Wisdom from words: insights from qualitative performance feedback

Every year, your firm collects thousands of words in performance reviews and feedback. Often these words have been agonised over, rewritten and debated. So much labour - financial and emotional - have gone into producing these words.

‍

How much of that rich, qualitative data is actually being used to its full potential? Hopefully it’s led to a great review conversation and encouraged deep reflection, but beyond that most of it would sit untouched, because analysing it is time-consuming, overwhelming, and expensive.

‍

Enter Natural Language Processing (NLP)

Natural Language Processing (or NLP for short) was born in the 1940s to solve the problem of language translation at scale. It’s since grown into many other applications - from smart email filters that detect malicious emails to banking tools that flag fraudulent transactions. We already trust NLP to guide attention and support decisions.

‍

In brief, NLP involves using computers to 'read' text and

identify and classify key words (topics, issues, people etc)
understand the emotions, opinions, and attitudes behind the words

‍

Large Language Models (the engine behind AI tools like ChatGPT, CoPilot, Gemini, Claude, DeepSeek) perform Natural Language Processing. These computer models are trained on billions of examples of text and can recognise words and patterns in language that indicate meaning and sentiment. For example, they can detect that "The workload is over overwhelming" expresses negative sentiment about the topic of "workload."

‍

It’s like having a team of assistants who can read through all your feedback, highlight the key takeaways, and tell you what’s really going on beneath the surface - except they do it in seconds and they are far more consistent than a team of human assistants (human judgement is inconsistent and noisy).

‍

How can it help you?

Please conduct topic and sentiment analysis on the sentences below. Then provide an overall score for each sentence, and suggest how HR can help this person. Put your output in a table.

‍

I used the above prompt in Microsoft CoPilot to analyse the three sample sentences in the the first column in the table below. The idea is to test how AI might understand qualitative comments that might be expressed at your firm. You should try it with your own examples.

‍

Sample Sentence	Topic/Sentiment	Overall Score	HR Suggestions
"Client work has been challenging and time consuming"	Client Work, Negative	2/10	Provide additional support, manage workload, and offer time management training.
"I’ve enjoyed the flexibility of working across timezones, even if it’s been intense at times"	Flexibility, Work Intensity, Mixed	7/10	Encourage work-life balance, offer flexible scheduling, and recognize efforts.
"Working in the healthcare sector has been mind-blowing. I have so much to learn"	Healthcare, Learning, Positive	9/10	Provide learning opportunities, mentorship programs, and acknowledge enthusiasm.

‍

By turning feedback, appraisals and objectives into actionable insights, topic and sentiment analysis helps you make smarter decisions about talent development, team structure, and organisational culture. For example:

In high-pressure environments, understanding the sentiment behind feedback can help identify burnout risks, improve employee satisfaction, and retain top talent.
Where precision and compliance are critical, sentiment analysis in feedback and reviews can uncover hidden frustrations or misunderstandings that might impact team dynamics or client relationships.
Where collaboration and communication are key, text analysis can reveal patterns in objective setting that point to opportunities for better alignment and performance.

‍

The potential is clear. The challenge is how to do this across large amounts of data (without losing context) in a quick and easy way.

‍

At Performance Leader, we're integrating AI to increase high quality feedback, support meaningful conversations and drive insights from data. The following approaches have show great promise in internal research and client testing.

‍

Weighted Net Sentiment Index

The Weighted Net Sentiment Index is a single metric that measures the overall sentiment in text data like feedback or reviews. It's calculated as the difference between positive and negative sentiment percentages, with a weight applied to the negative scores.

‍

Weighted Net Sentiment Index = (Positive - 2x Negative) / (Positive + 2 x Negative + Neutral)

We tested multiple approaches in our research and like this one because it

provided a simple metric that had an unambiguous meaning
emphasised negative scores, which surfaces risk

‍

Below are some examples of what the score would generate and what it could mean - the shaded columns capture the index score and what it could suggest based on the positive, neutral and average sentiment scores.

‍

Outcomes/Understanding	Positive Sentiment average score	Negative Sentiment average score	Neutral Sentiment average score	WNSI (-ve)
All positive	100	0	0	1.0
Mostly positive, some negative	75	5	20	0.6
Small positive and negative, mostly neutral	10	10	80	-0.1
Equal positive, negative, neutral	30	30	40	-0.2
Very divergent (mostly positive and negative)	45	45	10	-0.3
Mostly negative, some positive	20	60	20	-0.6
All negative	0	100	0	-1.0

‍

The charts below show how we display this sentiment index over your data and highlight the difference between sentiment and WNSI. In our research, admins found the chart on the right more useful because

they could quickly see where the overall sentiment stood and how it varied. The variability was an important insight (especially if used to compared different cohorts)

it took up less (physical and cognitive) space - it told more with less

‍

The Weighted Net Sentiment Index can also be used to explore sentiment across topics across different roles, departments or projects. The chart below extract skills and organisation names, and allows admins to apply that view over different departments or roles, so a leader can answer questions like

What are associates in corporate focused on this month, and is trending positive or negative?
What areas are my team receiving feedback on (or not) - which I can raise in my monthly check-in?
Has the launch of the new LX solution changed the conversation about skills?

‍

Key Topic - Descriptor Pairs

Key Topic - Descriptor Pairs are a way to analyse and summarise text data by identifying the main topics (key topics) and the specific words or phrases (descriptors) that describe or relate to those topics. It consists of a broad subject or theme (like "workload," "communication," or "career growth") and the specific words that describe it (such as "too high," "unclear expectations," or "lack of opportunities").

‍

Example feedback: "The workload is too high, and communication from leadership is unclear"

Key Topic: Workload → Descriptor: too high
Key Topic: Communication → Descriptor: unclear

‍

These are far more powerful than simple word clouds, because they add valuable context and thus insight into why certain topics are being discussed - for example, moving beyond simply noting "workload" to understanding that "workload is too high."

‍

Below is a preview in Performance Leader.

‍

Key topic - descriptor pairs are valuable tools in both performance reviews and client feedback contexts. In performance evaluations, they effectively identify strengths and areas needing improvement, such as noting "team collaboration" as "excellent during projects." Similarly, when analysing client feedback, they help pinpoint specific satisfaction drivers and areas of concern, like identifying "response time" as an aspect that "needs improvement."

‍

Key takeaways

Natural Language Processing and text analysis tools represent a significant leap forward in how organisations can utilise their qualitative data. By implementing these AI-powered solutions, firms can quickly extract actionable insights from thousands of words of feedback and drive meaningful improvements in performance management and organisational development.

‍

The combination of Weighted Net Sentiment Index and Key Topic-Descriptor Pairs provides a comprehensive framework for understanding not just what people are saying, but how they feel about it. This deeper level of analysis enables leaders to:

Make data-driven decisions about talent development and team structure
Identify emerging trends and potential issues before they become problems
Track the impact of organisational changes and initiatives over time

‍

It's never been easier to experiment with Natural Language Processing and sentiment analysis. Run text from different contexts in your favourite AI model and see what you can learn. Which types of comments stand out? What is not obvious, and how can you dive deeper?

‍

As we continue to develop and refine these tools at Performance Leader, our goal remains clear: to support more frequent, easier and better feedback interactions between people, and harness the full potential of your feedback data to create more effective, engaged, and successful teams.

‍