Homebase's September–October 2026 property-data research led us to judge BatchData, ATTOM, DataTree and PropertyRadar equivalent at the underlying source-data layer for our local house-buying purposes. We chose to do sifting, or derivatives, and stacking in-house around the local buyer's strategy. This was an operating conclusion, not a finding that every vendor's fields, estimates, dates or coverage were identical.
The studies explain that distinction. Basic building facts often corroborated one another. Estimated debt, vacancy and tax signals could disagree. Apparent missing properties sometimes turned out to be search exclusions. Current debt questions required evidence beyond a provider flag. None of this research measured response-rate, acquisition or profit improvement.
TL;DR
- Our local source-data conclusion moved strategy into in-house sifting and stacking, while leaving coverage and verification as active questions.
- Study denominators differ: 16 selected properties, thousands of matched records and 177 county-checked parcels cannot become one accuracy score.
- Building-fact agreement did not establish identical estimates, vacancy labels, tax status or financing conclusions.
- Separate source records, derived signals and current verification before using the results to buy data or construct a mailing program.
Study register: what was actually compared
These were selected diagnostics, not a random national survey. Results below describe the historical checks; public capability descriptions were consulted separately on November 14, 2026.
| Study and date | Actual comparison population | What it can tell us |
|---|---|---|
| Same-property diagnostic, September 30 | 16 properties across five markets; BatchData, ATTOM and PropertyRadar | Pairwise agreement on specific jointly populated fields |
| Financing challenge, October 2 | 24 selected contradictions plus 10 controls | Whether an alternative estimate resolved an apparent debt-free label |
| Apparent search gaps, October 5 | 20 selected properties across four states | Whether a missing result represented missing coverage or search behavior |
| DataTree study A, October 8–9 | 8,843 BatchData-matched properties from 18,981 DataTree export records | Basic-fact corroboration, with field-specific denominators |
| DataTree study B, October 8–9 | 5,857 neighborhood matches from 5,859 saved BatchData candidates | Building agreement and divergent signals in a separate study |
| County tax check, October 8–9 | 177 selected Arizona parcels, enriched for issues and including controls | Whether flags met a dated county-balance screen |
Basic facts corroborated; estimates diverged
The September diagnostic included 39 data points and 624 comparison rows. Those rows were not independent properties. ATTOM's three requests per property likewise produced 48 response groups without increasing the sample beyond 16.
Year built matched exactly across all three providers on all 16 properties. BatchData and ATTOM also matched exactly on floor area for all 16; each comparison involving PropertyRadar matched on 15.
For estimated loan balances, BatchData and ATTOM were within 2% on seven of 13 jointly populated properties. BatchData and PropertyRadar were within 2% on eight of 13. ATTOM and PropertyRadar were within 2% on 10 of 16. Missing BatchData numeric estimates on three properties were not zero debt. ATTOM and PropertyRadar AVMs agreed within 2% on six of 16; the tested BatchData package supplied no AVM.
The 2% rule divided absolute difference by the larger absolute value, with exact equality accepted. There was no independent appraisal, physical measurement or payoff target to establish which differing value was right.
Larger matches did not create a universal accuracy score
DataTree study A covered 12 requested neighborhoods, one blank and 11 usable in Arizona and Illinois. Within the matched records, floor area agreed within 5% on 99.97% of 8,830 populated comparisons. Tax-bill amounts agreed on all 8,804 populated comparisons. Those are corroboration figures, not proof of current debt or independent accuracy.
Study B used a different universe. Of 5,859 saved candidates, 5,857 matched within neighborhood searches, one more was found through separate address lookup and one remained unresolved. Year built agreed on all 5,857 neighborhood matches; floor area agreed on 5,856. Vacancy labels were much less aligned: BatchData flagged 60, DataTree 20, with six shared.
Report restrictions limited investigation. Study A used 25 TotalViews and five detail reports. Study B completed 40 search sets, but trial refusals prevented a planned 245-property history review from yielding additional usable TotalViews. These figures summarize saved historical reports, not an independent re-audit of every original response.
Current debt required a different test
The October 2 financing challenge selected 24 apparent contradictions and 10 controls. All 24 challenge properties returned nonzero ATTOM estimated balances. Nine controls returned data, two with nonzero estimates; one control returned no result. This enriched sample did not establish a general error rate, and an alternative estimate did not prove an enforceable balance or disprove the original label.
The county tax study supplied another warning. Its historical screen was at least $500 still due on 2025-or-earlier past-due bills. Among 177 investigated parcels, DataTree flagged 54 and 23 met the screen; BatchData flagged 39 and 17 met it; PropertyRadar flagged 130 and 44 met it. The cohort contained 45 county-positive cases. Flags overlapped, and changing the year or balance definition could change relative results.
An early 36-of-36 agreement on historical tax labels had looked reassuring. Later county balances showed why it was not proof of current debt. Cases outside the screen were not necessarily debt-free: small amounts, bill dates and payment arrangements mattered.
Search omissions were not always record omissions
All 20 properties selected for the October 5 gap investigation could be found in BatchData after relaxing constraints or using direct lookup. Owner classification, count interpretation and property or value rules explained apparent gaps, with overlapping causes. This supported debugging the query before changing vendors, not claiming that missing coverage never occurs.
Present-day products also require present-day package checks. BatchData describes search and API or bulk delivery; ATTOM documents property, tax and valuation resources; First American offers reports and document access through API as well as research tools; PropertyRadar documents lists and eligible API integrations. BatchData, ATTOM, First American, PropertyRadar.
What we changed in the strategy
We stopped treating a provider's derived label as the local buying strategy. A shared source-record foundation can support different calculations, and no comparison above proves exclusive access to secret source facts. Local city records and obituary research can still supply coverage absent from national county data.
Our Source, Sift, Stack, Skip framework starts with neighborhoods containing houses the buyer would purchase. Split owner-occupants from landlords, build one motivation per list, then combine lists rather than require every motivation at once. Remove duplicates and do-not-mail records, refresh before each send and mail consistently. Homebase handles that direct-mail work in neighborhoods investors choose.
For your own evaluation, keep ordinary records and known exceptions separate, give vendors the same properties and define each field before scoring it. Record missing values, dates, unavailable reports and unresolved discrepancies. Then decide which delivery and research package supports your strategy, without mistaking agreement for truth or a preset for a plan.
Direct mail to the neighborhoods where you want to buy
Homebase handles direct mail to property owners in the neighborhoods where you want to buy. Enter one address to find your neighborhood and get your first neighborhood analysis free. No card required.
Analyze a neighborhood free