Challenge

Our client, a public health research organisation, was conducting a large-scale study into vaccine hesitancy. This involved in-depth interviews with a diverse set of stakeholders, including healthcare workers, community leaders, and patients. The result was a rich but unwieldy body of qualitative data—hundreds of hours of recorded interviews—each containing valuable behavioural insights.

To extract meaningful findings, the research team needed to identify key quotations from transcripts, categorise them using a behavioural science framework, and structure the data for further analysis. Manual analysis at this scale would have been prohibitively slow and inconsistent. They needed a custom AI solution capable of producing accurate, repeatable outputs to support behavioural tagging, pattern recognition, and insight generation.

Approach

While off-the-shelf qualitative analysis tools exist, they often lack the flexibility and precision required for specialised behavioural coding. To address this, we designed and implemented a tailored AI orchestration system that enabled structured, semi-automated insight extraction at scale.

Key components of our approach included:

• Custom model prompting and orchestration: We iteratively refined few-shot prompt templates to align model outputs with the client’s behavioural framework. We broke down the task into smaller, model-friendly steps—such as identifying quotes, tagging them with behavioural themes, and generating summaries of grouped insights.
• Workflow automation and data structuring: We developed a backend pipeline to automatically process transcripts, extract and tag quotations, and store the outputs in a structured database. This created a single source of truth for filtered insight retrieval and cross-cutting analysis.
• Interactive insight generation: By structuring the data, we enabled follow-up querying and summarisation—allowing researchers to filter quotes by behaviour type, stakeholder group, or context and automatically generate summaries or trend analyses.

Results

The final solution was a robust, scalable AI-powered classification and insight pipeline. It enabled the client to:

• Process and analyse hundreds of hours of interview content in a fraction of the time traditional methods would require.
• Extract consistent, accurate behavioural insights across the entire dataset using a transparent, repeatable methodology.
• Quickly explore emerging themes and patterns, supporting high-quality reporting and evidence-backed recommendations.

By combining behavioural science expertise with advanced AI orchestration, we transformed a large, unstructured data set into a powerful, decision-ready research asset.

‍