I Used GPT-5 to Build an LBO Model
I used GPT-5 to build an LBO from scratch. Here’s what you need to know. Copy-paste prompts below: You can see the full prompt chain here. OpenAI describes GPT-5 as their smartest, fastest, and most practical model ever.
For the first time, a single model combines instant responses with deep reasoning, switching between quick outputs and detailed analysis depending on the complexity of your question. Every investor should master GPT-5. Why GPT-5 is a Breakthrough for Investors
I know GPT-5 doesn’t solve every finance pain. It’s not AGI. It’s not capable of replacing a human. It won't produce institutional-grade materials. But it can definitely make you much faster if you use it right. The cost of examining multiple opportunities and detailed analyses has dropped significantly; you can now quickly conduct preliminary version 0 screenings using AI before double-clicking manually where it makes sense. Use it anytime you need a back-of-the envelope analysis. Building an LBO Model Using GPT-5For private equity investors, rapid screening of LBO targets is important but incredibly time-consuming. Even preliminary, low-priority deals require integrating financials and manually tweaking assumptions / mechanics. GPT-5 can simplify this dramatically. Disclaimer: While not robust enough for creating your LBO for IC, GPT-5 provides a powerful first pass, significantly speeding up your workflow. Here’s how we set up our workflow to get best results:
Important: Activate “Thinking” Mode for noticeably improved results compared to the standard GPT-5 model. Step 1: The Data Room (Precise Data Retrieval)Accurate data forms the core of a solid financial model. GPT-5’s integrated coding sandbox allows direct parsing of documents, more accurately. It runs in the background without you needing to worry about it. We’ll use Under Armour for our example and drop in its latest 10-K filings, ensuring accuracy straight from the source. PROMPT 1: DATA RETRIEVAL You are a private equity analyst tasked with conducting an LBO for Under Armour. Your job is to extract the following data: - Current market capitalization and enterprise value - Last 3 years of financial statements (income statement, balance sheet, cash flow statement) Extract data only from these sources: - Market capitalization and enterprise value as of today based on live market data - Last 3 years of financial statements only from credible public sources (e.g., SEC filings, investor decks, financial databases). IMPORTANT: If filings are attached, use the attached filings. OUTPUT FORMAT - Show data in markdown tables, as well as a downloadable CSV file. - In the CSVs, each sections should be in a separate table: a) Current market capitalization and enterprise value, b) income statement, c) balance sheet, d) cash flow statement - Each of the financial statements should be output as reported How to use: replace with your target company and drop in its filings / financials. RESULT: Step 2: Teaching the AI to Build an LBO, And Then Building ItNow we move from data retrieval to actual analysis. Building an LBO demands niche domain knowledge, something LLMs are not inherently trained on. To bridge this gap, we use chain-of-thought prompting, an advanced technique where a clear, illustrative example is included within the prompt itself. This typical “paper LBO” example teaches the AI how to think step-by-step, exactly as a private equity associate would. The model learns the reasoning process from the provided example and applies it directly to the real-world data obtained in Step 1. To be clear, this initial output likely won’t suffice for your final IC memo at Blackstone. However, it’s an incredibly effective first pass, significantly accelerating your ability to screen more candidates quickly and identify key deal drivers early on. PROMPT 2: LBO CREATION Create an LBO analysis using the previous step's data as context. # SECTIONS ASSUMPTIONS: all of these will be assumptions from the user For each of these, suggest an initial value based on current market figures. Think hard about the assumptions. User may later choose to refine. - Purchase premium: the percentage - Debt / equity ratio (show implied Debt / LTM EBITDA) - Interest rate on debt - Revenue CAGR by exit (%) - EBITDA margin expansion (%) - Exit multiple (x NTM EBITDA) Fixed assumptions ** For these, base it off historical periods - Capex: in line with recent years - Net working capital:in line with recent years. Net working capital = current assets (excluding cash and cash equivalents) - current assets (excluding debt) - Tax rate: 21% - Assume all debt pay-down occurs at the moment of sale at the end of Year 5 - Exit at year 5 SOURCES & USES TABLE Sources: - Debt financing - Equity contribution ** You must make sure that sources = uses. Start with uses, then back solve for sources necessary Uses: - Purchase price of equity - Debt refinancing - Transaction fees (advisory) - Financing