From Rural Villages to Cities, Bengaluru Firm Latlong Bets Big on Better Geocoding Accuracy

In a major boost for India's homegrown geospatial technology ecosystem, Bengaluru based maps platform Latlong has announced that its Geocoding API is now four times more accurate than Google Maps for Indian addresses, following extensive in house research and development.

In 2023, a study verified by IIT Kanpur's National Centre for Geodesy found that Latlong's API was already twice as accurate as Google Maps for Indian addresses. The company now says the accuracy gap has widened further, especially outside major cities.

Why Geocoding Accuracy Matters

Geocoding, the process of converting addresses into geographic coordinates, is critical for services ranging from deliveries and emergency response to elections and e-commerce. In India, where addresses often depend on landmarks rather than street numbers, achieving consistent accuracy remains a challenge.

Latlong says it has rethought its approach to precision, drawing inspiration from India's long tradition of astronomical measurement.

The New Benchmark

The company tested its system on a much larger and more diverse dataset, with a strong focus on rural addresses. According to the results, Latlong's average geocoding error is 297 metres, compared to 1,280 metres for Google Maps. On the original 2023 dataset, Latlong recorded an average error of 231 metres, while Google Maps showed 1,067 metres.

Latlong also said 58 percent of test addresses were resolved within 100 metres, a key threshold for applications that require high location accuracy.

What Changed?

Latlong's engineering team rebuilt key parts of the system, acknowledging that earlier models had begun to overfit certain address patterns.

Three upgrades stood out.

First, the company revamped its address parsing and tokenisation approach, shifting to a method that focuses on maximum area matches through polygon chains. This allows the system to better handle incomplete or loosely structured addresses while narrowing down likely locations.

Second, the algorithm was redesigned to handle ambiguous address boundaries. In many cases, Indian addresses reference well known localities or cities for postal clarity even when the actual location lies outside official limits. Latlong said its updated system directly models this ambiguity, improving results in peri urban and fast growing regions.

Third, the company significantly expanded its data coverage. Latlong has more than doubled its Points of Interest database and added granular geographic data covering nearly all of India's 6.6 lakh villages, along with thousands of urban local bodies. This has improved accuracy across metros, towns and rural areas.

A Different Approach

Unlike platforms that adapt search or autocomplete systems for geocoding, Latlong treats geocoding as a spatial matching problem. The company argues that rearranging words in an address should not change the final location output, a limitation often seen in search driven systems.

Shreyas Bharadwaj, Senior Vice President at Varahe Analytics, one of Latlong's largest enterprise users, said the API has enabled hyper local targeting for campaign communication.

"Latlong's geocoding and location APIs have transformed how we identify and engage specific voter clusters with issue based messaging that truly resonates," he said.

Latlong's progress is notable given its scale. The company operates with a 30 member team based in south Bengaluru, competing with global platforms that have far larger resources. Latlong said its focused investment in research and development positions it to keep improving accuracy while offering cost effective solutions.