When Data Doesn't Come in Rows
1. When Data Doesn't Come in Rows
Welcome back! Previously, we used AI to analyze structured data like sales tables. However, in many consulting projects,2. Unstructured Data
some of the most valuable insights are hidden in unstructured data, such as customer feedback, meeting transcripts, or long-form reports.3. Example: Customer Experience Initiative
Imagine you're consulting on a customer-experience improvement initiative. Your client has conducted more than 200 customer interviews across different segments and regions. Each interview has been transcribed into text, but the client doesn't have the capacity to review them all. They've asked you to analyze the transcripts and draft an action plan to improve customer satisfaction.4. Example: Customer Experience Initiative
We begin by taking the dataset that the client compiled during the research phase, and reviewing a few sample entries. As with structured data, a key principle is to break the analysis into small, manageable steps. In each iteration, we should give the model a clear objective, context about the data, and guidance on the expected style.5. Example: Customer Experience Initiative
In this first step, we open a new chat with Copilot and upload our dataset. The most important part of the initial interaction is to provide rich context on the data and a clear goal. To get a better sense of the main recurring themes from the interviews, let's ask the model to generate a frequency table highlighting the key pain points mentioned by customers. Here's what we found! After some thinking, the model successfully provides a frequency table of the issues it discovered; things like wait times, usability, and reliability.6. Example: Customer Experience Initiative
Another step we can add to the analysis is classifying the overall sentiment of each interview as positive, negative, or mixed. Observing the model output, we realize that most interviews turn out to be negative or mixed. We should report this finding to the client and propose an action plan to increase the share of positive feedback.7. Example: Customer Experience Initiative
To do this, let's run a more detailed analysis of the Net Promoter Score to identify which themes are most common among detractors, and which appear more frequently among mixed respondents. This analysis reveals that support wait times are the most frequent complaint among detractors. Prompts like this allows us to combine structured data, like NPS scores, with insights from unstructured data, like interview transcripts, resulting in a more comprehensive analysis.8. Example: Customer Experience Initiative
We can ask the model to extract customer-experience initiatives directly from the data to address client dissatisfaction. It is important to instruct the model to stay grounded in the source material. This helps prevent hallucinations and allows us to trace each recommendation back to the underlying evidence. Here's what we found! The model effectively recommends reducing support wait times by improving call routing, fixing payment-flow reliability issues, and introducing a premium support offering, among other initiatives.9. Let's practice!
Time to put your skills in analyzing unstructured data to the test!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.