Overture Member Summit 2026

In this talk, we share Uber’s approach to evaluating Overture’s global POI and address datasets from a production maps perspective.

We will describe the goals and scope of our evaluation, including the international markets we focused on and the practical questions we aimed to answer as a downstream consumer of open geospatial data. The talk will outline how we assessed net-new coverage, data quality, and enrichment attributes, with an emphasis on understanding where open data provides incremental value and where gaps remain.

A key focus will be our evaluation methodology. We will walk through a multi-pronged approach that combines standalone sampling, comparative analysis against existing datasets, and trip-based coverage headroom to better approximate real-world user impact. Rather than relying solely on raw feature counts, this framework prioritizes where missing or incorrect data most affects actual usage.

Finally, we will share high-level findings and lessons learned from the evaluation, including observations on strengths, limitations, and areas where closer collaboration between data producers and consumers could improve outcomes.

A summary of all the cool things we've been doing with Overture Data

Increase awareness of the value proposition of connected geospatial data and how companies are exploiting that capability.
As geospatial data becomes increasingly critical for organisations seeking to make informed decisions, data enrichment, augmenting internal data with curated third-party data, is fast emerging as a key strategy.
In this presentation, we will show how using unique and persistent IDs, such as GERS, and the PreciselyID, makes it easier to integrate different geospatial datasets. This allows organisations to use and consume a wealth of trusted data for additional insight, including address and property data, points of interest, streets, demographics, boundaries, natural hazards, and more. We will also discuss the common challenges that organisations face when onboarding data and explain how connected data solutions solve these issues with our connected data solutions.
We will share examples of how leading companies use these capabilities to transform their operations and make geospatial insights accessible to everyone using a combination of Overture Places and Address GERs along with the extended attribution that can be connected to from Precisely.

At the beginning of 2026 Overture kicked off a new task force with a mission of improving places data using imagery. Initially this work has been focused on using street level imagery to improve the geolocation accuracy of POIs, and in the process confirming their existence. This session will share the early benchmark results from this work; testing the accuracy of POI placements to real world survey measurements. The conflation of street level imagery with places opens the door to several follow on opportunities 1) accurate conflation of buildings and places, 2) seeding the discovery of new/changed places and 3) generating AI native places data. For the last point we'll introduce the concept of vector map embedding tiles: a data format that augments Overture place and geometry layers with semantic and visual embeddings, entrance and signage primitives, and lightweight visual-language descriptions derived from street-level imagery. The goal is to demonstrate how Overture datasets can be transformed into agent-ready assets, small and simple enough to be utilized by on-device spatial memory that is suitable for modern AI assistants.

Let's dig into the Overture Place Categories taxonomy, covering its history, evolution, and plans for ongoing governance. We'll explain how the taxonomy is organized, introduce the concept of "basic level categories", and describe some of the ways the taxonomy can be used.

The Overture Places theme is focused on points of interest (POI)--real-world entities such as hospitals, museums, parks, and schools. A core component of the Places theme is a taxonomy of about 2000 categories, organized into a hierarchy to facilitate exploration and discovery.

The taxonomy has been significantly updated in recent months to improve ease of use, provide greater interoperability between different systems and platforms, and optimize the accuracy and coverage of POI classifiers. In addition to overhauling the taxonomy, we introduced the notion of "basic level categories"--a subset of Place Categories used for generalization and industry alignment. There are a wide range of user-facing and backend use cases for the taxonomy, ranging from labeling and iconography in mapping interfaces to AI-assisted data ingestion and classification.

A deep dive on the confidence model developed for assessing the quality of every input record.

Modern maps cannot be static. Places open, close, change sentiment, and shift relevance constantly. A single data point can be useful, but it is rarely trustworthy on its own. The real value emerges when signals are observed over time and in aggregate.

This session explores how historical and regular signals help create a self-healing Places dataset. Weekly signal updates allow the system to continually validate place confidence, improve accuracy, and prevent failures like navigating to a business that closed last week.

We will ground this discussion in the Overture Maps Foundation approach to Places and Signals, with a focus on the Contributor Club. Attendees will learn how the system aggregates signals from multiple providers to strengthen place confidence while protecting proprietary data.

This session is designed for members and contributors who want to understand how data fits into the pipeline, how signals compound in value, and what we can do to make contribution easier and more impactful.

This session will give an overview about ongoing efforts to increase the quality of our Places matching pipeline. Key topics:
1. Labeling with LLMs - Evaluating human alignment
2. Blocking Strategy
3. Feature Engineering - New features under development

Large language models are improving quickly, but place understanding still breaks in predictable ways: unclear building/POI containment, conflicting addresses, and duplicate or near-duplicate POIs. Graphlication turns Overture themes (Buildings, Places, Addresses, Transportation) into a provenance-aware spatial knowledge graph and retrieves small, evidence-bearing subgraphs for a specific query so an LLM can answer with traceable support. Rather than forcing early "golden record" merges, Graphlication keeps relationships (containment, adjacency, access, candidate conflation) explicit and audit-friendly. The talk shares a repeatable architecture and practical patterns that teams can apply to ship grounded place Q&A, streamline conflation review, and run systematic quality checks.