Anand Subramaniam, Chief Solutions Officer at KANINI, drives digital and data transformation journeys for our clients’ business success.
With the rise in popularity of ChatGPT, the business world has been examining new possibilities of generative AI (GenAI). One area that shows a lot of potential for GenAI is in the audit landscape, particularly in streamlining the intricacies of environmental, social and governance (ESG) audits.
An ESG audit is a complex, document-intensive process. GenAI holds the potential to address the central challenge faced by auditors: efficiently managing and analyzing the substantial volume of documents encountered every day.
In this article, I’ll share some of the ways that GenAI can streamline the ESG audit process as well as some of the challenges that ESG teams may face when using it for this end.
How GenAI Can Help With ESG Audits
Addressing the challenges of ESG audits requires auditors to hone specialized audit skills, collaborate with relevant stakeholders, foster transparency and accountability and advocate for the development of robust ESG reporting frameworks.
In this journey, generative AI can improve auditor’s efficiency and accelerate the audit process. For example, it can provide:
• Agility In Research And Planning Of ESG Audits: Generative AI can help auditors automate the complex information-mining processes of the research and planning stage of an ESG audit. It can automatically identify scoping entities such as GRI/SASB ESG Standards, Topics, Disclosures and Requirements by contextually comparing them with ESG-mandated standard documents.
• Productive Client Interactions And On-site Audit Actions: AI-powered document intelligence can automate data validation and document review processes, and auditors can extract valuable information embedded in images using the OCR capabilities of many AI solutions. GenAI can also offer support with their client interactions by transcribing meetings, summarizing key action points and generating plans for follow-up meetings.
• Streamlined End-To-End Data Collection And Analysis: GenAI can enable auditors to interrogate documents for specific insights for correct, consistent and complete ESG data.
• Enhanced Efficiency In Stakeholder Management Processes: GenAI can identify and map the right stakeholders from whom to collect the right information, gather a contextual understanding of stakeholder discussions and other data to draw comparisons for deeper analysis, and record and analyze stakeholder sentiments.
• Improved Audit Assessment Quality And Intelligent Recommendations: Generative AI’s predictive analytics capabilities can anticipate the relevant materiality decisions, highlight current and future audit risks based on trends and automatically generate effective recommendations for clients.
• Proactive Risk Management And Compliance Assurance: Generative AI can help auditors identify qualitative and quantitative risks more easily from documents such as CSR reports and 10-K reports. This is otherwise quite a challenge with risk charts for different organizations being diverse, based on their industry type and region.
• Comprehensive ESG Reporting and Effective Stakeholder Communication: Generative AI, with its ability to ingest and process large volumes of documents in minutes, can easily suggest the reporting scope and accelerate materiality decision-making.
Challenges Of Using Generative AI And LLMs In ESG Audits
While GenAI undoubtedly presents numerous exciting opportunities in the field of ESG audits, it does come with its own intricacies and limitations that must be navigated by auditors with caution, awareness and a commitment to ethical and responsible use. Some of those common challenges are:
• Hallucination Risk: GenAI is susceptible to hallucinations if the prompts and hyperparameters are not correctly defined. Therefore, auditors must know how to craft precise prompts to get accurate responses. The hallucination risk can be reduced by model monitoring and enabling a pipeline that can quickly allow users to change the prompting.
• Implementation Issues: The implementation of generative AI, particularly LLMs in the ESG audit context, can be quite complex. It requires a combination of technical expertise and careful planning in implementing processes such as:
1. Setting up automated pipelines for document cracking and vectorization of extracted data using different tools like Vector Database and LangChain Framework.
2. Tuning hyperparameters (e.g., temperature, the maximum number of tokens, Top-p and Top-k, and so on) for changing document corpora.
• Regulatory Concerns: While LLMs offered by leading cloud providers such as OpenAI, AWS Bedrock and Google’s LLMs have well-defined guidelines on security and regulations, this may not be the case with other open LLMs. Therefore, auditors must approach open LLMs by paying heed to all security and regulatory considerations.
• Data Security: A deep understanding of the data being used, whether it is personally identifiable information (PII) or other sensitive information, underscores the responsible and ethical use of LLMs. Secure data handling practices protect against unauthorized access or data breaches. When dealing with PII, strategically creating prompts that explicitly call for anonymization or using role prompting to control certain outputs can help.
Building Robust And Secure Generative AI And LLM Solutions
With these challenges in mind, here are a few tips to succeed when implementing GenAI into the ESG audit process:
• Employing LLMs as managed services within cloud platforms such as Azure, AWS or GCP is one way of automatically securing the applications built on them.
• In cases where custom training of LLMs is required on-premises, opting for LLMs from OpenAI and Google is a good choice.
• Using simulation environments like OpenAI playground to constantly test prompts can improve the outcomes in production.
• LLMs do not really require extensive data. They can be pretrained on a variety of corpora by leveraging in-context learning methodologies, such as zero-shot, one-shot and few-shot learning.
• Given the limited parameters available for fine-tuning LLMs for specific purposes, mastering advanced prompting techniques, including role prompting, chain of thought and retrieval-augmented generation prompting, enables maximized utility and precision of LLMs in various applications.
Getting Ready For The Future Of ESG Auditing
The convergence of auditor expertise and generative AI can create new benchmarks in ESG audits. With solid risk management and governance practices in place to harness generative AI for innovation securely, it emerges as the optimal path for enhancing efficiencies, making more intelligent decisions and nurturing a resilient, data-driven audit ecosystem.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?