How RATA Measures Geocoding Accuracy
Published 18 August 2026. This page describes RATA's own published process and figures; it makes no claim about what any regulation requires beyond what is quoted and cited.
The short version. Every address submitted to RATA's GeoPlus service goes through a thirteen-step process against a dozen proprietary, commercial and online geocoding sources, validated against current Census Bureau TIGER boundaries. RATA publishes 95%+ accuracy on the resulting geocodes. That figure is RATA's own measured and published performance — Regulation C does not set a numeric geocoding-accuracy requirement, so no vendor's percentage is a regulatory floor. What separates a real methodology from an adjective is whether the vendor will say how the number is measured, and what happens to the addresses the automated sources cannot resolve.
There is no regulatory accuracy percentage
Worth stating plainly, because vendor marketing in this category — ours included, until this was corrected this week — sometimes implies otherwise. Regulation C requires that the census tract, county, state and MSA/MD fields on a HMDA LAR be correct, and examiners test the accuracy of those fields the same way they test any other reported data element. But no rule publishes a number a geocoding process must hit. “95%+ accuracy,” “99% match rate,” and similar figures across the industry are each vendor's own measured and published performance, not a cited regulatory minimum.
That distinction matters for an evaluation. If a vendor tells you their accuracy meets “the FFIEC standard,” ask what standard they mean and where it is published. The honest answer, ours included, is that the standard is the vendor's own.
Two different measurements, often quoted as one
The single most common way a geocoding figure misleads is by not specifying which of two different things it measures:
- Match rate — the share of submitted addresses that receive a geographic code at all.
- Accuracy — the share of those codes that are actually correct.
A geocoder can post a high match rate and a mediocre accuracy rate simultaneously, by returning a low-confidence best guess for every address rather than declining to answer. That is not necessarily dishonest, but it means a match-rate figure and an accuracy figure are not interchangeable, and a vendor quoting one as though it were the other — intentionally or not — is giving you less information than it sounds like. Ask which one you are being given, and ask how it is measured.
The thirteen-step process
RATA's GeoPlus service runs every address through a defined thirteen-step process against a dozen proprietary, commercial and online geocoding data sources, validated against current Census Bureau TIGER boundary files. In outline, grouped into what each stage is actually doing:
- Standardisation. The address is cleaned and formatted before any matching begins, using USPS CASS-certified software as well as manual formatting rules, so that inconsistent input (abbreviations, misspelled street types, missing unit numbers) does not cause an otherwise-findable address to fail.
- Multi-source matching. The standardised address is run against the full set of proprietary, commercial and online reference sources, rather than a single engine. Because no one source resolves every address, comparing several against each other is what lets one source's miss become another source's hit.
- Boundary validation. Results are checked against current Census Bureau TIGER geographic boundary files, so a match reflects the boundaries in effect for the filing year rather than a stale reference set.
- Exception routing. Addresses none of the automated sources can resolve with confidence are set aside for the next stage rather than defaulted to a low-confidence guess or left blank.
- Interactive resolution. A geocoding technician works the set-aside addresses individually — correcting formatting issues, checking alternate spellings and new construction, and cross-referencing against a map — to reach a result the automated pass could not.
The full published process is thirteen discrete steps; the five stages above are the functional groups they fall into, not a renumbering of the thirteen. Every address RATA geocodes, whether through the GeoPlus file service or the QuickPoint real-time API, goes through the same process — the difference between them is whether addresses are submitted as a batch file or one at a time.
Why the manual step is the actual differentiator
Automation handles the majority of addresses quickly and consistently, and that is not where geocoding services actually differ from each other. They differ in what happens to the remainder — the addresses that are new construction, have inconsistent formatting, sit on a boundary, or simply are not in any reference source yet. A fully automated service returns its best guess or no result for those addresses. GeoPlus routes them to a person.
That is also the practical reason a geocode error matters more than it looks like it should: an address geocoded one block in the wrong direction can land in a different census tract, which is a different set of demographics on the LAR and, if it changes the assessment area or MSA/MD, potentially a different filing. The manual step exists because that failure mode is common enough on the addresses automation cannot resolve that skipping it would show up in the accuracy figure.
What to ask any geocoding vendor, including us
- Is your published figure match rate or accuracy, and how do you measure each? If a vendor cannot separate the two, they may not be tracking them separately.
- What happens to the addresses your automated sources cannot resolve? A default low-confidence guess, a blank field, or a person working the exception. The answer changes what the headline accuracy figure is actually describing.
- How many independent sources does the process use, and does it compare them against each other or simply take the first result? A single-source geocoder, however well-tuned, has no way to catch its own errors the way a multi-source comparison does.
- Is the figure the vendor's own measurement, or attributed to a regulatory requirement? As above, there is no numeric floor in Regulation C. A vendor citing one as though there were should be asked to produce it.
Where this applies across RATA's geocoding services
The full thirteen-step process, at GeoPlus precision, underlies more than the standalone geocoding service. It is the same process behind the BISG race, ethnicity and sex proxies (which start from a GeoPlus geocode before applying Census Bureau surname and tract demographic data), and behind the QuickPoint API for loan origination systems that need the geography at application time rather than in a batch file afterward. The full service comparison covers the other tiers — Fully Automated Processing for portfolio analysis that does not require GeoPlus precision, and Market-Grade for marketing use where completeness matters less than cost — and states plainly which one each is built for.
Frequently Asked Questions
Does Regulation C require a specific geocoding accuracy percentage?
No. Regulation C requires the census tract, county, state and MSA/MD fields on the LAR to be correct, and examiners test those fields for accuracy the same as any other reported data element, but no rule publishes a numeric accuracy floor a vendor's process must hit. "95%+ accuracy" and similar figures published by geocoding vendors, including RATA, are the vendor's own measured and published performance, not a regulatory minimum. Ask any vendor citing a specific percentage whether it is their own figure or a cited regulatory requirement, because the two are frequently blurred.
What is the difference between geocoding match rate and geocoding accuracy?
Match rate is the share of submitted addresses that receive a geographic code at all. Accuracy is the share of those codes that are actually correct. A geocoder can have a high match rate and a mediocre accuracy rate by returning a low-confidence guess for every address rather than declining to answer, and a vendor can quote either number without specifying which one you are getting. Ask for both, and ask how each is measured.
Why does RATA use multiple geocoding data sources instead of one?
Because no single address-matching engine or reference file resolves every address correctly on its own; each has its own gaps and its own errors. Running an address through a dozen proprietary, commercial and online sources and comparing the results means one source's miss is frequently another source's hit, which is why RATA's GeoPlus service processes every record through multiple sources before deciding an address cannot be resolved automatically.
What happens to an address that automated geocoding cannot resolve?
In RATA's GeoPlus service, an address that the automated sources cannot resolve goes to a geocoding technician, who works it interactively -- correcting address formatting, checking alternate spellings, and cross-referencing against a map -- rather than the record being left blank or defaulted to a low-confidence guess. This manual step is the main reason GeoPlus differs from a fully automated geocoding service: automation handles the majority of records quickly, and a person handles the remainder.
This methodology description may be republished with attribution to RATA Associates and a link to rataassociates.com. If you believe a figure here is inconsistent with something else published on this site, tell us — that is exactly the kind of thing we would rather fix than be cited on incorrectly.
