AI comes up in almost every deal flow conversation here, so I thought I'd share my full stack & prompts you can use. The first four you may already be doing. The fifth is the one that completely changed how I search. Link to Prompts: redacted Use these prompts as a starting point that you can feed into your agent of choice to customize to your needs. I have stripped out the specific information for my own search and tried to standardize them so you can customize them for yourself. Just copy and paste one into your agent with "Let's customize this for what I'm looking for ..." and let the agent expand upon the template. Starting Point: For context, my background is in technology. I've tested every model, API, Agent Harness... you name it, I've tried it. I've settled on this setup: I pay for both Claude Max and ChatGPT Pro. Claude Max is my workhorse. My ChatGPT is connected to an Hermes Agent that scours many brokerages & listing sites every day for listings. I only use ChatGPT for this because Anthropic has restricted account usage in agent harnesses like Hermes. 1. Listing alerts — redacted Have Claude Scheduled Tasks (or OpenClaw, Hermes, etc.) check listing sites on a schedule and dedupe against what you've already seen. I have a separate task for every brokerage I come across, so instead of checking 15 websites every morning, I wake up to one clean list of only the new opportunities published in the last day. You can also have AI summarize each listing, flag anything that matches your buy box, and ignore industries or deal sizes you don't care about. It saves real time and eliminates the mental overhead of constantly refreshing brokerage sites. The limitation: you're still competing for the same public listings every other buyer sees. I have shared on prompt for a brokerage so you can copy modify for your brokerages. 2. Deal analysis skills — redacted Build reusable AI skills (or prompts) for the analyses you perform on every opportunity. Mine automatically review financials, analyze the industry, summarize customer concentration risk, identify competitors, research local market trends, and highlight operational red flags. The consistency is almost more valuable than the speed. Every CIM gets evaluated using the same framework, so it's much easier to compare opportunities instead of relying on intuition. It's also surprisingly good at finding details that are easy to miss. I've had it uncover a major competitor opening up, identify road construction that would significantly impact traffic to a retail business, and point out customer or supplier risks buried deep in documents. Those aren't reasons to reject a deal on their own, but they're exactly the questions I want answered before spending days in diligence. 3. Financial modeling with the Claude Excel extension — redacted This has become one of the biggest time savers in my process. Seller financials almost always arrive in different formats: PDFs, QuickBooks exports, handwritten Excel files, inconsistent account names, and years that don't line up. The Claude Excel extension lets me pull P&Ls directly from PDFs into spreadsheets, standardize messy financials, clean inconsistent categories, and build the same model for every business I evaluate. I merge every year's statements into one workbook so I can compare performance across time without constantly flipping between files. Then I can quickly do analysis like- "Generate a Jan–Dec monthly revenue chart stacking each year on top of one another." "Highlight months where gross margin deviated more than 5%." "Find any expense categories growing materially faster than revenue." Instead of spending an hour building spreadsheets, I spend that hour understanding the business. 4. Document generation — redacted This prompt is meant only as a starting point. Several law firms have LOI, IOI, and other templates freely available. I would use those and use Claude + Word extension to quickly fill it in. All of these are great when you already have a CIM in hand, or a business you know you want to analyze. However how do you find those companies? An SBA loan broker told me something a while back that I haven't been able to shake & others validated earlier, roughly 50% of deals never hit a listing site at all. Which brings me to #5. 5. Sourcing off-market companies — Instead of using AI to react to listings faster, I built visibility into EVERY small business in the US. Top-down instead of bottom-up. A massive database of every company in the US, mapped them out, and estimated their SDE based upon PPP data. To say this is a big project, it has taken months and several iterations. How it works: it started with PPP data— * business name, * employee count, * payroll, and * NAICS code for millions of established companies. I layered industry revenue and margin benchmarks on top (adjusted for inflation) to estimate revenue and SDE for each one. For context, my background is in software (founding & selling Tindie.com, Netflix, Yelp, and other companies in Silicon Valley). Now I can filter millions of companies by location, industry, estimated SDE, number of employees, and much more. My own search: 50 miles from home, $750k–$1.3M SDE, exclude the following industries... From millions of companies, it becomes < 1k companies that I can reach out directly to the owners. How do I do that? I connect my Claude account to the data, and I can quickly research every company, find the website, find the owners, and draft an email to contact them. I'm no longer competing with everyone under the sun and have had great results so far. If you’d like access to it when it’s available, please leave a comment on this post. After I gave access to a few searchers, and had tremendous response, I figure it’s getting time to finally share with more people. Anyway, those are the prompts and tools I use. Hopefully it was helpful! (Screenshot of the businesses) redacted