Whether searching for dinner places nearby on a mobile phone, tagging a social post to a location, or selecting a suggested location as the destination in a rideshare app - today’s consumers are constantly utilizing local search functionality. Also known as “discover near me” search, this capability has become an integral part of daily life, helping individuals discover and navigate to nearby places of interest - from restaurants, to flower shops, to hospitals, and more.
At its core, local search helps consumers perform daily tasks more efficiently with the assistance of location-based applications, enhancing the user experience by providing relevant and reliable information about nearby establishments or places. But building a reliable local search application is not always straightforward; the data powering both the geographic and categorical elements of local search is complex to create and maintain in a dynamically changing world.
To help product managers and application developers understand these complexities, we’ve created this ultimate guide to local search and the points of interest (POI) data needed to make a complete, accurate, and up-to-date tool that users can trust. The information in this guide will leave you feeling like an expert in local search and empower you to choose a data provider with clear evaluation criteria in mind.
In this guide, we’ll cover:
It’s important to note that the success of local search relies heavily on the accuracy of the information provided. A high level of accuracy contributes to a seamless user experience (UX), allowing individuals to make informed decisions about where to dine, shop, or visit.
On the other hand, low accuracy can lead to a disappointing experience that erodes user trust - like recommending dissolved businesses or providing directions to a place that never existed in the first place. Maintaining accuracy is not just a matter of user satisfaction; it is essential to building brand credibility and ensuring success in an increasingly competitive market. No matter the ultimate use case, accurate local search is critical to the user experience.
Imagine a person arriving in New York City for the first time. They exit Grand Central Station after a long day of travel and want their favorite Chipotle meal, so they open their rideshare app to call a car to Chipotle. The intuitive expectation is that when searching for a Chipotle destination, the prepopulated POIs listed by the app are closest in proximity to their current location. This user likely also expects the locations listed are open. However, when they are on their ride they pass two open Chipotle stores, and find that the one they eventually arrive at has already closed.
This poor UX not only disrupts the consumer’s day, but also reflects poorly on the rideshare app and its functionality. In today’s hyper-connected world, users expect reliable local search tools that do not disappoint, and this experience would be enough to lose this user to a competitive app. To avoid such user churn, it’s important to choose a robust and precise POI dataset to power a local search engine.
Accuracy in local search presents unique challenges for product managers or application developers. It involves trust in capturing real places, ensuring completeness, and validating that attributes such as store hours, web address, phone number, street address, and coordinates reflect the latest available information.
The importance of these metrics vary based on the specific use case for local search. For example, navigation companies prioritize precision, ensuring coordinates are correct for accurate directions. On the other hand, search providers optimize for recall, developing comprehensive listings with accurate attributes like phone numbers and websites that enhance user engagement, increasing the likelihood of consumers using the service again. In reality, the best local search functions have just the right mix of precision and recall.
However, pursuing both precision and recall can sometimes be counterproductive, so it’s important to recognize the natural trade-off between quantity and quality. By addressing and understanding these complexities, local search providers can confidently prioritize the right metric for their use case. In doing so, they can deliver reliable and positive experiences that specifically meet their users’ expectations for exactly what they are trying to accomplish.
In the simplest terms, precision refers to whether or not something is true. For a data product, that means making sure the entries included in the dataset are as correct as possible. In local search, precision can be examined from two dimensions: row precision (the existence of a place) and column precision (the facts about places that exist).
Row precision means having confidence in the authenticity and accuracy of each entry in the dataset representing a place in the physical world with relevance to users. Fundamentally, row precision addresses the question of whether a record in a dataset should even exist. It helps determine if the information presented relates to an actual place that users can interact with in the physical world, or if it will contribute to a poor UX.
While this may seem straightforward, the neverending volume of digital information means that data that was once accurate may no longer be correct or relevant. Additionally, the age of user-generated data, online reviews, and e-commerce has introduced the challenge of distinguishing between real and fake information on a scale never seen before. This is particularly an issue for crowdsourced or open source data.
For instance, let's consider a boutique jewelry business. While it may have a website and a phone number, it could actually be a small business operating on Etsy and run from a private residence. In this case, it doesn't function as a physical storefront where visitors are welcomed and therefore shouldn't be considered a local boutique in a given area. It also should not be displayed as a potential destination for a navigation engine, or a potential competitive location for another brand performing site selection analysis.
Ensuring that each row in a POI dataset corresponds to a real and currently operational place lays the foundation for accurate and predictable local search experiences. Once row precision has been established, additional accuracy metrics can be evaluated - like column precision.
Column precision refers to the accuracy of specific attributes in a given row. This involves evaluating the truth of information that allows users to engage with a place, such as phone numbers, open hours, and website URLs. However, it’s important to note that column precision builds upon the foundation of row precision. A row must be identified as a real and current place before assessing the accuracy of the data included for that record.
It's also essential not to mistake a dataset’s fill rate as a measure of column precision. For example, merely populating every phone number field with a brand’s corporate headquarters’ number may achieve a 100% fill rate, but does not guarantee accuracy or provide any value for users.
