Research snapshot · 09 Oct 2026
Address → calibrated exterior draft → UniFi

Build the pipeline. Don’t buy the fantasy.

No turnkey product was found that converts an address into a dimensioned exterior-wall plan for WiFi design. The strongest budget path is public GIS, Microsoft footprints, a Google Maps satellite cross-check, a scaled PNG, and field measurements.

85confidence in the
assemble verdict
Recommendation: assemble the workflow.Free data gets a calibrated draft. Two control measurements test scale; final verification requires every critical wall, corner, annex, and roof-to-wall offset to be checked. The UniFi handoff remains manual.
REFERENCE CASE · ~114 × 82 FTN ↑
114.0 ft ~82 ft PNGSCALED
Parcel context — not a wallML / traced footprintOn-site control measurement

The six-stage field pipeline

Every stage has one job. The chain separates location evidence, shape inference, file production, and final measurement so no single data source can masquerade as truth.

CRITICAL DISTINCTION

A parcel polygon is the legal/tax lot boundary. It is not the building outline and should never be used to calculate exterior wall lengths.

01 / LOCATE

Geocode address

Use Census or Esri to convert the street address to a point.

CONF · 90
02 / CONFIRM

Find the parcel

Query county GIS by point. Confirm the correct lot and ownership context only.

CONF · 70
03 / OUTLINE

Footprint + satellite

Pull Microsoft’s polygon, then cross-check its shape and staleness against Google Maps satellite imagery. Use the imagery to inspect—not as an export API.

CONF · 80
04 / RENDER

Make the plan

Export a scaled PNG with dimensions, north arrow, scale bar, and parcel overlay.

CONF · 85
05 / MODEL

Load into UniFi

Upload the PNG, calibrate one printed reference length, then draw materials and APs.

CONF · 90
06 / VERIFY

Measure on site

Use two controls to test scale, then measure every critical segment and annex before calling the plan field-verified.

CONF · 95

Grounded measurements

  • Laser, phone LiDAR, RTK drone, and photogrammetric reports measure physical geometry.
  • County parcels derive from assessor/deed records, but are not survey-certified and do not represent walls.
  • Two controls calibrate scale; only complete critical-segment checks justify a field-verified label.

Estimated geometry

  • Microsoft footprints and Google Solar are ML inference from imagery.
  • Manual satellite tracing still inherits image registration and roof-overhang error.
  • Use these sources to draft, never to certify.

Tool stack, with the marketing removed

Filter the candidates. “Grounded” means measured or compiled from records. “Estimated” means ML or imagery interpretation. Pricing marked as conflicting stays unresolved.

GOOGLE MAPS ROLE

Use satellite view as the human visual cross-check for footprint shape, additions, and obvious staleness. This research did not verify a supported Maps satellite-export endpoint, capture date, per-site ground resolution, or reuse terms for a UniFi deliverable—so those remain open and the recommended automation source stays Microsoft’s downloadable footprint data.

Microsoft Building Footprints

ESTIMATED
999M GeoJSON polygons
2017–2022 imagery
98.5% precision · 92.4% recall · IoU 0.86
Free · STAC / downloads / tiles

Best free outline source. There is no hosted point-query API; local quality varies, especially rural versus urban.

County ArcGIS

GROUNDED*
Parcel / tax-lot polygon
County-dependent; not survey grade
REST directory blocked
Free where exposed

*Grounded in assessor/deed records—not wall geometry. Wabash now embeds ArcGIS Experience Builder, so the hosted feature service must be discovered per county.

UniFi Design Center

CONSUMES PLAN
JPEG / PNG
Manual reference line
No import API found
Walls + materials + APs

The pipeline endpoint is a dimensioned PNG. Upload, calibrate, draw attenuation materials, then model 2.4/5 GHz coverage.

EagleView

MEASURED
98.77% accuracy
~1-inch GSD
$15–$87 residential; $75+ commercial
Not self-serve

Strong paid verification for large jobs, but accuracy is tier-dependent and exact commercial pricing is quote-based.

Nearmap

MEASURED
Ortho, oblique, 3D, AI layers
1.7–2.8 in GSD in US
~$220/mo/user, aggregator
Tile + AI, enterprise

Potentially useful in covered metros. Rural Midwest coverage is likely sparse, and no independent measurement-accuracy figure was found.

Regrid

GROUNDED*
156M+ parcels
Matched Building Footprints
$375/mo API or $300/county
$29/mo listed elsewhere

*Parcels are grounded; matched footprints are a separate product. API derivative works require written consent. Pricing conflict remains unresolved.

RTK drone + Pix4D / DroneDeploy

MEASURED
±0.1–0.3 ft
~$1,200/year + hardware
Yes
Capture + Part 107

The accuracy ceiling. Use for large or high-value sites; it is operationally and financially excessive as the default pre-sales path.

Google Solar API

ESTIMATED
Roof segments, areas, pitch
5/5 within ±10%; 1.8% avg
~$0.20/report
Residential-leaning

Roof Recon proved the chain can work, but it is a hackathon roof estimator—not a wall-plan tool—and church rooflines are outside its sweet spot.

Google Open Buildings

ESTIMATED
Polygon + confidence + area
~90% precision · ~70% recall
None
Exclude

Useful elsewhere, but V3 does not cover the United States. It cannot serve rural Midwest jobs.

