Rethinking Tracking Studies: Unlocking the Power of AI with Semi-Structured Qualitative

Traditionally, tracking studies in pharma, such as the well-known ATU (Awareness, Trial, and Usage) surveys, have naturally relied on quantitative methodologies. While quant is seen as the ‘gold-standard’ for tracking it does not come without challenges. Sample sizes are often restricted due to the scarcity of specialized physicians and patients with specific conditions. This scarcity can make it challenging to determine when a change in the data is truly meaningful or merely a ‘bobble’ in the data. Further, the nature of quantitative surveys often fails to capture the nuanced “why” behind observed changes. Together these often mean delays in taking any action as a result.

At the same time, qualitative tracking can feel like nearly an oxymoron. You simply need numbers to track, right? How else do you avoid subjectivity and truly measure trends and changes. Well, we believe today’s AI fundamentally changes this calculation. We believe by tapping into the power of AI and adopting a different approach to qualitative data collection, we can achieve genuine ‘tracking’ of critical market changes to inform business action.

The key to unlocking the potential of AI to enable qualitative tracking begins with how we collect the data. While traditional qual follows a loose discussion guide, we envision qualitative tracking to use ‘semi-structured data’. By that we mean, asking the same questions in the same way each time and clearly linking answers to their corresponding questions. While we would leave room for additional probing and exploration as needed, this ‘semi-structured’ approach will be critical in creating the right data-set to unlock the full potential of AI-supported analysis.

With a semi-structured format, AI can more effectively identify patterns, trends, and anomalies across interviews and time periods. The consistent core questions provide a stable framework for comparison across time periods and groups, while the flexible elements allow for the capture of emerging themes and unexpected insights. This method amplifies AI’s scalability advantages. As the volume of data grows over time, the semi-structured format ensures that new data seamlessly integrates with historical information, allowing for ever-richer longitudinal analysis.

And while previous ‘text-analytics’ or ‘sentiment trackers’ were often underwhelming; the current AI models are a totally different ballgame. This is not counting key words or assigning sentiment tags to responses. Today’s AI models can effectively decode the context and meaning of responses, while simultaneously putting those responses in context of hundreds or thousands of other responses.

For example, consider the standard battery of association attributes one encounters in a traditional brand tracker or ATU. Rather than selecting a point on a scale, participants might be asked to select 3-5 that best describe the brand and explain why. We hear their actual thinking on the brand rather than guessing how the interpreted attributes, yet we’ve guided the participant to topics we need their reaction to.

From this semi-structure, AI can not only track trends in responses but also help us analyze themes in their explanation of why to uncover if there is a brand strength or weakness we might have otherwise missed.

While this type of qualitative tracking will not be right for every situation, we see several potential advantages: