​​​Executive Summary

Grupo TX implemented a Generative AI solution on an Amazon Web Services (AWS) serverless cloud infrastructure for IDAAN as part of a larger digital transformation initiative aimed at increasing revenue. The project specifi cally addressed critical defi ciencies in IDAAN's customer database that directly impacted billing capability and revenue collection. The existing database suff ered from three major problems: missing customer information resulting in unbilled water consumption, outdated records containing non-existent customers and locations, and lack of data on unauthorized connections and water theft. These issues collectively prevented eff ective revenue management and accurate service delivery.

Field teams conducted an extensive census to physically identify and document water meters and infrastructure throughout service areas. The Generative AI component was instrumental in addressing the slow, tedious data standardization process. Grupo TX developed a specialized AI agent that processed fi eld information captured daily written in colloquial Spanish, including location descriptions resembling verbal directions. The agent successfully standardized variations in street naming conventions (such as "street" vs. "st." vs. "avenue") while maintaining proper syntax in the geodatabase.

​​

​

​

Challenge

Instituto de Acueductos y Alcantarillados Nacionales (IDAAN), Panama's national water and sewerage authority, faced signifi cant challenges in their billing capabilities that directly impacted revenue collection. The organization's customer database suff ered from three critical defi ciencies: 
● Missing customer information resulting in unbilled water consumption
● Outdated records containing non-existent customers and locations
● Lack of data on unauthorized connections and water theft

These database issues prevented eff ective revenue management and accurate service delivery, creating a substantial gap between water produced and water billed. The challenge was particularly complex due to fi eld data being collected in colloquial Spanish with inconsistent address formats and verbal-style location descriptions, making standardization nearly impossible through conventional means.

​​

​

​

​

Solution​

​

Grupo TX designed a comprehensive Generative AI solution as part of a larger digital transformation initiative. The initial step involved establishing an ESRI Geodatabase to manage customer and infrastructure data. Subsequently, fi eld teams performed a comprehensive physical census to record water meter and infrastructure details in the fi eld. Finally, a specialized AI agent, built on AWS's serverless architecture, was created to process the collected fi eld data.

 

​

The solution leveraged a multi-stage AWS architecture:

● CSV fi le upload via API Gateway for raw fi eld data
● ETL transformation through Lambda functions
● DynamoDB for data storage with stream processing
● SQS queue implementation to prevent data loss
● Amazon Bedrock integration with a custom LLM agent designed to:

• Process fi eld information written in colloquial Spanish

• Standardize variations in street naming conventions

• Interpret location descriptions resembling verbal directions

• Normalize information into standardized ESRI GDB format

 

Advanced prompt engineering techniques including few-shot prompting and chain-of-thought prompting were applied to optimize the AI's performance in handling the complex linguistic variations in Panamanian address data.

​

​​

​

Results

​​

The implementation of Grupo TX's Generative AI solution resulted in a reduction of address standardization processing time of more than 90%; the process, which was previously handled manually by a team of quality assurance specialists, became completely automated.
 

This processing time reduction resulted in a diffi cult to measure cost optimization due to quality assurance specialists being able to focus on more important tasks and, in turn, improved customer satisfaction.​​​

​