Building a Property Valuation Agent: Automated CMAs with AI Analysis

What Is an Automated CMA and Why Automate It?

A Comparative Market Analysis (CMA) is the backbone of real estate pricing. Agents compare a subject property against recently sold comparable properties ("comps") to estimate fair market value. Manually, this takes 1-3 hours per property — pulling data, adjusting for differences, and formatting a report.

An AI valuation agent compresses this to minutes by automating comp selection, adjustment calculations, and report generation while keeping a human in the loop for final review.

Finding Comparable Properties

The first tool searches for comps within configurable parameters.

flowchart LR
    CALLER(["Buyer or Seller Lead"])
    subgraph TEL["Telephony"]
        SIP["Twilio SIP and PSTN"]
    end
    subgraph BRAIN["Real Estate AI Agent"]
        STT["Streaming STT<br/>Deepgram or Whisper"]
        NLU{"Intent and<br/>Entity Extraction"}
        TOOLS["Tool Calls"]
        TTS["Streaming TTS<br/>ElevenLabs or Rime"]
    end
    subgraph DATA["Live Data Plane"]
        CRM[("CRM and Notes")]
        CAL[("Calendar and<br/>Schedule")]
        KB[("Knowledge Base<br/>and Policies")]
    end
    subgraph OUT["Outcomes"]
        O1(["Showing scheduled"])
        O2(["Lead routed to agent"])
        O3(["Pre-qual handed to broker"])
    end
    CALLER --> SIP --> STT --> NLU
    NLU -->|Lookup| TOOLS
    TOOLS <--> CRM
    TOOLS <--> CAL
    TOOLS <--> KB
    NLU --> TTS --> SIP --> CALLER
    NLU -->|Resolved| O1
    NLU -->|Schedule| O2
    NLU -->|Escalate| O3
    style CALLER fill:#f1f5f9,stroke:#64748b,color:#0f172a
    style NLU fill:#4f46e5,stroke:#4338ca,color:#fff
    style O1 fill:#059669,stroke:#047857,color:#fff
    style O2 fill:#0ea5e9,stroke:#0369a1,color:#fff
    style O3 fill:#f59e0b,stroke:#d97706,color:#1f2937

from dataclasses import dataclass
from typing import Optional
import math

@dataclass
class ComparableProperty:
    address: str
    sale_price: float
    sale_date: str
    sqft: int
    bedrooms: int
    bathrooms: float
    lot_size: float  # acres
    year_built: int
    distance_miles: float
    property_type: str

def haversine_distance(lat1: float, lon1: float, lat2: float, lon2: float) -> float:
    """Calculate distance between two coordinates in miles."""
    R = 3959  # Earth radius in miles
    dlat = math.radians(lat2 - lat1)
    dlon = math.radians(lon2 - lon1)
    a = (
        math.sin(dlat / 2) ** 2
        + math.cos(math.radians(lat1))
        * math.cos(math.radians(lat2))
        * math.sin(dlon / 2) ** 2
    )
    return R * 2 * math.asin(math.sqrt(a))

async def find_comparables(
    subject_lat: float,
    subject_lon: float,
    subject_sqft: int,
    radius_miles: float = 1.0,
    sqft_tolerance: float = 0.2,
    max_results: int = 6,
    pool=None,
) -> list[ComparableProperty]:
    """Find recently sold properties similar to the subject."""
    min_sqft = int(subject_sqft * (1 - sqft_tolerance))
    max_sqft = int(subject_sqft * (1 + sqft_tolerance))

    rows = await pool.fetch("""
        SELECT address, sale_price, sale_date, sqft, bedrooms,
               bathrooms, lot_size, year_built, latitude, longitude,
               property_type
        FROM sold_properties
        WHERE sqft BETWEEN $1 AND $2
          AND sale_date >= NOW() - INTERVAL '6 months'
        ORDER BY sale_date DESC
        LIMIT 50
    """, min_sqft, max_sqft)

    comps = []
    for row in rows:
        dist = haversine_distance(
            subject_lat, subject_lon,
            row["latitude"], row["longitude"],
        )
        if dist <= radius_miles:
            comps.append(ComparableProperty(
                address=row["address"],
                sale_price=row["sale_price"],
                sale_date=str(row["sale_date"]),
                sqft=row["sqft"],
                bedrooms=row["bedrooms"],
                bathrooms=row["bathrooms"],
                lot_size=row["lot_size"],
                year_built=row["year_built"],
                distance_miles=round(dist, 2),
                property_type=row["property_type"],
            ))
    comps.sort(key=lambda c: c.distance_miles)
    return comps[:max_results]

Applying Valuation Adjustments

Raw comp prices need adjustments for differences in size, features, and condition.

