Simple Solutions for Powerful Impact: A Case Study
Sometimes, simple solutions are the most effective. When your budget is tight and your teams are small, leveraging technology can seem daunting or even impossible. But, in a world where everyone is looking for the next ChatGPT, at Mercy Corps we’re testing effective, low-cost technology solutions that still have a powerful impact.
In 2022–2023, the Mercy Corps Research and Learning (R&L) team wanted to test a straightforward theory: that anticipatory humanitarian action would reduce the negative effect a disaster has on households. In this case, they wanted to test whether households would experience less loss and damage from hurricanes if they received financial resources prior to the storm hitting their region.
Mercy Corps partnered with a trusted provider of digital financial services to pilot the Remittances for Anticipatory Action model in Central America. Through this model, the partner’s clients in the U.S. who regularly send remittances to the study area would receive an early warning message for a forecasted tropical cyclone along with a financial incentive to send. Remittance recipients in Central America could then take that money to prepare for the coming storm: to stock up on supplies, reinforce property, or even evacuate.
Testing this theory required monitoring the path of storms throughout the hurricane season (July-November). In 2022, R&L manually monitored forecasts. A diligent consultant followed a daily monitoring procedure based on the National Hurricane Center (NHC) dashboard, increasing monitoring to every 12 hours or fewer based on tropical cyclone activity. This process required 22.5 consultant working days over the course of the hurricane season.
For 2023, Mercy Corps’ Technology for Development (T4D) team developed an application called Anticipatory Action for the Americas (AAAStorms) that leveraged the NHC’s RSS feed to detect critical decision points. Using a lightweight data engineering strategy and AWS tools, we monitored the RSS feed and parsed text reports looking for key pieces of information such as countries or regions that have released watches and warnings. The R&L team developed criteria for two statuses: pre-trigger and trigger. If one of these statuses is detected, the system pings the WFP ADAM LiveMap API looking for additional information. Then, the information is packaged in an HTML-formatted email and sent to decision makers to review and take action.
Technical Solution
The AAAStorms application uses AWS Lambdas to run each stage of the ETL process. The functions include:
etl— calls the NOAA RSS GIS feed and parses the information; saves the raw information to an S3 bucketetlTriggers— retrieves the storm data and creates a trigger dataset based on the available information; saves the trigger dataset to an S3 bucketreport— retrieves the storm triggers and builds an HTML report using a Python Jinja template; the report function uses Amazon SES to send a formatted report to each email in theconfig.pyfile. The reports are saved to a final S3 bucket
The AAAStorms application is deployed to AWS using the serverless framework. The AAAStorms ETL and report are a single service. Each function is containerized using Docker and uses a function-level .yml file to support Lambda configuration.
The AAAStorms application is orchestrated using AWS Step Functions. It runs on two separate loops, a 12-hour loop to check for storms in pre-trigger status and a 6-hour loop to check for storms in a trigger status. These loops invoke slightly different state machines in AWS Step Functions. The 12H loop collects all storm data that is in trigger OR pre-trigger status. The 6H loop collects only storm data that is in trigger status. Once a storm leaves trigger or pre-trigger status, it is no longer reported.
Over the course of the hurricane season, the AAAStorms automated system saved hundreds of dollars in consultant fees and hours of personnel time. The system costs about $0.30/month, with the AWS Elastic Container Registry accounting for most costs ($0.10/month). This reduced the cost of monitoring over the course of the hurricane season from $9,000 to less than $10.
Conclusion
The model was active for two hurricane seasons (2022–2023) during which time no storm impacted the study area. With this limitation, the R&L team. did not have the opportunity to collect data to evaluate the model’s efficacy and the primary research question. However, we were able to capitalize on a unique learning opportunity in reducing the time and financial resources needed for future efforts of this kind.
Many nonprofit needs can be met through simple solutions that save time and effort. Automation of this kind provides an opportunity to leverage cloud services that are priced-for-scale to grapple dynamic data sources at low cost. Identifying opportunities for low-cost, low-lift applications such as AAAStorms can be a huge benefit to nonprofits with tight budgets and low bandwidth. Sometimes simple solutions are the most practical way to get information into the hands of decision makers.
Interested in our work? Reach out at dataforimpact@mercycorps.org
