Generative AI for Geospatial Insight at a leading US telco

Agentic AI | Retrieval-augmented generation | Geospatial analytics

A major US telecoms provider needed a clearer view of how its fibre products compared with those of its competitors across the country. We built a map-based web application with a generative AI capability to query geospatial competitor insights. Analysts can now explore complex geospatial data and ask questions about it in plain English.

The Challenge

The client already held the data it needed to understand its competitive position. That data was spread across several large, complex geospatial datasets covering competitor speeds and coverage, demographics and population density. Getting answers depended on specialist GIS and data skills, so the insight was slow to reach the people who needed it. 

The client wanted an easy way for employees to visualise the competitive landscape, explore it by location and ask questions of it. The insight would feed marketing decisions and help identify where to upgrade the network next.

Our approach

We combined the competitor, coverage, demographic and population density datasets in a geospatial database, so analysts could run location-based queries efficiently across all of them together. The web application shows competitor fibre speeds as interactive map layers. Users can zoom from national down to local level, switch layers on and off, and spot gaps and opportunities straight away, without the delays that come with loading large raw datasets.

An Agentic AI setup was integrated with the data through a retrieval-augmented generation pipeline. Descriptions of each dataset, its fields and its meaning were turned into embeddings and stored for vector search. When an analyst asks a question, the system finds the most relevant datasets and passes that context to the model. Each answer is grounded in the underlying data, so analysts no longer need to write queries or understand how the data is structured.

The Agent is geospatially aware, with the current map area passed to the model with each question, so a question such as "how do we compare here?" is answered for that specific area. The interface was designed so people with different roles and levels of data expertise can ask their own questions. Executives can get quick summaries, while planners and engineers can drill into the detail. That opened the insight up beyond specialist data teams.

The impact

• Faster competitive insight: analysts can explore large volumes of geospatial data and get summarised answers in moments rather than days.
• Insight for more people: employees at every level of data expertise can explore and question the data directly.
• Better-informed investment: a clearer view of competitor strength by location supports marketing and network upgrade decisions.
• A platform to build on: a production solution in the client's own environment that is ready to take more data and advanced analytics as needs evolve.