It also has great value in internal communication. Some intranet providers have integrated corresponding tools into their applications. Large companies, which hardly differentiate between internal and external communication, also include the voices of employees on LinkedIn & Co. in their analyses. However, sentiment analysis solutions have one weakness: they are not cheap and only produce informative results if there is a large amount of data.
Sentiment via ChatGPT
In our case, we were dealing with a medium-sized, internationally active company. A few thousand employees in Germany – enough to have reasonable internal communication and some interaction, but too few for the sensible use of one of the “big” solutions. For a dashboard on internal communication, we still wanted to track a house number on sentiment. There were too many comments for a manual analysis, so we decided to carry out the analysis with ChatGPT.
A few appropriate prompts and attempts later, we had an evaluation that was quite something: Listing of the categories with percentage share at the beginning including visualization via pie chart and then the respective categories with a bullet-point list of the comments – for the sake of clarity, only the first 5 letters in each case, but enough to be able to check the quality of the output (spoiler: the output was of very high quality).
In addition to a sentiment analysis, we now have the following:
- A prompt template for carrying out sentiment analyses in the future.
- Learned something again.
When we ask ChatGPT to complete a complex task for us, it is not uncommon for it to abort in between and stamp the entered data as incorrect. But with enough persistence, the AI will find the solution. And the good thing is that if you open a new chat from time to time, ChatGPT won’t be offended if you ask it the same question for the umpteenth time.
We are happy to share the prompt on request at hello@montua-partner.de