2025 was the year that artificial intelligence (AI) really broke through to the public. Some board members may be wondering if they can outsource their reserve planning to AI and if it is a reasonable idea, so we are going to explore that idea in this topic. First, do AI models (Chatgpt, Perplexity, Copilot, Gemini) even know what a reserve study is? Yes, the AI models do understand the concept of a reserve study. Okay great, but can AI models spit out a realistic funding model for our association? From my experience, it is not close to producing something credible and reliable.
AI Reserve Study Issues
Let’s look closer at why AI models struggle to produce a credible reserve study. First, all of the proper items have to be inputted into the model. This places a ton of responsibility on the person prompting the AI model. This would include all of the specific items and their associated quantity sizes. The models would have no idea of what items should be included and what their quantities are. Second, the model has no way of knowing the condition of each item. The only way that an AI model could develop a remaining life of an item is to deduct the number of years from when the community started, but would have no way of knowing any damage/deterioration/partial replacements of the item. Third, AI models have a very hard time of pinning down a specific cost for an item, as prompting the models the current cost of an item produces a very large range of costs. This effect over many items would produce a huge range of required reserve funding. In other words, the models lack nuance to accurately find the proper replacement cost of an item. On the contrary, professional reserve specialists are going to have a much more specific database that will utilize specific costs.
Broader AI Issues
AI models are best for non-specific ad relatively vague things like composing an email, creating a template, and summarizing information. However, they struggle much more with precise and critical things like medical advice, legal advice, and financial advice. In fact, Open AI (owner of Chatgpt) had an internal report that suggested they were wrong 40% of the time. The user of these services has to be careful to what the AI model’s inputs were and that there was no data poisoning occurring. The best way to think of AI models for reserve planning is to use it as a starting point to gather information to get a range of costs or lifespans, but do not count on it for detailed and precise results.