fees (on debt) FINANCIAL FORECASTS - Periods: Show two years of historical information + 5 years of forecasts for each of the below - Revenue and growth - EBITDA and EBITDA margin - Net Income and net income margin - Capex - D&A - Net working capital = current assets (excluding cash and cash equivalents) - current assets (excluding debt) - Increase in Net Working Capital - Debt Amortization - Free cash flow = Net Income + D&A - Increase in net working capital - capex - debt amortization RETURNS MATH - Enterprise value: NTM EBITDA x exit multiple - Ending debt balance: entry debt - cumulative free cash flow (used for debt paydown) - Sponsor equity value at exit: enterprise value - cumulative debt paydown - MoM = sponsor equity value at exit / entry sponsor equity contribution - IRR = based on the time period SENSITIVITY TABLE - Sensitivity table sensitizing a) exit multiples on one axis (range of +- 20%, 40%) and b) entry premium on one axis (range of +- 10%, 20%) # FORMATTING - Each section should be its own separate table - Put years in columns, metrics in rows ALWAYS #LBO LOGIC EXAMPLE Q: Build a markdown table with the entry sponsor equity value, exit sponsor equity value, MoM multiple, and IRR. Assume the following parameters for transaction and company. - Entry multiple of 5x NTM EBITDA at the end of year 0 - Debt / Equity for acquisition is 60:40 - Interest rate on debt: 10% - Revenue: $100M in year 1, with 10% YoY growth thereafter. - EBITDA margin: 40%, expected to stay flat - Capex: 15% of revenue each year - Operating working capital: increases by $5M each year - Depreciation: $20M each year - Tax rate: 40% - Exit: after year 5 at same EBITDA multiple used at entry (5x NTM EBITDA) - Assume all debt pay-down occurs at the moment of sale at the end of Year 5 A: The purchase price is $200M (5x NTM EBITDA x $40M EBITDA in Year 1 ($100M Revenue in Year 1 x 40% margin). Sponsor entry equity is $80M (40% x $200M total entry value). Revenue at year 6 is $161M ($100M growing 10%) and EBITDA will be $64M (revenue x 40%). Cumulative free cash flow will be $34M, calculated by taking EBI + D&A - Capex - Increase in net working capital for each year. Enterprise value at exit is $322M ($64M of EBITDA in Year 6 x 5x multiple). Ending debt balance is $86M, which is $120M entry debt ($200M purchase price x 60% debt). Sponsor End Equity Value = $236M ($322M enterprise value - ending debt of $86M). MoM multiple is 3.0x ($236M ending sponsor equity / $80M entry sponsor equity), leading to a ~27% IRR. RESULT: Step 3: Interactive Dashboard (Dynamic Analysis)Convert your static analysis into an interactive tool where you can play around with assumptions. GPT-5’s advanced coding environment can build and run code, allowing to vibe code a web-app with a single prompt. The below prompts the AI to take the completed LBO and turn it into a dashboard where we can explore various scenarios by adjusting assumptions dynamically. Important: Now disable thinking mode. I found that GPT-5's default model yielded better results for coding. PROMPT 3: DASHBOARD CREATION Create a single-page web-app with the following requirements. # BASICS - Name: LBO dashboard - Purpose: To create a dashboard showing an leveraged buyout model for private equity investorsInput: the context of this chat, with a fully modeled LBO # SECTIONS 1) Assumptions: The user should be able to tweak each of the assumptions below, and that should carry forward in the calculations for the subsequent sections - Purchase premium: the percentage - Debt / equity ratio (show implied Debt / LTM EBITDA) - Interest rate on debt - Revenue CAGR by exit (%) - EBITDA margin expansion (%) - Exit multiple (x NTM EBITDA) 2) Sources and Uses 3) Financial Forecasts 4) Returns Math 5) Sensitivity Tables # FORMAT - Dark color - Clean, modern aesthetic - User assumption boxes should be prominently highlighted in yellow - Each section should have appropriate headers - Each section should be its own separate table - Put years in columns, metrics in rows ALWAYS RESULT: You can see the full dashboard here. Next Steps: From Model to Investment MemoThis is pretty amazing for what is possible with just a few prompts, each in a single shot. If you find errors when you run this, try guiding the AI to make the appropriate changes for your use case. Now that you have an interactive model, what’s next?
The key is thinking of AI as steps in a chain, using each output as the input for the next, more sophisticated task. The Bottom LineChatbots answered questions. Agentic AI executes your workflow. GPT-5 takes us another step closer. While the quality isn’t yet at a final, institutional-grade level, it produces a sophisticated first-pass tool that is incredibly useful for rapid screening and analysis. We'll learn more about how to make this better and better over time. Until then, happy screening. Upcoming events: |
Enjoyed this article?
Get more AI for finance content delivered to your inbox.