggle to showcase the value delivered by Data Management, but aligning Data Management activities with AI governance can enhance success. Additionally, governance should be integrated throughout the entire AI lifecycle to ensure comprehensive application and oversight.

When governing AI within an organisation, it is essential to address several key aspects: determining the appropriate timing for AI implementation, establishing guidelines for AI procurement that include contractual requirements and obligations, and focusing on specific governance needs for AI adoption. Additionally, organisations must consider governance requirements at various stages of AI development, including design, deployment, and ongoing operation, ensuring that each phase aligns with strategic goals and regulatory standards.

Figure 5 Deep Dive

Figure 6 Business Objectives to Business Value

Figure 7 AI Governance Framework

Understanding Governance Frameworks in AI Initiatives

Mario shared that the framework is the culmination of collaborative efforts aimed at recognising the necessity of orchestrating various governance facets rather than managing them in isolation. He highlighted that a staggering 95% of Generative AI initiatives fail, emphasising the importance of understanding and prioritising relevant elements specific to an organisation.

Organisations do not need to adopt every component of governance frameworks—rather, they should identify the critical aspects pertinent to them. Additionally, it is essential to consider various governance pillars, including Data Management, as well as the specific standards and regulations that apply, such as ISO 42000, to ensure a comprehensive and effective governance strategy.

Leveraging chatbots creates a powerful business solution for organisations to enhance efficiency and customer engagement. However, while the potential benefits are substantial, it’s crucial to avoid analysis paralysis caused by an overwhelming amount of information. To ensure successful implementation, organisations should focus on identifying critical starting points and governance components that will streamline processes, ultimately leading to innovation and growth rather than stagnation.

Figure 8 Laws, Standards and Policies

Figure 9 “Where to start?”

The Importance of Governance Requirements in AI Deployment

When establishing governance requirements for AI implementation, it is essential to consider the specific use case, such as developing First Notice of Loss (FNOL) capabilities with a chatbot versus conducting fraud detection on structured databases. The critical components of governance vary based on the AI properties and techniques employed.

It is crucial to prioritise the most relevant governance specifications according to the intended application. Additionally, one must assess the organisation’s risk tolerance, particularly in sensitive sectors like finance, where the potential for false positives can lead to significant consequences, such as inadvertently sending incorrect information to clients. Understanding these factors is key to successful AI governance.

Risk tolerance plays a crucial role in determining the prioritisation of governance aspects, as those with low risk tolerance may require stricter governance measures. Additionally, the compliance context is significant; for instance, organisations operating under regulations like India’s DPDPA or California’s CCPA must consider how these privacy laws affect their projects.

Existing capabilities also influence governance deployment; newcomers to AI should start with foundational governance elements that align with their current capabilities, gradually enhancing governance as their maturity develops. This approach ensures that governance intensity is appropriately adjusted based on the project’s context and the organisation’s progression.

Balancing Compliance and Risk in AI Project Development

Many projects often commence without a thorough assessment of compliance and risk factors, which can lead to potential pitfalls down the road. While the initial allure may tempt organisations to dive in without a clear plan or understanding of their capabilities, this approach can result in challenges and false starts.

Engaging in a well-defined framework can help navigate these complexities; however, many may not be familiar with the various frameworks available. Understanding these elements is crucial in avoiding unnecessary setbacks and ensuring a successful project initiation.

The balance between compliance and experimentation is crucial in legal practices surrounding the AI Act. This, Matio noted, would reveal a spectrum that ranges from strict adherence to risk and compliance frameworks to a more flexible approach of experimentation, where practitioners might adopt a “let’s just do it and see what happens” mindset.

While an unstructured approach can result in failure, the primary challenge lies in identifying the minimum necessary elements to achieve initial objectives without inundating the process with excessive regulations. Ultimately, adopting this pragmatic assessment allows for effective action that still acknowledges essential compliance factors, fostering innovation within a compliant framework.

To make informed decisions about deploying an AI system, it is crucial to identify the key parameters and associated risks that stakeholders consider when evaluating approval for production. Failing to address these risks upfront can lead to significant delays and potential failure.