Ensuring high column precision increases the reliability of local search results, encompassing not only the comprehensive list of prospective places, but also the associated metadata that enhances the UX. This enables users to confidently rely on accurate phone numbers when they call a business to book an appointment, trust that a store will be open upon their rideshare’s arrival, and seamlessly follow a website link to learn more about their potential destination. The reliability of these interactions enhances the user’s overall engagement while exploring the places they seek, and thus translates into increased satisfaction with the product or application being used to perform local search.
Local search creators must also factor in recall. Recall is the ability to find all relevant records within a dataset. In relation to POI data, recall refers to the confidence that all possible places meeting a certain criteria are represented in the dataset. For example, a POI dataset with perfect recall would include all possible restaurants, hospitals, national parks, etc. in a given geography.
Let’s say a consumer wants to know all of their options for selecting a top restaurant nearby. How often is it that a local search function will return ALL possible restaurants - even the one that just opened yesterday? This example helps highlight the difficulty in attaining perfect recall. Points of interest open and close every week, making perfect recall a constantly moving target. Although it is difficult to attain, users expect high recall when searching for places nearby. To meet these expectations and deliver a strong UX, local search products require a frequently updated POI dataset as an input.
Even with up-to-date POI data, recall is difficult to measure. It requires a truth set to compare against, and in practice, none exists. It’s virtually impossible to know, at any given moment, every single place that exists in the world. But innovations in data science and curation enable local search providers to get as close to a truth set as possible when building their products and applications.
Due to these difficulties in measuring recall, it is important to beware of overreaching when sourcing or creating a POI dataset. Just like it is possible to have less than 100% of possible POI in a given dataset, it is also possible to have over 100%. Extraneous entries only dirty the dataset, leading to inaccurate local search results and a negative UX. It is important to consider the trade-offs that occur when more value is put on recall over precision, and vice versa.
For the purpose of local search, precision can be thought of as quality, whereas recall represents quantity. As with anything with a delicate balance, by greatly improving one metric, substantial pressure is put on the other. For example, the more data you have, the harder it is to ensure all of the data is precise. This is true of both row and column precision in relation to recall.
It is critical to consider whether recall or precision will provide the most value for a particular local search use case. To understand this, consider whether it is more important to ensure all possible results are returned in a search, or that all results returned in a search are real and up-to-date. Or alternatively, determine which bad UX (poor precision or poor recall) would be worse. For example, would it be worse for a user to navigate to a closed POI, or not find one in the first place?
Any good dataset attempts to strike a balance between quality and quantity. But deciding which holds higher value for your use case will determine the priority of recall versus precision as you evaluate data providers.
Technology companies in the consumer mapping, social discovery, and navigation spaces rely on local search to power their platforms and improve their user experience. Each of these groups uses local search for slightly different purposes, which we explore below.
Consumer mapping companies make it easy for people to find points of interest by enriching user GPS locations with highly precise POI data.
Today’s most commonly used consumer mapping applications and products include:
Companies with social discovery applications and products have also begun to invest in local search to create highly personalized and interactive features for their users. Although they may not be focused on local search as their main offering, it has become an influential feature on many platforms.
Examples of social discovery brands leveraging local search include:
Local search is also an integral part of products and applications that help consumers navigate around the world. Many of the key players in today’s navigation space are taxi and rideshare apps, or companies that develop technology to power them.
Popular examples of modern navigation companies who leverage local search functionality include:
To meet this demand for data to power these products and applications, companies that specialize in mapping and geospatial data creation often develop POI datasets specifically for local search. You can see a full list of POI data providers here, but some popular examples include:
At SafeGraph, we understand the unique challenges faced by providers wanting to deliver accurate local search results. We are committed to being the trusted data partner for product managers and application developers, providing reliable and up-to-date information about our dynamically changing physical world.
To ensure we provide as close to a truth set as possible to our clients, we maintain a maniacal focus on both precision and recall through a mix of the following:
SafeGraph performs regular reviews and updates of our Places data, constantly recompiling information to ensure the ongoing accuracy of our monthly releases. This process crosschecks numerous sources and helps identify any outliers or outdated information that may have crept into the database.
To validate data accuracy, SafeGraph conducts manual reviews of sampled data. This hands-on approach helps identify potential errors or inconsistencies that automated processes might overlook. These manual validations then fuel our row and column precision models, and we consistently increase our precision as the training data continues to accumulate.
SafeGraph keeps a close eye on business openings and closings, ensuring that our database reflects the most up-to-date information. By promptly updating our data, we ensure that local search functions can minimize the risk of recommending closed establishments to users. For more information, check out our blog on the importance of timely open and close metadata.
POI data is inherently messy, and any provider that pretends it isn’t is not being honest with their customers. SafeGraph is committed to being transparent with our users about the Places data we create. Each month, we publish release notes and accuracy metrics that help product managers and developers understand both precision and recall so they can make informed decisions for local search.
We value customer input and feedback, and actively engage with local search providers to address any concerns, suggestions, or requests. This customer-centric approach allows us to improve data accuracy based on real-world usage and feedback, and it also helps narrow our focus on improving what is most meaningful to our customer’s customers.
Precision and recall are not just “nice to haves,” but fundamental requirements to local search capabilities. As a data partner, SafeGraph recognizes the importance of striking the right balance between quantity and quality. Get in touch with our team today to see how the right POI data can enhance your local search product, elevate the user experience, and accelerate growth.