Thank you for your interest!

Thank you for signing up for our webinar: Sustainability Data Meets Agentic AI: Introducing Persefoni Analytics Agent

We hope you enjoy the on-demand recording!

← Back to webinar details

To view this video, please enable cookies or you can watch it directly on Vimeo.

1. As a carbon accounting software curious to know if any work has been done to measure the emissions from the LLM to give users an estimate of emissions during the use of this tool?
   
   1. AI-related emissions are becoming one of the fastest-growing areas of energy use for companies globally, but today there is still no reliable, accounting-grade way to estimate the emissions impact of individual prompts, tokens, or the use of a specific LLM-powered tool.
      The main challenge is data availability. AI providers like OpenAI and Anthropic generally do not disclose the level of information needed to calculate emissions with confidence, including model-specific energy use, infrastructure details, data center location, grid mix, utilization, and how workloads are allocated. Without that transparency, any estimate is likely to rely on broad assumptions.
      We do expect to see more organizations publish methodologies for estimating AI-related emissions, and those can be useful directional indicators. But at this stage, we would treat them as estimates rather than reliable measurement methods for carbon accounting.
      Persefoni is actively monitoring this space and developing our own approach as better datasets and methodologies become available. For now, our view is that AI emissions are important to track, but the market does not yet have enough provider-level data to measure them with the same rigor expected in carbon accounting.
      ‍
2. I’m still concerned about hallucinations. Can you speak more on what checks you’ve built in to prevent them, and to detect them if they pop up?
   
   1. The reality is that — at least at present — hallucinations aren’t 100% preventable with any LLM. However, we’ve done a lot of work to give the agent some domain knowledge to try to prevent misunderstandings and inaccuracies. We also have a robust testing suite to make sure that for common questions, the agent is getting the questions correct — and we plan to keep augmenting this testing data as people use the agent.
      ‍
3. Does the agent store and remember chat history between sessions?
   
   1. Yes, those chat histories are stored when you log into Persefoni!
      ‍
4. Is the LLM trained/versed in common reporting frameworks (e.g. TCFD) to be able to prompt against the rules?
   
   1. At present, we have reporting tools for pulling out key emissions and energy metrics from ledger data that are separate from the agent (e.g., for CDP, CSRD, SECR, etc.). We’re continuing to build new functionalities for the agent. Our roadmap for additional agent functionality is continuing to evolve — so we’re really interested in people’s feedback about what additional functionality would be helpful.
      ‍
5. Does this agent work with forecasting emissions or reduction scenarios?
   
   1. The Analytics Agent can perform forecasting and emissions reduction scenarios through data that you provide. We’re continuing to build out functionalities for the agent around different customer use cases based on feedback we receive.
      ‍
6. Which LLM is the backend for your agent?
   
   1. The Persefoni Analytics Agent uses Snowflake Cortex to query against the ledger data that sits within Snowflake. The underlying LLM being used us typically Opus 4.6; however, we are experimenting with different models and will continue to do so, to see if we can achieve better accuracy and performance with other models.
      ‍
7. How does the system treat the Emissions consolidation of an American company based in California, that has subsidiaries in Europe?
   
   1. The Analytics Agent will query information from your CO2e Activity Ledger. If you have built out your organizational structure, or catalogs, within Persefoni to align to this use case, you are able to prompt the agent to pull that specific information together.
      ‍
8. In terms of audit trail, can't the agent reference a certain location in the ledger to show traceability?
   
   1. Yes, the Analytics Agent is excellent at helping with audit trails. You can ask the agent to provide you with all of the calculation inputs associated with specific transactions in the ledger, as well as ask questions to try to find a specific transaction based on other information.