@dataclass
class AdjustedComp:
    comp: ComparableProperty
    sqft_adjustment: float
    bedroom_adjustment: float
    age_adjustment: float
    adjusted_price: float

def calculate_adjustments(
    subject_sqft: int,
    subject_beds: int,
    subject_year: int,
    comp: ComparableProperty,
    price_per_sqft_market: float = 250.0,
) -> AdjustedComp:
    """Apply standard CMA adjustments to a comparable property."""
    sqft_diff = subject_sqft - comp.sqft
    sqft_adj = sqft_diff * (price_per_sqft_market * 0.5)

    bed_diff = subject_beds - comp.bedrooms
    bed_adj = bed_diff * 10_000

    age_diff = comp.year_built - subject_year  # newer comp = negative adjustment
    age_adj = age_diff * 1_500

    adjusted = comp.sale_price + sqft_adj + bed_adj + age_adj

    return AdjustedComp(
        comp=comp,
        sqft_adjustment=sqft_adj,
        bedroom_adjustment=bed_adj,
        age_adjustment=age_adj,
        adjusted_price=adjusted,
    )

Each adjustment uses a simplified linear model. In production systems, you would derive these multipliers from local market regression analysis rather than hard-coding them.

Wiring Into an Agent Tool

from agents import Agent, function_tool, Runner

@function_tool
async def generate_cma(
    address: str,
    sqft: int,
    bedrooms: int,
    year_built: int,
    latitude: float,
    longitude: float,
) -> str:
    """Generate a Comparative Market Analysis for a property."""
    comps = await find_comparables(
        subject_lat=latitude,
        subject_lon=longitude,
        subject_sqft=sqft,
    )
    if len(comps) < 3:
        return "Insufficient comparable sales found. Try expanding the search radius."

    adjusted = [
        calculate_adjustments(sqft, bedrooms, year_built, c)
        for c in comps
    ]
    prices = [a.adjusted_price for a in adjusted]
    avg_price = sum(prices) / len(prices)
    low_estimate = min(prices)
    high_estimate = max(prices)

    report_lines = [
        f"## CMA Report for {address}",
        f"Estimated Value: ${avg_price:,.0f}",
        f"Range: ${low_estimate:,.0f} - ${high_estimate:,.0f}",
        f"Based on {len(comps)} comparable sales\n",
    ]
    for a in adjusted:
        report_lines.append(
            f"- {a.comp.address}: Sold ${a.comp.sale_price:,.0f}, "
            f"Adjusted ${a.adjusted_price:,.0f} "
            f"({a.comp.distance_miles} mi away)"
        )
    return "\n".join(report_lines)

valuation_agent = Agent(
    name="PropertyValuationAgent",
    instructions="""You are a real estate valuation specialist.
    When given property details, generate a CMA report using the tool.
    Explain the adjustments clearly. Always note that this is an
    estimate and recommend a professional appraisal for lending.""",
    tools=[generate_cma],
)

Generating a Professional Report

The agent's LLM output serves as the narrative section. For a formatted PDF, you can add a report generation step.

from datetime import date

def format_cma_report(
    subject_address: str,
    cma_data: str,
    agent_narrative: str,
) -> dict:
    """Structure a CMA report for PDF generation."""
    return {
        "title": f"Comparative Market Analysis - {subject_address}",
        "date": str(date.today()),
        "sections": [
            {"heading": "Executive Summary", "body": agent_narrative},
            {"heading": "Comparable Analysis", "body": cma_data},
            {"heading": "Disclaimer", "body": (
                "This CMA is an estimate based on recent sales data. "
                "It is not a formal appraisal and should not be used "
                "for lending purposes."
            )},
        ],
    }

FAQ

How accurate are automated CMAs compared to manual ones?

Automated CMAs are typically within 5-10% of manually prepared analyses when sufficient comparable data exists. The main limitation is that they cannot account for interior condition, renovations, or curb appeal without additional data inputs like inspection notes or photos.

How do you handle markets with few comparable sales?

The agent expands the search radius incrementally (1 mile, then 2, then 5) and widens the square footage tolerance. If fewer than three comps are found even after expansion, it reports insufficient data rather than generating a misleading estimate.

Should the AI agent replace professional appraisals?

No. Automated CMAs are excellent for quick pricing guidance, listing price recommendations, and initial buyer analysis. Formal appraisals required by lenders must still be performed by licensed appraisers who physically inspect the property.

---

#PropertyValuation #CMA #RealEstateAI #Python #MarketAnalysis #AgenticAI #LearnAI #AIEngineering