embedded mixed-methods approach

The embedded mixed-methods approach is an advanced research design originating from social sciences, combining the strengths of quantitative and qualitative methods. Its core definition involves embedding a secondary, different type of data collection and analysis phase (e.g., small-scale qualitative interviews) within a primary research framework (e.g., a large-scale quantitative survey). The goal is not to give equal weight to both methods, but to use the secondary data to enhance, explain, or expand upon the primary study's findings. In AI risk management, this approach is crucial for implementing the 'Measure' function of the NIST AI Risk Management Framework (RMF), which emphasizes assessing AI impacts in real-world contexts. This cannot be achieved by quantitative metrics alone. For instance, when evaluating AI fairness, beyond calculating metrics like accuracy, one can embed in-depth interviews with affected users to understand the qualitative impact of algorithmic decisions on their lives, providing richer, contextualized risk insights than either method alone and supporting compliance with ISO/IEC 42001's monitoring requirements.

In enterprise AI risk management, the embedded mixed-methods approach effectively deepens the understanding of algorithmic bias, fairness, and societal impact. Practical implementation steps include: 1. Establish the primary assessment framework, e.g., quantitatively testing an AI recruitment system's resume screening pass rates across demographic groups. 2. Design and embed the secondary study, e.g., conducting semi-structured interviews with a subset of applicants from groups with significant pass-rate disparities to explore contextual factors. 3. Integrate data for risk response, using qualitative insights to explain the 'why' behind the quantitative 'what'. For example, a multinational bank found its loan model had high rejection rates for a specific immigrant community (quantitative). Embedded interviews revealed this group lacked traditional credit histories due to cultural preferences (qualitative). This insight led to feature engineering adjustments, reducing the risk misjudgment rate for this group by 15% and improving audit evidence.

Taiwan enterprises face three main challenges. First, a 'talent skill gap': data science teams excel at quantitative analysis but often lack training in qualitative methods like ethnography or interviewing. Second, 'resource and time costs': qualitative research is more labor-intensive than automated quantitative tests, posing a challenge in fast-paced tech environments. Third, 'methodological integration difficulty': rigorously integrating disparate data types to derive actionable insights requires mature methodologies. To overcome these, enterprises should: 1. Form cross-functional AI ethics teams (data scientists, UX researchers, legal experts) and provide targeted training. 2. Adopt a risk-based approach, piloting the method on high-risk AI systems to demonstrate value and secure management buy-in. 3. Implement standardized operating procedures (SOPs) and software tools to ensure consistent quality in data integration and analysis, gradually scaling successful practices.

Winners Consulting specializes in embedded mixed-methods approach for Taiwan enterprises, delivering compliant management systems within 90 days. Free consultation: https://winners.com.tw/contact