Writing Proposals with AI: Navigating the Promise and Pitfalls

We’re all excited about getting rid of mundane and repetitive work. For those who ensure that their NGO has enough grant funding, this dream looks like “let the AI write the bid, and let me know when the donor is ready to sign”.

Unfortunately, we’re not quite there yet. But what exactly can AI tools like ChatGPT, Bing, or Google Bard achieve? The short answer is that they can do exactly what we tell them to do – nothing more, nothing less.

After several months of using generative AI tools in various applications, from EU and USAID to GIZ and foundation projects, we’re here to share our experience. Overall, we have used AI to draft over 50 small and several large grant and commercial contract applications, including project proposals to venture capitalists and private sector organisations. We here share a roadmap along the business development path from opportunity identification to contracting. Along the way, we provide insights and tips for successfully deploying generative AI in proposal writing.

1. Identifying the right opportunities with AI

  • AI can be utilized as a powerful search tool to browse online data and generate lists of partners, overviews of their projects, potential calls, and more.
  • AI can even generate a list of donors with calls for proposals for a particular location.
  • While this is all helpful, the output suffers from two problems:
    • Firstly, as AI tools are based on pattern recognition rather than a true understanding of the subject matter, they miss a lot of opportunities. We found AI tools useful in collating the first list of potential donors, especially when looking at a specific location. Still, our colleagues usually had to manually add a lot of donors afterwards. Still, some time was saved.
    • Secondly, since donors tend to write more about completed projects rather than forecasted ones, AI tools miss donors that have no (or little) experience in a particular location. ChatGPT basically looks back, not forward. But it is new donors who are often of particular interest. Still, AI tools do save time collating that first list.

2. Understanding the donor’s rules & requirements

  • NGOs need to understand the compliance requirements and regulations of a (new) donor to successfully attract donor funding and expand it.
  • When asking Bing or Google Bard about the compliance rules of this donor vis-a-vis this particular grant, the answers of our AI friend sound as nice and comprehensive as they are incomplete. In most cases, we noticed that the AI tool omitted significant regulations and requests.
  • Nonetheless, the chatbot gave a nice introduction to the requirements which can be used as a starting point to “ease oneself into the rulebook”, as one of our funding analysts used to say.

3. Writing a concept note or proposal

Generating convincing and well-written text is precisely what generative AI is made for, and unsurprisingly this is where it performs best. We find AI very useful for translating figures to narrative stories, such as when prompted with, “Read this budget here and summarize it in narrative form in no more than 400 words.” This is not surprising, as the AI’s ability to comprehend simple, large amounts of data is exactly what it was designed to do.

Where AI tends to be less effective is in crafting the “value for money” section, which always results in a lot of obvious wording looking like what you expect a first-year student of development economics to write. However, it is pretty good at reviewing written narratives and highlighting opportunities to say something in fewer words. For instance, when prompted with, “Read this and flag any opportunities to shorten the text without losing any information or conclusions,” it performs well.

4. Contract negotiation with donors

This is where AI continues to play no significant role. It is all down to the relationship with the prospective donor and knowing what budgetary, temporal and programmatic priorities the NGO has. Still, it is nice to get some help from the AI formulating these long emails.

Some warnings

  • AI-generated text can sound very convincing. That does not make it true. We frequently found AI “hallucinating”, stating something as a fact that is just plain wrong. Hence, reviewing with a very critical eye and fact-checking everything is crucial! That takes more time and might offset some of the time savings gained by using AI in the first place, but it is absolutely essential.
  • Data protection is key: AI processes information, including sensitive information. Never enter personal information or proprietary details.

Conclusion

Overall, the use of generative AI depends on the prompt and therefore the competence of the user. We are basically all programmers now. Yet, it’s worth noting that using it incorrectly might make outputs worse, not better. That is always a danger when adopting new technology, although the sheer immense benefits of AI render outright rejection of these tools simply impractical.

On another note: AI has been around for several years, with generative AI like ChatGPT being just the latest innovation. Undoubtedly, it will continue to evolve week by week, but it’s worth remembering that specific AI tools are already very good at managing those mundane tasks we want to get rid of. MzN, for example, implemented AI-driven finance tools, data-collection schedules and task managers at various NGOs with good success – worth looking into.