How to Use AI for Thematic Analysis of Qual Data: 6 techniques to improve accuracy and defensibility

If your team is using Microsoft Copilot or similar AI tools to analyse qualitative datasets, there is a reasonable chance your conclusions are harder to defend than you think. Not because AI tools are useless, they are not, but because they struggle with large datasets, but won’t tell you. 

This can lead to themes without evidence trails, outputs that change if you re-run them, and analysis you cannot reproduce when leadership questions it.  

In this practical session, Allen + Clarke's AI team will demo where generic AI tools fail for thematic analysis and give you a six-step method for doing it properly with the tools you have access to and are already using. We will also be honest about the limitation of this method, and what the answer looks like when your dataset is too large or too high-stakes for off-the-shelf tools to handle. 

What you'll learn: 

• Why Copilot and ChatGPT fall short: Five failure modes with demonstrations 
• The importance of your codebook: The analytical foundation that makes everything else work 
• The six-step method: Practical techniques you can apply to your next dataset immediately 
• Where the limits are: An honest account of what off-the-shelf tools can and cannot deliver 
• When to escalate: How to know when your dataset needs more than this technique
   

Perfect for: Government agencies, local councils, and regulated organisations handling large volumes of consultation submissions, stakeholder surveys, or open-text feedback — where findings face intense scrutiny by leaders or the public.