We are excited to feature a post by Dr. Larry Lowe with RippleWorx in our AI for HR series. We’ve been fortunate to work alongside RippleWorx with mutual clients, and Larry and I were classmates in Leadership Greater Huntsville’s Flagship Program. Larry is wicked smart, but better than that, he is a really great guy!
We trust Larry’s in-depth insights on AI for HR and how they (and you) can utilize it to your advantage to understand your workforce’s needs and impact organizational culture. Enjoy!
Guest Blogger: Dr. Larry Lowe, Chief Scientist at RippleWorx (larry.lowe@rippleworx.com)
Major Changes Are Coming to Your Organization
When your organization faces significant changes, a common first step is to send out a survey to understand your workforce’s views on specific topics. Your survey will likely include Likert scale questions, Net Promoter Score (NPS) questions, and some open-text questions. While Likert and NPS questions are straightforward to analyze, open-text questions pose a unique challenge. These responses can be messy in terms of length, sentiment, context, content, format, spelling, and even include emojis and text speak (SMH). Despite this messiness, open-text questions often provide the most context and insight. Distilling them into common subject categories is difficult and time-consuming. It is mentally draining to read and categorize thousands of responses, and keeping biases from influencing our decisions is challenging. If only there was a tool to help create structured insights from unstructured data…
RippleWorx has cracked the code to actionizing real data insights to drive meaningful change in organizations. With years of experience analyzing customer feedback, RippleWorx has developed the right AI models to drive continual organizational performance improvements.
The Power of Generative AI
If the problem of analyzing a large amount of employee feedback data sounds familiar, good news! One of the greatest benefits of Generative AI in the workplace is its ability to create structured insights from unstructured data. Let’s clarify some terms.
Structured Data: These are items that fit neatly into rows and columns, like a well-organized Excel spreadsheet where the columns contain consistently formatted data. With structured data, it is straightforward to calculate averages, count categories, or identify outliers. The structure naturally leads to clear insights.
Unstructured Data: These are items that do not have a predefined format or structure, such as the varied responses to open-ended survey questions. The lack of structure makes deriving insights extremely challenging and sometimes misleading.
The key to analyzing the open-ended feedback questions from your employees’ surveys is to generate structured, actionable insights from highly unstructured data. Different analytic approaches can be applied, but there are trade-offs. Let’s explore a few.
Traditional Methods to Analyze Open Text Responses
Traditional methods for analyzing open text responses include:
· Manual Coding: Reading each response and categorizing it into predefined themes or codes.
· Content Analysis: Reading the entire corpus to determine patterns, themes, and meanings.
· Statistical Text Analysis: Counting word frequency or creating word clouds.
While statistical text analysis is expedient, it often lacks understanding and semantic meaning across all responses. Manual coding and content analysis are both complex and time-consuming endeavors. When the unstructured data set is large, the human brain cannot equally consider all expressed thoughts. We often get tired and start “seeing” our biases in the data.
A New Method: Generative AI
By now, I hope everyone has experimented with the latest chat completion models, such as GPT-4, Claude 3.5, and Gemini 1.5. These models excel at summarizing large corpora of text into easily interpreted bullet points or narrative paragraphs. If the open text responses are saved as a PDF, follow these steps for effective summary insights:
1. Attach the PDF in the prompt window.
2. Write the following prompt into the chat window:
“You are a helpful HR assistant. I have attached a document that includes open text survey questions along with all the responses aggregated across the entire organization. I need you to summarize the top three most mentioned themes in the open text responses. The summary output format should be bullet points, each less than 200 words.”
Two key benefits arise from this approach:
1. Semantic Interpretation: The models semantically interpret all open text responses simultaneously, resolving the “messiness” of varied responses. This addresses human fatigue associated with processing large amounts of information, as the language model interprets every response equally and almost instantaneously.
2. Coherent Output: The model connects extracted themes from the responses and generates a coherent summary following the provided instructions.
These models’ ability to identify threads and concepts from numerous responses is remarkable. Adjusting your prompt can extract additional information from the PDF. For example, you can ask the model to summarize the top “positive” and “negative” themes mentioned or to develop an action plan addressing the top issues in the responses.
While these models significantly improve and expedite the summarization of open text questions, there are important considerations. Uploading corporate information into a public chat completion model poses risks. Sensitive topics discussed may not be intended for public disclosure. This data could be used to train future models, or your prompts and attached data could potentially be hacked and published later. Ensuring data security should be paramount when using Generative AI in your workflows.
An Even Better Method: Generative AI Mapped into a Performance Taxonomy
For even greater insight, integrating an organizational performance taxonomy into the prompt allows the model to categorize responses into different dimensions of the organization before summarizing actionable insights. This approach provides more precise results by highlighting not just the overall organizational strengths and weaknesses but pinpointing strengths and weaknesses to specific areas within the organization.
RippleWorx has created a model for organizational performance called the Performance Chain. In the Performance Chain, an individual addresses a role, roles combine to form teams, and teams combine to form the organization. A performance taxonomy accompanies each link in the chain. The taxonomy for the individual includes motivation and well-being concepts. The taxonomy for the role covers hard and soft skill proficiency and employee readiness. The taxonomy for the team covers collaboration and tactical task execution. The taxonomy at the organizational level covers strategic leadership, culture and climate setting, and key performance metrics.
Embedding the performance taxonomy within the prompt flow results in more precise insights within the organization. For instance:
General Prompt Response: “Communication is an issue in the organization.”
Performance Taxonomy Prompt Response: “Multiple middle managers are having trouble communicating action plans with their teams.”
The general prompt provides a broad level of actionable insight, but the prompt with the performance taxonomy offers deeper insights, such as the need for targeted training for middle managers. The primary goal of assessing actionable insights is to implement targeted interventions that increase organizational performance.
The Wrap-Up
Organizing and analyzing open text survey responses is just one example of how RippleWorx is utilizing Generative AI to transform organizational performance. The Performance Chain framework also integrates external surveys, performance evaluations, and key performance indicator data into our Generative AI prompt workflows. Including this information along with a performance framework provides an even greater level of resolution for actionable insights. The additional resolution aids leaders across the organization in creating targeted action plans that keep individuals motivated and increases organizational momentum.
www.rippleworx.com