The focus should be on determining the minimum requirements needed for a successful initial iteration, rather than attempting to address every possible aspect before starting. This approach enables progress and the opportunity to demonstrate the value of the AI system, thereby avoiding a situation where excessive planning leads to stagnation.

Complexities and Regulations of Chatbots

The management of chatbots, especially in the context of FNOL, involves intricate complexities and stringent regulatory requirements. Mario points out that Peking University has developed a comprehensive framework that surveys these necessary governance structures, underscoring the extensive oversight required for effective chatbot development. Although he intended to present this framework during the discussion, he acknowledged the need to explore that aspect further, mentioning a compliance playbook that could provide valuable insights into the essential policies for chatbot implementation. Overall, understanding these regulatory frameworks is crucial for ensuring that chatbot systems are both compliant and effective in their operations.

Figure 10 Insurance Automated Claim

Considerations for Data Safety and Stewardship in Chatbot Development

When creating a chatbot, it’s essential to address several key requirements, particularly those related to data security. A collaborative document from Peking University and a U.S. university emphasises the importance of pre-training data safety. This involves employing various techniques for data filtering to eliminate undesirable content from the training dataset. Additionally, Organisations must ensure that, when using publicly available data, any private or sensitive information is carefully removed to safeguard user privacy and enhance the chatbot’s safety and reliability.

To ensure that unsafe domains are excluded from the training data for chatbots, it is essential to implement a comprehensive filtering approach. This involves defining specific domains to block and applying various techniques to eliminate undesirable content from the input dataset. One key aspect is model-based filtering, which focuses on identifying and managing content duplication, as repeated content can inadvertently direct the chatbot’s attention. Ultimately, the choice of techniques and the extent of their application should consider both technical and contextual factors, rather than being purely technical decisions.

Understanding the implications of stewardship is essential for assessing an organisation’s needs and tolerance levels regarding the development of chatbot systems. A critical aspect of LLM (Large Language Model) safety involves recognising the potential for these systems to induce deception in users.

It is important to explore available techniques for measuring and addressing deception to ensure safety. The responsibility for these measures extends beyond just the developers; it requires a contextual evaluation of the organisation’s appetite for risk and the potential negative outcomes specific to their environment.

Figure 11 Large Language Model Safety

Figure 12 Large Language Model Safety pt.2

Figure 13 Large Language Model Safety pt.3

Figure 14 Large Language Model Safety pt.4

Evidence-Based Trust in USAI Systems

In developing chatbots, it’s essential to recognise the various frameworks that apply to their creation, each with specific requirements. Prodago plays a crucial role by integrating these frameworks and translating them into actionable tasks that align with best practices. For instance, one key action identified is the need to remove undesirable content prior to training the chatbot. This approach ensures that projects are guided by appropriate standards while facilitating the execution of necessary steps within the development process.

The orchestration of various frameworks, such as GDPR and technical requirements for chatbots, is crucial for ensuring project success by meeting critical minimum standards. Organisations must identify key elements tailored to their risk tolerance to navigate these complexities effectively, thereby enhancing their project’s viability. Furthermore, addressing concerns about transparency and risk mitigation in the creation of AI systems is essential, as stakeholders seek assurance of compliance. By employing evidence-based artefacts, organisations can foster trust and demonstrate their commitment to responsible and compliant project management.

To ensure effective oversight of a project, it is essential to identify the necessary artefacts that demonstrate the successful execution of specific activities, such as removing undesirable content before training. Project teams should provide evidence of these actions, which may include a list of blocked domains and documentation of measures taken to mask personal information within the training dataset. This practice of producing artefacts as a gating requirement fosters evidence-based decision-making and varies depending on the project’s context and the type of data involved.

Implementation of Data Governance of AI Systems on the Prodago Platform

Recent research indicates that ensuring the safety and reliability of chatbot technology requires comprehensive methods to identify and mitigate deceptive behaviour. This includes training systems on specific behaviours to avoid and tagging these as areas of concern. By focusing on these negative behaviours, developers can better manage chatbot responses. Furthermore, advancements in AI now enable the alignment of chatbot behaviour with established policy parameters, thereby enhancing overall system integrity and user trust.