LightBox / ATTOM

PARCEL DATA
Evaluation / enterprise
Derivative-work restrictions
Client deliverables blocked
Exclude

The legal terms conflict with producing client-facing plans from the data. Capability does not rescue the licensing mismatch.

Accuracy ladder

Expected error for a roughly 114 × 82 ft commercial building. Select a method to see what the number means and where it comes from.

Grounded measurement · rank 1

RTK drone photogrammetry with ground control

±0.1–0.3 ft

The gold-standard verification path. Published guidance places RTK horizontal accuracy around ±2 cm; 1-inch GSD drone mapping yields roughly 2–6 inches of vertical accuracy. Best reserved for high-value or complex sites.

Basis: sourced measurement claims. Hardware, software, field capture, and FAA Part 107 make it too heavy for the default MSP workflow.

Discrepancies & hard stops

These are not footnotes. Each one can turn a precise-looking plan into a wrong answer, an unavailable workflow, or a deliverable the MSP cannot legally reuse.

Parcel boundary mistaken for a wall

HARD STOP

Parcel geometry describes the lot. Rural tax maps can be displaced by tens of feet and are not survey-certified. Use the parcel to locate the building, never to measure it.

EagleView’s 98.77% is tier-specific

CONFLICT

The vendor cites CompassData verification; a 2025 contractor review says a lower report tier missed six squares and EagleView pointed to Premium. Do not apply the headline figure to every report.

Regrid pricing and rights do not line up

UNRESOLVED

2026 terms research found $375/month for the API and written consent for derivative works; an aggregator lists $29/month. Treat $375 as the credible API figure until Regrid confirms the exact client-deliverable tier.

Microsoft data may predate the building

STALE RISK

The US refresh is based on 2017–2022 imagery. New construction and later additions can be missing. Compare every candidate polygon to current aerial imagery before rendering.

Heartland reference case is not a rectangle

4.6 FT SPREAD

The two traced short walls were 84.7 ft and 80.1 ft. “~82 ft” is only their rounded summary, not a common measured width. The older north annex was excluded from the footprint. With ±3–6 ft satellite-trace uncertainty, the drawing remains preliminary until field checks.

Roofline is not wall line

SYSTEMATIC BIAS

Satellite and ML sources see roofs. Eaves can add roughly 1–2 ft of outward bias. A laser check removes that error; more zoom does not.

Wabash GIS changed architecture

INTEGRATION WORK

The bhamaps REST directory returned AccessDenied on 9 Oct 2026. The site now wraps an ArcGIS Experience Builder app, so its underlying hosted feature service must be identified separately.

Nearmap rural coverage is an open question

VERIFY PER SITE

The capture program focuses on populated areas. No independent measurement-accuracy figure was found, and specific rural Illinois coverage requires a site-by-site map check.

Google Open Buildings has no US coverage

EXCLUDE

V3 covers Africa, South Asia, and parts of Europe with 2026 additions in Southeast Asia—but not the United States. It is not a fallback for Midwest jobs.

Confidence register

The headline 85 is the research score for the specific “assemble, don’t buy” conclusion—not an average of this register. The entries below grade separate recommendations; high confidence still does not mean zero field verification.

95

Verify with laser / LiDAR

Measured data outranks ML. Independent LiDAR tests support ~50 mm simple-room and ~170 mm complex-room accuracy; exterior range is extrapolated.

95

Exclude Google Open Buildings

No US coverage; the geography mismatch is decisive.

90

Deliver a dimensioned PNG

UniFi accepts JPEG/PNG and requires manual scale calibration. No import API was found.

85

Assemble instead of buy

No WiFi-focused turnkey address-to-plan product was found. Roof Recon is the closest public analog, but it is roof-focused and hackathon-grade.

85

Use RTK drone only on large jobs

Excellent accuracy, but capture, hardware, software, and Part 107 overhead make it a premium path.

80

Use Microsoft for free outlines

US-wide with published metrics, but rural variance and 2017–2022 imagery require visual and field checks.

75

Paid aerial as optional upgrade

EagleView and Nearmap are measured, but pricing, coverage, and self-serve access do not fit every presales survey.

70

Use county ArcGIS for context

Free and grounded, but every county is different and Wabash’s public directory is blocked.

70

Prefer Regrid county data over API

Likely cleaner for internal tooling, but price tiers and derivative-work rights need direct confirmation.

65

Manual trace can beat ML

Plausible when imagery is current and a known reference exists, but the ±3–6 ft range is an engineering estimate, not a single published benchmark.

How the answer was checked

The research pass separated live observations from indexed source material, reported conflicting claims instead of averaging them, and labeled every derived error range.

01 · CLASSIFY

Every source is tagged as grounded measurement, grounded parcel record, inferred geometry, or an RF-design consumer.

02 · TRIANGULATE

Vendor accuracy claims are paired with independent or contrary evidence where the research found it.

03 · PRESERVE HEDGES

“Claimed,” “aggregator,” “derived,” and “engineering estimate” stay attached to the numbers they qualify.

04 · FAIL CLOSED

Missing coverage, blocked endpoints, uncertain license terms, and unavailable prices remain open questions—not invented answers.