CASE 01 · PRODUCT
AI-generated investment memos in ninety seconds
Most deal screening still runs on a spreadsheet, an analyst, and forty-eight hours of context-switching. The Deal Intelligence Engine collapses that loop into ninety seconds: ingest deal data, score against a proprietary framework, output an IC narrative ready for a partner’s first read. Built, deployed, and operational.
- 90s
Per memo, end-to-end
ingest to output
- $0.001
Marginal cost per analysis
Gemini 2.5 Flash inference
- FDI
Proprietary scoring framework
founder dependency, four indicators
- Live
Deployed and accessible
production environment on Vercel
Deal screening is the most repeated workflow in private equity. It is also the worst documented.
An associate receives a teaser. They open a spreadsheet, copy financials, hunt down comparables, summarize the thesis in a Slack thread, and forty-eight hours later produce a one-page memo that a partner will read in four minutes. The output is structured. The process is not.
The cost of this is invisible because it is distributed. Two thousand associate hours per year per firm spent on context-switching between PDFs, models, and IC templates. Inconsistency in how deals are framed. Memos that reflect the analyst’s narrative more than the firm’s framework. And founder-dependency risk, the single highest predictor of post-close underperformance in lower-middle-market deals, almost never appears on the first pass.
A production tool, not a prototype. Next.js, Gemini 2.5 Flash, deployed on Vercel.
The Engine ingests three input paths: a deal teaser PDF, a manual data entry form, or a synthetic deal generator for demonstration. It runs the inputs against a proprietary scoring framework that includes the Founder Dependency Index, a four-indicator structural assessment of key-person risk, and outputs an investment narrative aligned with a standard IC memo template.
The architecture is intentionally minimal. Frontend in Next.js 14 deployed on Vercel. Inference through Gemini 2.5 Flash at a marginal cost of roughly one tenth of a cent per analysis. No vector database, no fine-tuning, no infrastructure overhead. The intelligence sits in the prompt engineering and the scoring framework, not in the stack complexity.
What the Engine demonstrates sits in the operating posture more than the technical sophistication: the same person who runs the deal sourcing logic also writes the code that automates it. The gap between “I have an idea” and “I have a deployed tool” is hours rather than quarters.
The intelligence sits in knowing which four questions a partner will ask before opening the memo, and answering them on page one. The model is execution. The questions are the work.
The most valuable AI applications in PE are built by operators who can encode their judgment into systems.
A generic LLM cannot screen a deal. A well-prompted LLM, scoped against a proprietary framework written by someone who has read three hundred memos and run two operational transformations, can produce a useful first pass. The encoding does the work. The technology is commodity.
The Engine is the live demonstration of that thesis. The Founder Dependency Index inside it is the operational lens, applying the same diagnostic logic that drives the Post-Acquisition Value Creation Playbook. The screening tool and the value creation framework are two visible surfaces of one underlying judgment about what determines returns in middle-market PE.
WHAT STAYED WITH ME
The cost of deploying production AI in a focused domain has collapsed. What has not collapsed is the cost of knowing what to deploy it on. The next decade of value creation will reward people who can diagnose the operational bottleneck and ship the tool that resolves it in the same week, with the same hands.
REQUEST THE FULL CASE FILE
The Engine is live and accessible. Try it with a sample deal at deal-intelligence-engine-eight.vercel.app, or request a walkthrough including the FDI scoring framework, the prompt architecture, and the integration path for a fund’s proprietary deal flow.
Request access