Two commercial real estate principals shaking hands over a property model and market maps

A solo brokerage replaces manual deal research with a nightly scan of 180+ counties

180+
Counties scanned nightly
14
Motivated-seller signals
700–1,250
Hours returned per year
205
Sources cited in research

Client

Confidential

Asset type

Mobile home parks

Size

Solo operator, ~$150M+ AUM

System built

Sourcing Engine

The operation before the build

A brokerage run entirely on manual research

A deal travelled from a target county to an offer through six manual steps: find parks on a mapping overlay, pull the owning entity from a county GIS map, chase it through a state business registry, resolve it to a person, look up a phone number, and call every number until someone answered.

~100
Cold calls a week, reaching about 5 people
~50%
Of looked-up numbers wrong or out of service
1–2
Leads a week worth nurturing, none ready sellers

The data problem

180+ counties, 6 states, no two publishing records the same way

The audit verified the data landscape state by state and applied a four-tier access framework. Two integrations cover most of the footprint at zero data cost; a licensed parcel dataset covers only the gap.

72
Counties in one free statewide layer

Owner name and mailing address, queried nightly through a public ArcGIS FeatureServer.

59
More on a free opt-in layer

Including every metro county in that state. The rest fall to county portals.

38–66
Gap counties needing a paid license

Everything else runs on free state and county portals where they exist.

Research depth

205 sources
County records access, six state business registries, seller-signal sources, skip-tracing vendors, scraping feasibility.

Signals scored

14 across 3 tiers
Tenure, personal-trust ownership, tax delinquency, absentee ownership, mortgage age, recorded liens, lis pendens.

Audit duration

14 days
Diagnostic, deep analysis and solution mapping, roadmap, and hand-off.

What was built

The Sourcing Engine

The control flow is deterministic code. One bounded AI call handles PDF extraction when structured data is unavailable. The model never decides what to do, never computes a number, and never invents data.

Nightly pipeline

Acquire county data, identify parks by land use code, score each one, classify the owner, resolve entities, skip-trace, and write the ranked call list.

Owner resolution

Six state registry handlers plus assessor-address and deed fallbacks. Entities resolve to a person 35 to 50% of the time from registry data. Unresolved parks are marked, never fabricated.

Verified contact

Skip-tracing returns ranked, DNC-scrubbed numbers at about $0.07 per match, replacing directory lookups.