Recent advancements in chatbot training have shifted from addressing issues individually to establishing defined parameters for AI development. A key focus is ensuring that chatbots do not promote violent or inappropriate responses, particularly in scenarios such as inquiries about making bombs.

Effective strategies include refusing to provide answers to harmful questions and redirecting users toward appropriate support or resources. For instance, a chatbot designed for first notice of loss should be equipped with strict guardrails to prevent engagement in unrelated or sensitive discussions, maintaining safety and relevance in interactions.

Prodago’s specialised platform offers a comprehensive suite of playbooks tailored to navigate critical regulatory frameworks like GDPR and ISO 42001. By implementing a structured approach to AI projects, the platform facilitates essential activities such as risk assessment and evidence collection, which are vital for effective AI governance. This structured methodology not only enhances understanding and trust in AI initiatives but also ensures compliance with regulatory standards. To provide a clearer understanding of its functionalities, a demonstration of the Prodago platform has been proposed as a valuable next step.

ISO 42001 outlines the framework for establishing an AI management system, focusing on the overarching management of AI systems rather than their technical development. It emphasises the importance of implementing an AI policy, which involves defining guidelines for the development and use of AI systems. Additionally, ISO 42001 requires organisations to assess the potential impacts and risks associated with the deployment of AI systems. In this context, organisations, such as Prodago, analyse these requirements to determine the necessary actions for compliance and responsible AI management.

The process of translating frameworks into actionable steps is essential for determining the minimum requirements necessary for project success. This involves assessing potential impacts and producing relevant artefacts, such as a probability severity of ARM (Application Risk Management) assessment. The aim is to simplify the system development process through the use of well-defined templates.

The platform orchestrates various frameworks, including ISO, GDPR, D-CAM, and DMBoK, and serves as a playbook for AI projects, such as fraud detection and prevention systems. It includes a questionnaire to gather project context, data sources, existing data quality, and model characteristics, facilitating the automatic identification of required activities.

Figure 15 Knowledge Management and Oversight

Figure 16 ISO 42001 – Information technology

Figure 17 ISO 42001 – Information technology pt.2

AI Governance and Evidence-Based Compliance

AI governance requires effective management through various laws, frameworks, and standards, which can quickly become overwhelming for projects. To streamline this process, developing playbooks that guide teams in addressing the necessary requirements is essential. These playbooks simplify project workflows by providing questionnaires and a checklist of actions to assess potential impacts, aligning with standards such as ISO and NIST.

The objective is to maintain a manageable complexity at the governance level while delivering minimal requirements tailored to specific initiatives, ensuring that projects, such as developing a chatbot for FNOL, meet the crucial criteria to avoid falling into the category of initiatives that fail to reach production.

The process involves completing a questionnaire to gather data relevant to the project, which then facilitates the suggestion of operational procedures aligned with various frameworks. This approach alleviates concerns about specific frameworks by guiding users through essential operational practices necessary for compliance. At the project’s conclusion, users receive a clear overview of the required tasks and evidence needed to demonstrate compliance, ensuring a straightforward path to achieving project goals while effectively managing associated risks.

Mario outlined the concept of evidence-based governance in project management, emphasising the importance of identifying critical data and collecting artefacts throughout different project phases. Trust in the delivered system is correlated with the percentage of approved artefacts, which serve as proof that essential activities have been completed. By ensuring that these elements are collected and validated by the appropriate stakeholders, organisations can establish a reliable framework for trust in the developed AI system. Ultimately, demonstrating the production of this evidence often becomes a gating requirement before the system can move into production.

Figure 18 Fraud Detection and Prevention System

Figure 19 Fraud Detection and Prevention System pt.2

Figure 20 Fraud Detection and Prevention System pt.3

Figure 21 Fraud Detection and Prevention System pt.4

The Integration of Agentic Frameworks in AI Initiatives

Prodago serves as the orchestration layer for integrating various technologies within agentic frameworks, such as N8N. It facilitates automation by enabling the execution of required activities—such as documenting business terms or managing AI cases—across various tools, including Purview, Calibra, or CDGC. By coordinating these tasks, the platform provides a cohesive method to manage metadata requirements and collections, ensuring that activities are clearly linked to the appropriate technologies for efficient execution and monitoring.

