Methodology guide
How DealPrism grounded AI works
The math is locked. The AI explains it. This page shows how DealPrism keeps explanations tied to deterministic underwriting outputs instead of letting the model improvise financial analysis.
In plain English
DealPrism uses deterministic underwriting calculations as the source of truth, then layers AI on top to explain what the current assumptions and outputs mean in plain English.
The AI is not supposed to invent its own math. It works from structured deal data, labeled assumptions, allowed values, and validation guardrails so the response stays anchored to the live analysis.
What the system is trying to do
Rental property underwriting becomes hard to trust when explanations and calculations drift apart. DealPrism's architecture is designed to avoid that split.
The goal is simple: let the deterministic engine calculate the numbers, then let the AI help the investor understand the tradeoffs, sensitivities, and assumptions behind those numbers.
Grounding layers
| Layer | What it does | Why it matters |
|---|---|---|
| Deterministic math engine | Calculates cash flow, cap rate, cash-on-cash return, DSCR, and related outputs from the current assumptions | The numeric source of truth stays reproducible |
| Structured input data | Passes the current assumptions, outputs, and context into the AI request | The model works from the live deal, not from a generic template |
| Allowed-number registry | Restricts numeric references to values that exist in the analysis payload | This reduces unsupported or hallucinated figures |
| Schema and repair validation | Checks that responses fit the expected structure and retries or repairs invalid output | The explanation layer stays constrained and auditable |
What the AI should do
- Explain which assumptions are driving a result right now.
- Translate metrics like DSCR or cash-on-cash return into investor language.
- Call out assumptions worth verifying before making a decision.
- Describe how scenario changes would likely affect the current deal.
What the AI should not do
- Invent new deal numbers that are not present in the current analysis.
- Replace lender, legal, tax, or investment advice.
- Act like a single explanation is more reliable than the assumptions behind it.
- Override the deterministic underwriting engine.
Why this matters for investors
A trustworthy explanation layer is most useful when a deal is close, fragile, or assumption-sensitive. That is when investors need the system to point back to the exact rent, expense, or financing input creating pressure.
This is also why DealPrism treats assumptions as first-class context. If the assumptions move, the numbers move, and the explanation should move with them.
Frequently asked questions
- What does “grounded AI” mean in DealPrism?
- Grounded AI means the explanation layer is supposed to describe the current underwriting data, assumptions, and computed outputs instead of inventing its own math or unsupported numbers.
- Does the AI replace the underwriting engine?
- No. DealPrism uses deterministic calculation logic as the source of truth for metrics like cash flow, cap rate, cash-on-cash return, and DSCR. The AI layer explains those results in plain language.
- How does DealPrism reduce hallucinated financial numbers?
- The system constrains the AI response to the current input data, validates structured output, checks referenced value keys, and rejects or repairs responses that introduce unsupported numeric content.
Related resources
See the explanation layer on a live deal
Open a property in DealPrism and inspect how the assumptions, metrics, and grounded AI explanation line up against the same analysis.
Analyze your own dealDealPrism provides educational analysis based on available data and user assumptions. AI explanations are intended to describe the current analysis, not replace due diligence or professional advice.