Which attributes should buyers look for in a POI dataset?
The attributes that matter most in a point of interest (POI) dataset are a stable identifier, name and brand, coordinates and address, category, operating status, a freshness date, contact fields and source provenance. A long column list proves little. Ask for the fill rate of each field in your own countries and categories.
How to use this article
Read the checklist here, then use the linked hub and product pages for procurement citations.
Two POI files can hold the same number of rows and still be very different products. The difference sits in the attributes: which fields exist, how complete they are and how recently they were checked. This guide groups the fields a buyer should inspect. It uses the public schemas of OpenStreetMap and the Overture Maps Foundation as reference points, because anyone can read them. Neither schema is a standard that every provider follows, so treat them as a way to frame questions, not as a checklist a vendor must match field for field.
Which POI attributes matter most?
POI attributes fall into six groups: identity, location, classification, status and freshness, contact details, and delivery terms. The table lists what to check in each group and the warning sign to watch for.
POI attribute groups, what to check and the red flag for each, as of November 2026. Source: GSDSI editorial analysis of the public OpenStreetMap and Overture Maps schemas.
Attribute group
What to check
Red flag
Identity
Name, brand or chain, and an identifier that persists between releases.
Identifiers change every delivery, so files cannot be joined over time.
Location
Coordinates, address parts, and how places inside other places are linked.
One global address fill rate with no country breakdown.
Classification
Category depth, a published taxonomy and a rule for uncategorized records.
Categories that cannot be rolled up or mapped to your own.
Status and freshness
Operating status, a confidence score and a last-checked date per record.
Every record marked open, or no date fields at all.
Contact details
Phone, website, social links and opening hours, each with a fill rate.
Fields listed in the schema but empty in the sample.
Delivery terms
File format, refresh schedule, snapshot or change file, and license.
A license that does not name your intended use.
Identity fields: can you tell which place is which?
Identity fields say what a record is. Overture's place schema carries a primary name, optional common names, a brand and an identifier. Ask whether names are normalized or kept as submitted. The answer changes how well the file joins to your own.
Brand does work that a name cannot. The OpenStreetMap Wiki entry for the brand key says the tag can represent a common identity for individually owned stores. A brand field lets you count every location of one chain. On November 10, 2026, Taginfo counted 3,305,869 objects with a brand tag, against 116,750,153 with a name tag.
A stable identifier lets you join this month's file to last month's. Overture's Global Entity Reference System assigns each feature an identifier that is meant to stay the same across releases. Ask any provider whether its identifiers persist. Ask too what happens to them when two records merge.
Location fields: where exactly is the place?
Location fields start with coordinates. Overture's place schema allows one geometry type, a point. A point cannot say where a store begins and ends. If your use case needs a building outline, ask for a POI polygon and ask how it was drawn.
Address completeness varies widely. The OpenStreetMap Wiki's address guidance asks mappers for a street and a house number at minimum. City and postcode are treated as the wider locality. Overture's schema stores a freeform line, locality, postcode, region and country. Ask for address fill rates by country, because one global figure hides the weak markets.
Shared addresses need a rule. A mall and the stores inside it can appear as one record or as many. Ask whether a parent record sits in the file beside its tenants, and how you can tell them apart.
Category fields: how are places classified?
Category is the field most buyers filter on. Overture's places guide, read on November 10, 2026, describes a hierarchy of roughly 2,300 categories. Each record carries a primary category, the full path above it, optional alternates and a simpler basic category.
Depth matters because it saves mapping work. A hierarchy lets you roll restaurants up to food and drink, or drill down to one cuisine. OpenStreetMap takes a different route, with separate keys such as amenity, shop and tourism. Our recount, OpenStreetMap amenity tags are not businesses, shows why a category filter comes before any count.
Status and freshness fields: is the place still there?
Status and freshness fields usually decide whether a dataset is usable. Overture's schema has an operating status with three states: open, temporarily closed and permanently closed. The schema notes that the field says nothing about opening hours. A separate confidence score estimates whether the place exists at all.
