Home / Coordinates and Location Formats

Geocoding in ArcGIS Pro: Build a Locator, Geocode a Table, and Rematch What Failed

Drone view of the street grid and rooftops of Puebla, Mexico

Photo: Mikhail Nilov / Pexels

Every address table hides a map. A spreadsheet of customers, clinics, inspections, or deliveries is a point layer waiting for its coordinates, and geocoding is the machinery that converts the words into the points. In ArcGIS Pro that machinery is deliberately explicit: instead of one opaque button, Pro gives you a locator, a persistent, inspectable object that encodes what reference data to match against and how, and then runs your table against it. That design is the whole lesson of this guide. A geocode is only as good as the locator behind it and the table in front of it, and Pro makes you build both properly. We will take the workflow in its real order: clean the table, build the locator with the Create Locator tool, geocode, and then do the unglamorous work that separates a professional result from a dotted guess, rematching what failed.

Interactive tool

From Table to Map: The ArcGIS Pro Geocoding Checklist

Check off each step as you finish it. Your progress stays on this page.

0 of 7 complete

0% done

Act Zero: The Table Decides the Match Rate

Locators get the glory, but tables decide outcomes. Before any tool opens, look at your address data the way the locator will. The strongest tables carry one field per address part: house number and street in their own column, city, state or province, and postal code each separate, because the locator's field mapping, the heart of Act One, works best when it can be told exactly which column is which. The weakest tables carry one merged column of everything, commas and all, which forces the locator to parse what you could have separated for free. Clean the obvious wounds now: postal codes stored as numbers that lost their leading zeros, street types spelled six ways, the row where the address is a PO box, which no street locator can place honestly, and the rows with no address at all. Note your expectations as you clean: a table of urban deliveries should match nearly completely, a table of rural routes and historic addresses will not, and knowing which kind of table you hold is what lets you judge the results in Act Three instead of merely receiving them. Our explainer what geocoding is covers the concept itself; this guide is the Pro workshop that follows from it.

Act One: Build the Locator With Create Locator

Open the Geocoding toolbox and run Create Locator. The tool asks for the decisions that define your match quality. First, the role: the locator type that describes your reference data, such as street address, point address, or postal code, chosen to match what your reference layer actually is. A locator built with the wrong role is a fluent liar, matching confidently against the wrong idea of an address. Second, the primary table: the reference feature class itself, your parcels, address points, or street centerlines. Third, field mapping: for each role field the locator needs, you point at the column in your reference data that supplies it, house number, street name, city, postal code, and the rest. Map every field deliberately rather than accepting guesses, because a swapped city and postal field produces a locator that fails in ways that look like data problems. Name the output locator like infrastructure, because it is: a good locator, built once from good reference data, serves every table your team geocodes this year. Save it with the project or in a shared folder, with its reference data version noted, so next season's geocode can be compared honestly with this one.

Drone view of the street grid and rooftops of Puebla, Mexico
Drone view of the street grid and rooftops of Puebla, Mexico

Act Two: Geocode Addresses, on a Copy

With the locator built, run Geocode Addresses: your table in, the locator against it, and a feature class out. Work on a copy of the table, always, because the tool writes match fields back into its output and your original is evidence. The output carries the fields that make Act Three possible: a status for every row, matched, unmatched, or tied, a score expressing how well the match fit, and the matched address the locator settled on, which is the field to actually read. A high match rate with low scores is not success; it is a locator being agreeable. Read the matched addresses against your input for the rows you know best, and look for systematic sins: a suburb folded into the neighboring city, a street type normalized wrongly, a whole postal district shifted. Geocoding errors are rarely random. They are patterns, and patterns have causes in the field mapping or the table, both of which you can still fix. If your addresses are in QGIS instead, our QGIS geocoding guide runs the same three acts with different tools.

Act Three: Rematch Addresses, Where Quality Is Decided

Every real table leaves a residue: the unmatched, who matched nothing, and the tied, who matched two or more candidates equally well. Rematch Addresses is the interactive tool built for them, and the time you spend in it is the difference between a geocode and a guess. Work the residue in order of value. Tied records first: the tool shows the candidates side by side, and a human who knows the territory resolves in seconds what no scoring rule can, this Maple Street, not that one. Then the unmatched, loosening one constraint at a time: correct the spelling the table got wrong, supply the postal code the row lacked, recognize the address that is a landmark rather than a street number and place it by hand. Some rows will stay unmatched because they deserve to: the demolished building, the address invented by a form at midnight. Leave them unmatched and documented rather than forcing them onto a map that will be believed. A geocode with an honest residue and a note explaining it is a professional product. A geocode at one hundred percent, achieved by lowering standards until everything matched something, is a liability with coordinates.