The orchestration layer above technology allows organisations to evolve their processes without being constrained by specific tools. Even if teams are using basic solutions like Excel, they can still perform necessary activities and adapt to new technologies over time. The governance role involves coordinating efforts among project teams to ensure that tasks, such as documentation and privacy impact assessments, are completed efficiently, often using different tools like OneTrust. Effective orchestration is crucial for identifying responsibilities and addressing potential gaps that could hinder the success of initiatives.

Coordinating activities within a technological framework is essential for improving efficiency and ensuring compliance through automation. By integrating third-party tools into an agentic framework, organisations can systematically execute repetitive tasks and conduct overnight checks against established checkpoints.

This approach not only reduces manual effort, particularly during frequent updates, but also enables the identification of any violations that arise from changes. Ultimately, such automation fosters trust and reassurance by ensuring that assessments related to potential impacts on individuals and the environment align with internal policies and directives.

Many elements of a process are suitable for automation, and it is essential to automate them as much as possible. However, for those elements that remain manual, it is crucial to define a clear journey towards full automation. While tools such as APIs must be executed and structured effectively, it’s important to maintain an overarching understanding of how both automated and manual components fit into the broader initiative. This comprehensive perspective ensures that all necessary tasks are addressed, even in the absence of complete automation.

Figure 22 Fraud Detection and Prevention System pt.5

Project-Based Activities and Data Quality Management in a Platform

The integration of project-based activities and standardised templates within a platform is crucial for enhancing data quality and user guidance. An attendee emphasised the need for clear templates for submissions, drawing from his experiences with the DMBoK framework and previous work with the NDMO and A15 domains. They highlighted how these templates can help users produce high-quality artefacts. Mario then acknowledged an attendee’s insights and the necessity of linking activities to operational procedures and artefacts to strengthen the platform’s functionality further. Overall, the collaborative efforts aim to create a more structured and effective approach for users, promoting better data quality and operational efficiency.

Prodago offers an extensive repository of approximately 3,000 predefined operating practices, accompanied by artefacts and templates to facilitate their implementation. These practices encompass a range of topics, with a primary focus on quality management, offering essential resources for developing a data quality management strategy. Key frameworks such as DCAM, DMBoK, and BCBS 239 Principle 27 are integral to these practices, serving as foundational elements for effective governance. This vast inventory serves as the actionable material necessary for organisations to enhance their operational processes.

Figure 23 DAMA -DMBoK

Figure 24 Operating Practices

Figure 25 Operating Practices pt.2

AI Techniques in Operational Procedures and Evaluation

The integration of AI techniques, such as anomaly detection and risk scoring, into operational procedures is crucial for effectively addressing targeted use cases. This approach allows for a high degree of flexibility and customisation, catering to unique organisational requirements. For instance, the NDMO’s experience with approximately 600 templates highlights the importance of defining metadata, which contributes to a more structured artefact design. Furthermore, scorecards play a critical role in evaluating the effectiveness of these artefacts, ensuring that submissions adhere to quality standards rather than being accepted without scrutiny. Ultimately, these elements together enhance the operational integrity and success of AI integration.

Training in Business Contexts and AI Governance

Effective hands-on training with software is essential for successful AI governance insurance, as highlighted by an upcoming session with MetLife that builds on existing relationships and expertise. This training will underscore the importance of effective encoding as critical proof points, while also cautioning against the high costs and risks associated with manual execution, which can impede AI initiatives. Striking a balance between defensive risk management and demonstrating the value of AI is crucial;. In contrast, robust defensive strategies are necessary, they must not overshadow the need for business support in advancing AI projects.

The core value of the initiative lies in automating various checks by allowing agents to define templates and perform scoring, which can enhance outcomes, especially since not all agents are native English speakers. This approach not only aims to streamline processes but also focuses on effective knowledge management by consolidating key artefacts, such as glossaries and policies, into a comprehensive knowledge base. Furthermore, the goal is to encourage actionable project work while maintaining an understanding of existing frameworks without becoming overly reliant on them. By abstracting these frameworks, the initiative seeks to eliminate unnecessary debates and complexities that can hinder progress, particularly in the context of AI initiatives.

