Our new AI assessment of engagement sentiment

The Global Equity ESG has long integrated EOS at Federated Hermes Limited, our dedicated stewardship and engagement team, into our investment process. EOS provides both qualitative insights from thousands of company meetings and quantitative metrics that enhance ESG analysis and decision-making.

These insights feed directly into our proprietary QESG Score, which blends data from voting outcomes, milestone tracking, and engagement conversations to assess companies’ progress on material ESG risks – those that could impact long-term shareholder value.

Our focus on ESG improvers – companies demonstrating strong or improving ESG standards – has been central to our philosophy since inception. EOS’s forward-looking intelligence complements third-party data from providers like MSCI, Trucost, and Sustainalytics, and is used not only in stock selection but also in ongoing company reviews and investment validation.

The history of EOS

EOS at Federated Hermes Limited (EOS) is a world-leading stewardship service provider.

Founded in 2004 on a legacy dating back to 1983, EOS advises on more than US$2.2tn in assets to deliver corporate engagement and proxy voting services.

EOS at a glance:

>45 people with diverse backgrounds, nationalities and sector specialisms
994 companies engaged in 2024
14,701 voting recommendations for meetings made in 2024
87 discussions held with regulators and stakeholders in 2024

Source: Federated Hermes, as at 30 June 2025.

AI-powered engagement analysis

In 2024, the team made a breakthrough by applying large language models (LLMs) to over 20,000 EOS engagement transcripts. This innovation enables:

Sentiment extraction across 16 ESG themes
Unsupervised machine learning for time-series analysis
Clustering of key engagement topics
Quantification of sentiment shifts over time

These insights are now embedded into the subjective layer of the investment process, offering a more nuanced view of company trajectories. For instance, an industrial holding previously flagged for weak climate disclosures showed improving sentiment in EOS engagements – which was later validated by market recognition.

What are Large Language Models (LLMs)?

A large language model is a type of AI program trained on huge data sets – hence the name – to understand, predict and generate human language. LLMs form a key part of the ongoing ‘AI boom’, with the largest and most capable LLMs being generative pretrained transformers (GPTs) which are used in popular generative AI chatbots such as ChatGPT.

In 2024, the team made a breakthrough by applying large language models (LLMs) to over 20,000 EOS engagement transcripts.

These findings are visualised in the ESG Dashboard, a proprietary tool first launched in 2010 and continuously refined over time.

EOS’s evolution – from advising a handful of clients to representing over US$2.2tn in assets under advice – has amplified its influence, enabling deeper engagements and more impactful stewardship.

What is the ESG Dashboard?

Our proprietary ESG Dashboard provides a concise overview of a company’s ESG profile across a variety of metrics, compared to peers. It helps direct the team towards areas that may need further investigation, such as controversies, product involvement or differences in ratings provider assessments, as part of our due diligence.

We can dynamically select the peer group, going from a 10,000ft view of the universe, down to the sub-industry and country level to understand the company’s overall position and how it compares to its closest peers.

The Dashboard also provides an overview of historic and current EOS engagement activities. It provides information on where progress has been made, the duration of the engagement and the associated meeting notes, helping direct our conversations with the EOS engagers in our regular meetings.

Benchmarking ESG performance: Sector-level insights

The analysis also allows us to benchmark companies more accurately against sector peers, enhancing our ability to identify leaders and laggards in ESG performance.

For example:

Governance sentiment in Financials has trended negatively over the past five years.
One portfolio holding, Visa, however, stands out with marked improvements in Governance sentiment.

Top engagement topics in Financials include:

Composition & structure
Board & management
Enterprise risk management
Long-term strategy
Pay design & transparency

These topics have all been discussed with Visa, where they have strong averages versus other financials.

Another portfolio holding example is Nvidia. Despite not seeing engagement objective progress recently, we can still use the LLM to determine that sentiment is positive across several issues based on our ongoing conversations with the company. Topics discussed with Nvidia since start of 2024 include:

Access & affordability
Basic shareholder rights
Board and management effectiveness
Business purpose
Climate opportunities
Composite & structure
DEI & innovation
GHG emissions
High geographic risk
Long-term strategy
Pay design & transparency
Succession & stability
Water stress

Expanding our coverage

The LLM model now covers companies previously lacking engager sentiment ratings – expanding the team’s coverage by approximately 50%. This enhancement allows for:

Sentiment scoring across ESG themes, even without direct qualitative ratings
Benchmarking against existing engager sentiment data
Bias detection in ESG rankings

The model occasionally flags negative governance sentiment where human engagers rate the engagement positively. This divergence is visualised in the figure below, plotting the LLM governance score vs. engager sentiment, highlighting the model’s ability to uncover hidden risks.

Conclusion

Federated Hermes continues to lead in ESG innovation, using AI not just to inform but to drive alpha. We plan to continue making use of this innovation as part of our stock selection and portfolio monitoring process. Additionally, we will continue to interrogate the data and test our investment thesis over time as part of our ongoing alpha model improvement. In the current environment, sentiment around ESG topics will be crucial in identifying long-term improvers from an ESG perspective.