Composite Locators and the World Geocoding Service

Two extensions complete the picture. A composite locator chains several locators into one: your precise local address points first, street centerlines behind them, postal code centroids as the last resort. Geocode Addresses runs the table through the chain in order, so every record gets the best match it qualifies for and the fallback is explicit in the output fields. Build one when your reference data has layers of quality, which is to say, build one for almost every real jurisdiction. The alternative to building anything is Esri's World Geocoding Service, the hosted locator behind ArcGIS Online, which covers the world and removes the reference-data burden entirely, at the price of sending your addresses to a service and spending your organization's credits. It is the right answer for a one-off table in a country you hold no data for, and the wrong default for a recurring national workflow you could own. Either way, the three acts do not change: the table still decides the rate, the output still needs reading, and the residue still needs a human. Readers who arrived here chasing a blue dot for one address rather than a layer should detour to finding coordinates by address, the single-address version of this craft, and anyone whose points land in the wrong place entirely should check the map's own settings with our guide to finding the coordinate system in ArcGIS.

Maintaining the Locator: Reference Data Has a Birthday

A locator is a photograph of your reference data on the day you built it, and streets refuse to hold still. New subdivisions pour their addresses into the world monthly, roads are renamed, postal boundaries move, and every one of those changes is a future unmatched record blaming your table for the locator's age. Professional geocoding therefore treats the locator as infrastructure with a maintenance rhythm. Rebuild it when the reference data updates, and version it honestly: a locator named for its data's date lets next year's analyst discover that a match-rate change was the city growing, not the process rotting. Keep the field mapping documented beside the locator, because the colleague who rebuilds it in two years should inherit your decisions, not reverse-engineer them from the tool's memory. And re-run a known test table after every rebuild, a few hundred addresses whose correct answers you have verified by hand, and compare its results against the previous locator's. A rebuild that quietly worsens the test set is a regression wearing an update's clothes. This is the unglamorous truth the whole workflow leans on: geocoding quality is not a setting in a dialog. It is a habit of maintenance, applied to the table, the locator, and the residue alike, for as long as the addresses keep meaning something. Pro simply gives that habit excellent tools, and expects, as this guide has, that you will do the rest.

Composite Locators: When One Table Is Not Enough

The last tool in the kit answers a structural problem: real address collections are not one table. The modern streets live in one dataset, the historic parcels in another, the rural routes in a third, and no single locator covers the territory your addresses roam. A composite locator is Pro's answer: a locator of locators, consulted in an order you set, so the geocoding run tries the street locator first, falls back to the parcel locator for whatever the streets missed, and finishes with the rural routes. The geocoder experiences one tool; you have quietly built a cascade. Build composites when coverage genuinely fragments, and document the order as carefully as any field mapping, because the order is a policy: whichever locator answers first wins the record, and a permissive locator placed early will happily claim addresses a stricter one would have placed correctly. Test the cascade with records known to belong to each tier, and re-test it whenever a tier is rebuilt. Composites also retire gracefully: when the day comes that one unified reference dataset replaces the fragments, the single locator steps in and the cascade becomes a paragraph in the project history. Until that day, the composite is how honest coverage gaps are engineered around, openly, in the tool where everyone maintaining the process can see them.

Related Guides

Readers working through this topic usually need its closest companions too. These coordinate guides cover the neighboring ground:

Frequently Asked Questions

What is a locator in ArcGIS Pro?

A locator is the object that encodes how addresses should be matched: the reference data to match against, the role describing what that data represents, and the field mapping between them. You build it once with Create Locator and reuse it across tables.

Why did so many of my addresses come back unmatched?

Usually the table, not the tool: merged address fields, misspelled street names, missing postal codes, or addresses like PO boxes that no street locator can place. Clean the table, then work the residue in Rematch Addresses, loosening one problem at a time.

What does a tied match mean?

The locator found two or more candidates that fit the input equally well, often the same street name in neighboring towns. Rematch Addresses shows the candidates so you can pick the right one with local knowledge.

Should I use the World Geocoding Service or build my own locator?

Use the World Geocoding Service for one-off tables in places where you hold no reference data. Build your own locator, or a composite of several, for recurring work in your own jurisdiction, where you control the data and the match rules.

What is a composite locator?

A locator that chains several locators in order: precise address points first, then street centerlines, then postal centroids as a fallback. Every record gets the best match it qualifies for, and the output records which level matched it.

Photo credits: Mikhail Nilov (Pexels); Alan Vega (Pexels). Photos are served from this site; follow the credit to the library page for the original.