Identifying use cases is crucial when evaluating a large number of Operational Processes (OPs), such as 4000 OPs, especially in the context of implementing NLP chat systems. A key factor in this evaluation is understanding risk tolerance, which plays a significant role in prioritising these OPs. With regulations like the European AI Act imposing strict compliance requirements, organisations must carefully assess their most pressing needs by identifying high-risk elements and determining the necessary maturity levels to navigate these challenges effectively. In conclusion, a strategic approach to use case identification and risk assessment is vital for successful implementation and compliance in operational processes.

The focus of governance should be on aligning business objectives with existing capabilities rather than solely building governance capacity. It is essential to determine the critical subset of actions necessary for success while considering the organisation’s current maturity level. The goal is to facilitate agility in connecting these governance requirements with business needs, ensuring that only the minimum necessary elements are pursued to achieve success. Evidence of accomplishment should be gathered to demonstrate progress, avoiding the common pitfall of remaining at the proof-of-concept stage.

Figure 26 Governance Requirements

Ethical Considerations and AI Governance

Navigating complex operational frameworks while ensuring compliance and accountability is essential in today’s landscape. Participants in a recent discussion emphasised the importance of adhering to established guidelines, such as ISO 42001 and the NIST AI Risk Management Framework, while also highlighting the need to prioritise actionable steps over becoming overwhelmed.

The ethical divide among organisations regarding compliance with robust frameworks presents a significant challenge in maintaining integrity within the industry. This disparity is evident across various entities, from individuals to state-backed organisations, highlighting the varying commitment levels to ethical practices. Consequently, it underscores the urgent need for a collective effort to enhance compliance and foster a culture of integrity across all sectors.

Flexibility in governance frameworks is crucial for defence organisations such as the U.S. Department of Defence and the Canadian Department of Defence. While adhering to various guidelines and standards, such as those from NATO and ISO, these organisations should treat these frameworks as tools that promote agility rather than inflexible rules.

By selectively adopting necessary components and allowing for exceptions when needed, they can avoid dogmatism and better demonstrate their value over time. Ultimately, this adaptive approach enhances the effectiveness of defence governance in a rapidly changing environment.

An effective agile mechanism is essential for successfully managing AI initiatives, allowing teams to seek expertise when needed and promoting flexibility between governance and operational efforts. Not all AI projects require the same guidelines; some may demand stringent measures, while others can be more lenient.

It’s crucial to establish essential quality checks, Data Management practices, and suitable AI techniques for each initiative. For instance, when exploring anomaly detection using one-class SVM, it’s important to clearly explain to the business the rationale behind classifying a claim as an anomaly and differentiate between various fraud detection techniques. Providing clear guidelines will help teams understand their tasks and make informed decisions.

Prodago’s framework, which is currently under development, includes comprehensive guidelines that may not be necessary immediately, but are designed to prevent oversights that could lead to adverse outcomes. Mario categorises these guidelines into stages—crawl, walk, and run—helping teams start with manageable tasks and gradually progress to more complex implementations.

The Role of AI in Data Management and Governance

The ongoing challenges in data projects highlight the persistent issue of disorganised data, even in the face of technological advancements over the past few decades. An attendee recalled their early experiences with data infrastructure and notes that similar problems continue to hinder progress today.

Mario underscored that the primary obstacle is not just the technology itself, but rather the need for the right motivation to address these challenges; without genuine drive, progress may remain stagnant. Ultimately, addressing the complexity of data requires both innovative tools and a dedicated effort to drive change.

The management of Data Governance poses significant challenges, often leading to organisational chaos. However, leveraging AI presents a compelling opportunity to drive business value and motivate action towards resolving these issues. By aligning specific Data Management transformations with defined business objectives and AI use cases, organisations can prioritise efforts effectively.

This approach enables streamlined Data Management capabilities and direct connection to strategic goals, thereby fostering a more efficient transformation process. An example of this can be drawn from my experience working at a central bank, where such strategies were implemented to enhance Data Management practices.

