FIGURE 2 Sentiment Analysis of Transcript (excluding Q&A)
0.6 0.4 0.2 0.0 -0.2 -0.4
080 Sentence Number
20 40 60 100 120 140
to understand some of how these companies have governance around their ESG function, which could inform how we consider organizing our ESG function. “It’s important to note that these AI tools can ‘hallucinate’ and
make things up, so you have to check facts. But it is a good starting point and it was directionally specific. It was efficient, taking about 45 minutes compared to much more time I would have spent read- ing and analyzing each of these sustainability reports. “AI tools are not replacements; they are assistants. Tey create
tremendous efficiencies. You still need to do your homework. We are still the subject matter experts. But this can greatly accelerate your work.”
Conducting Sentiment Analysis Lampf also used ChatGPT to conduct a sentiment analysis of an earnings call transcript, using the following prompts: Prompt 1: Analyze the transcript and provide a sentiment analysis for everything but the question-and-answer section. Provide the report back in the form of a summary, bullet points, and conclu-
niri.org/ irupdate
sion and create a simple chart to give me a visual representation. Prompt 2: Act as an equity portfolio manager. What is the specific language that corresponds with the top five most negative scores? In Prompt 1, Lampf used the ChatGPT data analytics plug-in
that generated the chart shown in FIGURE 2. Ten in Prompt 2 he asked ChatGPT to assume a persona that generated some insights and alternate language options. “We all know stocks can move during earnings calls, so I want to anticipate that,” Lampf explains. “Tere are sentiment analysis tools out there including dashboards and heat maps. I’ve always found them interesting but am not always sure precisely what to do with them. Tey’re not proactive. “So if you upload a draft script into a secure AI environment,
it can provide more context and perspective and help you under- stand nuances. It helps us anticipate sentiment and consider what AI-recommended language we may want to use.” In the transcript analysis, Lampf also posed these two prompts
regarding the Q&A session to ChatGPT: Prompt 1: What are the negative questions that are asked?
IR UPDAT E ■ FA L L 2 0 2 3 1 1
Sentiment Score
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36
