Large Language Models' Potential for Grants Management: Opportunities and Challenges

The Large Language Models (LLMs) that power generative AI tools like OpenAI’s ChatGPT are rapidly emerging as a transformative technology with the potential to revolutionize many industries, including grants management. While LLMs offer a wide range of applications across various domains, their true potential shines when applied to an organization’s unstructured data. 

POTENTIAL APPLICATIONS

In the context of grants management, this translates into a wealth of possibilities. Here are a few examples:  

• Answering questions about grant manuals and policies: LLMs can serve as virtual assistants—providing accurate and context-aware responses to inquiries from grantees, subrecipients, and internal staff—ensuring consistent interpretation and understanding of complex grant requirements. 
• Summarizing Notices of Funding Opportunities (NOFOs): LLMs can quickly distill the essence of lengthy NOFOs, allowing grant professionals to quickly assess eligibility criteria, application requirements, and funding priorities, thereby streamlining the proposal development process. 
• Analyzing awarded vs. non-awarded applications: By ingesting historical application data, LLMs could be used to identify key success factors and patterns that differentiate successful proposals, thus providing valuable insights to improve future submissions. 
• Tracking program success and risk through subrecipient reports: LLMs can continuously monitor subrecipient reports—flagging potential risks, identifying areas of concern, and highlighting successful initiatives—enabling proactive risk management and program optimization. 

CHALLENGES

While the potential of LLMs in grants management is vast, there are several challenges that must be addressed. Data security and control are paramount concerns, especially when leveraging off-the-shelf LLM solutions that involve uploading potentially sensitive data to third-party servers. Model reliability and the potential for updates to break existing functionality are also issues that need to be carefully managed. The “hallucination problem” inherent to LLMs, where they can generate plausible-sounding but factually incorrect outputs, necessitates robust validation and quality control measures to ensure the accuracy and integrity of LLM-generated insights. 

One of the most significant challenges is perhaps structuring your unstructured data. Documents need to be organized and turned into a format that a machine can easily understand. Organizations using a document management system or grants management software that allows for the addition of custom metadata or tagging of documents and text segments have a head start with this. 

GETTING STARTED

For organizations eager to harness the potential of LLMs in their grant programs but uncertain where to begin, an excellent starting point is exploring the features and limitations of readily available off-the-shelf solutions. This initial hands-on experience can provide valuable insights into the capabilities and nuances of LLMs. Concurrently, organizations should invest time in learning about the various components of the generative AI workflow and how model tuning can be used to ground responses to a source of truth, thereby minimizing inaccuracies and biases. Crucially, conducting comprehensive data audits and readiness assessments is a critical prerequisite to ensure that the data fueling AI initiatives is accurate, reliable, and poised to drive meaningful outcomes aligned with your organization’s goals and priorities for grants management. 

CONCLUSION

As the capabilities of LLMs continue to evolve, their potential applications in the grants management domain are vast and exciting. By harnessing the power of these cutting-edge AI technologies, organizations can streamline processes, gain invaluable insights, and drive more effective and impactful grant programs. However, realizing this transformative potential requires a thoughtful and strategic approach that addresses key challenges around data security, model reliability, and factual accuracy. Future blog posts will delve deeper into these topics. Please feel free to reach out if you’re interested in exploring solutions tailored to your organization’s needs. 

Authors