Ask how a score is produced before you trust a threshold. Overture's guide says its confidence values mix provider signals with internal scoring. It adds that they are not calibrated to be comparable across providers. The guide advises treating the score as a relative filter, not a precise probability.
A last-checked date is the most practical staleness test. In OpenStreetMap, the check_date key records when the existence of an object was last checked. Taginfo counted 3,740,991 objects carrying it on November 10, 2026. Vacant premises can be marked with the disused prefix. Ask a provider for the equivalent fields, and for their fill rates.
Provenance belongs in the same group. Overture records the source dataset, an update time and a confidence value for each contributing source. Knowing where a record came from helps you judge how far to trust it.
Contact fields: phone, website and opening hours
Contact fields support enrichment, matching and verification. Overture's schema has lists for websites, social links, emails and phone numbers. Its guide says the matcher relies on names, addresses, websites and phones. Records missing them are more likely to persist as duplicates.
Open data shows how sparse contact fields can be. On November 10, 2026, Taginfo counted 4,873,055 objects with a website tag and 3,752,975 with a phone tag. Opening hours appeared on 4,923,948 objects, while 7,294,222 objects carried a shop tag. A website is often the quickest way to confirm that a business still exists.
Format and license: can you use the fields?
Format and license decide whether the attributes reach your own stack. Overture publishes its data as GeoParquet. GeoParquet is an incubating Open Geospatial Consortium standard that adds geospatial types to the Parquet file format. Confirm that a provider's format loads into your pipeline. Ask whether you receive full snapshots or changes only, and how the refresh cadence is set.
Licenses differ even between open datasets. OpenStreetMap data is under the Open Database License, which requires attribution and share-alike for adapted databases. Overture's places guide says its places data is published under CDLA Permissive 2.0 and Apache 2.0, depending on the source. The CDLA Permissive 2.0 text is short enough to read in full. For commercial files, the data licensing agreement red flags guide covers the terms to negotiate.
What to ask for before you buy a POI dataset
A short request list turns the six groups into evidence. Ask each provider for the same items.
The full schema, with a data dictionary entry for every field.
Fill rates for each attribute, broken down by the countries and categories you need.
How closed and low-confidence places are flagged or removed.
The last-checked or last-refreshed date for a sample of records.
How chains, malls and other parent and child places are represented.
The delivery format, the refresh schedule and the license terms.
Then run a sample extract for one city and one category you know well, and check it against the street. The POI data quality guide lists accuracy tests. The POI provider comparison guide covers how to weigh providers. A data dictionary is the first document to request.
Every source on this page was read on November 10, 2026. Research by Catarina Costa Carvalho.
Frequently Asked Questions
What fields should a POI dataset include?
At minimum, look for a stable identifier, a name, a brand or chain field, coordinates, a structured address, a category, an operating status and a date showing when the record was last checked. Contact fields such as phone and website help with matching and verification. Source provenance helps you judge trust.
What is a fill rate in POI data?
A fill rate is the share of records where a given field holds a value. A schema can list a phone or website column that is empty for most rows. Ask for fill rates per field, split by the countries and categories you plan to use, because a global average can hide weak markets.
Why does a stable identifier matter in a POI file?
A stable identifier lets you join one delivery to the next and see what was added, changed or closed. Without one, every refresh has to be matched again by name and address, which is slow and error-prone. Ask whether identifiers persist across releases and what happens when two records merge.
How can I tell whether a POI record is out of date?
Look for an operating status, a confidence score and a date showing when the record was last checked or refreshed. Then test a sample. Pick one city and one category you know well, and compare the file with what is on the street. Records with no date fields cannot be judged for staleness.
Does the license affect which POI attributes I can use?
Yes. A license sets what you may do with the data, including whether you can redistribute it, combine it with other sources or build products on it. Open datasets carry different licenses from one another, and commercial files add their own terms. Check that the license names your intended use before you rely on any field.
✓ Opt-Out Request Honored via Global Privacy Control
We use cookies and similar technologies to improve your experience. No marketing cookies are pre-selected. Learn more in our Privacy Policy. Your Privacy Choices.