Raise the AI Barbell

By Mike Ryan, BPN Co-Founder and CEO

If you are an investor or corporate decision-maker, I strongly doubt that AI will replace you, but you could lose your job to someone who has mastered using AI agents more effectively than you have. The speed at which AI can read and write is undeniably powerful, but off-the-shelf LLMs do a very poor job of producing the kind of insightful investment memo or compelling slide deck needed to make a mission-critical decision, and spreadsheets are notoriously challenging for LLMs.

When we ask investors, “How are you using AI currently?” we typically hear responses like, “I’d love to be more efficient, but our work is bespoke…Other than summarizing documents or calls that I’ve already identified as important or getting basic information on topics I know little about, I found that AI can be a waste of time…LLMs can be dangerously wrong or effectively useless for important questions, so I end up doing it myself anyway.”  

Is it possible to harness the incredible efficiency of generative AI to produce useful, instead of wasteful, research to help you quickly build a model that frames the story with numbers to inform your decisions? We believe the answer is yes–but first, the AI agent needs to understand your thesis, learn ‘what you mean’ when you ask for certain facts or figures, and navigate your spreadsheet knowledgeably. You will need a purpose-built AI agent to do these specialized tasks well, and you will want a human in the loop to own the assumptions and review the memo, model, and slide deck before you put the conclusions into action -  just like you do in real life.

A purpose-built AI agent can automatically generate useful prompts and map useful answers to the financial model you are using to frame the story. That “model” could be super-simple (like estimating next year’s revenue and assigning a multiple based on similar companies), or it could involve a very detailed revenue buildup across multiple segments with lots of P&L line items and working capital assumptions -- either way, the decision-maker knows what they’re looking for, and the AI agents need to be on the same page. 

A purpose-built AI agent can produce a useful first draft of the memo–not just one that summarizes obvious points, regurgitates the company’s IR deck or makes off-topic detours–but a memo that lays out your specific thesis, captures your thinking about the key drivers, quantifies the risks, and produces a logical cash flow model that frames a meaningful (if tentative) conclusion, supported by hard evidence and real numbers.   

Is that AI-driven first draft memo and model sufficient for final decision-making?... Of course not.

However, it can be extremely useful if purpose-built AI agents can produce a draft 20x faster and deliver the memo and the model in an interactive format that makes it easy to update with human insight and additional AI research over time. The key is making sure AI actually understands the thesis and can map evidence to the assumptions in a model that produces the numbers you will eventually rely upon to reach your conclusion. If you are considering several cases, the AI agent should map evidence to build out those cases with hard numbers.

An objective team of AI agents that map evidence to the assumptions in your thesis and quickly frame logical cash flows with evidence can also mitigate cognitive bias. You can determine what future growth and profitability is priced into today’s valuation, capture information to illuminate blind spots in your assumptions and avoid jumping to conclusions that anchor you before the real facts and figures are available.

If the framing is on target and the thesis is well supported, you can take ownership of the assumptions and quickly proceed to a final decision with conviction. If the evidence conflicts with the thesis, or if there are simply too many holes, you can either deepen your research or move on to a more promising opportunity. In either case, delivering a useful framing with hard numbers mapped to current evidence - before the decision is largely baked - is a big advantage that frees up time for decision-makers and their highly-paid teams to spend their time doing what they do best.