The CEO decided to halt various data projects, focusing instead on funding only AI initiatives. However, the rationale behind this choice remains unclear and may not have been the best decision. There’s a growing sentiment among people who are weary of being compelled to manage data solely for regulatory and privacy reasons. Instead, they are now motivated by the potential of AI to transform businesses, viewing Data Management as a positive opportunity rather than a compliance chore. This shift in perspective highlights a universal trend where organisations are increasingly embracing AI as a critical element of their operational strategy.

The process of anomaly detection involves identifying unusual claims and requires human expertise to explain the reasons behind these anomalies to customers. While AI can effectively highlight anomalies, collaboration with data stewards or subject matter experts is essential to provide comprehensive answers.

An illustrative example is Immigration Canada’s initiative to automate visa decision requests, where they used a decade’s worth of historical data to train a model. This model demonstrated greater accuracy in decision-making than human officers, showcasing the potential of AI to enhance operational efficiency.

A recent case highlighted the importance of integrating governance requirements early in the decision-making process for deploying AI technology. Initially, a significant investment was made in a technology that ultimately failed to provide the necessary explanations for decisions, resulting in a year of wasted resources and a loss of several million dollars when the decision was subjected to judicial scrutiny. This situation underscores the crucial need to identify and assess explanatory requirements upfront, ensuring that selected AI solutions can meet these governance standards and avoid costly setbacks.

Figure 27 Claim Anomaly Detection

Data Anomaly Detection and Interpretation

In his analysis of anomaly detection, Howard utilised advanced techniques such as one-class SVM (Support Vector Machine) and Isolation Forest, which revealed unexpected anomalies within the data. Despite the initial success in identifying these anomalies, he faced difficulties interpreting the results due to the lack of data normalisation, preventing him from clarifying why certain data points were flagged while others were not.

This experience underscored Howard’s increasing dependence on statistical methods to uncover potential issues, like fraud, while also emphasising the necessity of understanding the underlying metrics, such as score and distance, that determine anomaly thresholds. Ultimately, this process highlighted both the potential and challenges of using statistical analysis in data interpretation.

Understanding the scoring system is crucial for accurately interpreting results and effectively communicating findings to others. While the scoring routines can provide valuable insights, relying on them without comprehending their underlying logic can lead to misunderstandings and potential issues. Expanding the scoring framework to identify specific conditions that trigger anomalies could streamline the analysis process, eliminating the need for manual investigation and enhancing overall clarity and efficiency in data interpretation.

The Role and Impact of AI in Decision Making

The integration of generative AI in organisations presents a transformative opportunity to enhance internal productivity and efficiency. While AI is frequently leveraged for customer interactions, its real value lies in supporting decision-making processes rather than replacing human judgment.

By detecting anomalies and analysing their underlying causes, organisations can foster deeper inquiries into their data, ultimately leading to more informed and strategic decisions. This strategic application of AI not only empowers users but also drives overall organisational effectiveness.

To effectively assist insurance agents in addressing anomalies detected in claims, it’s essential to provide clear explanations to customers regarding the issues identified. Leadership must recognise that while AI is often viewed as a comprehensive solution, it serves merely as another tool that must be used appropriately.

Misusing tools can lead to complications, much like attempting to hammer a nail with a wrench can damage both the tool and the nail. Ultimately, decision-making should reside with humans, as AI should never take on this responsibility; it ought to be utilised correctly and for the right reasons to enhance the claims process.

AI has significantly streamlined data analysis by compiling results and assessments, yet human judgment remains crucial due to the lack of context. While AI enhances individual productivity, it does not inherently improve business outcomes. Instead, AI serves as a tool to help humans make informed decisions more efficiently by clarifying data anomalies and facilitating a deeper understanding of the data.

This reflects a broader trend in data science where professionals traditionally spent about 80% of their time determining the right data and only 20% on analysis. With improved processes, governance, and data quality, data scientists can now access accurate and categorised data, allowing them to focus more on their core work and reduce inefficiencies.

Mario then highlighted the critical role of collaboration in driving successful projects, particularly through the careful selection of relevant frameworks and tools. He stressed the need to focus on a specific subset of tools that align with project goals and emphasised storytelling and automation as essential mechanisms for achieving outcomes. By addressing broader factors beyond technology, Mario underscored the importance of a holistic approach to facilitate effective AI adoption and minimise the risk of project failure.

Figure 28 Key Findings