CARE leverages data and Azure AI to prepare for global emergencies

Crises are compounding globally. Food insecurity and inflation are skyrocketing. Climate-borne disasters are increasing. In 2023, one in six people were estimated to have been exposed to conflict. The global development nonprofit CARE knows that women bear the brunt of crisis in impacts to their livelihoods, health, and safety. Yet women are also on the front lines of crisis response. After surveying more than 15,000 women in 15 countries, CARE found that 91% led or actively participated in community response. They provide meals and shelter, advocate for change, and offer health services—all to make communities safer.

CARE works in 109 countries worldwide to save lives, fight poverty, and advance equality. It reaches 167 million people a year to advance health, education, climate resilience, and more. It also responds to natural and human-caused crises, helping people find shelter, food, healthcare, and other vital services. CARE focuses on serving women and girls, who are often the most vulnerable and who also hold immense power to transform their communities.

While many emergencies cannot be foreseen, CARE strives to prepare for crises so staff in the field can react immediately. A quick response can lessen the human impact of a flood or armed conflict—and can save thousands of lives. To deepen its understanding of its emergency preparedness in high-risk countries, CARE teamed up with Microsoft Tech for Social Impact and Microsoft partner Valorem Reply. Together, they utilized Microsoft Azure OpenAI Service to build a sentiment analysis solution that evaluates staff survey responses at scale. Leadership can now use the insights the tool surfaces to inform decisions and gear up for any crisis.

"We have been blessed to have Microsoft as a partner,” says Cherian Varghese, Director of Data Science and Analytics at CARE. “With the greater awareness from the sentiment analysis tool, country offices can act on what they need to do to reach 100% emergency readiness.”

Identifying gaps in preparedness

As a nonprofit deeply committed to evidence-based interventions to promote social justice and battle poverty, CARE collects mountains of data. Comment fields and questions on surveys for staff and program participants alike gather nuggets of wisdom that complement quantitative data. “Qualitative data is rich in expressing the struggles and needs of staff and people in crises, but because of its volume and open-ended nature, it’s very hard to extract meaning out of it,” Cherian explains. Staff struggled to mine this data, leaving insights untapped.

CARE brought on Valorem Reply to understand this qualitative data with a sentiment analysis model, and Microsoft provided an Azure Innovate grant to support its development. “We have limited resources internally and not enough people with the skill set needed to build this proof of concept,” Cherian says. “With its expertise and resources, Valorem Reply was a great partner to see results fast.”

Within two months, CARE and Valorem Reply built and applied the sentiment analysis tool to process the data from open-ended comment fields within the global emergency readiness survey. They tested it with the survey’s mid-year check-in, hoping insights would enable country and global leadership to address gaps in readiness.

The tool uses Azure AI Services natural language processing (NLP) to map unstructured survey responses. It categorizes comments into positive and negative sentiment and extracts the most meaningful remarks. Azure AI Services’ translation capabilities enable the tool to correctly categorize and analyze comments in languages such as Spanish and French.

The CARE team spot-checked the AI results for accuracy, working with a Valorem Reply engineer to refine the model. The solution now classifies comments with the same accuracy as people would, but in a fraction of the time.

“There’s so much data that can help nonprofits to do more good, but they’re also struggling to understand that data while keeping operational costs down,” explains Nicholas Burton, a manager at Valorem Reply who worked on the sentiment analysis solution. “We came in with the technical capabilities to give CARE the tools to focus on their mission.”

Turning insights into action

CARE incorporated the Azure OpenAI Service sentiment analysis tool into a Power BI dashboard that displays the quantitative results of the emergency preparedness survey. Now in addition to seeing a country’s precise readiness percentage, leadership can read the top ten positive and negative comments from the readiness survey, pulled by the sentiment analysis tool. This qualitative data provides the context behind each country’s readiness score.

Positive comments enable CARE to replicate factors that improve emergency preparedness. For example, if a given country conducts virtual courses to ensure field staff are up to date on the latest emergency training, leadership can recommend that tactic to other countries.

Other comments surface ways to empower leadership to proactively address gaps. For example, remarks may reveal that a country office is struggling to meet readiness KPIs because of insufficient staff. Further, survey responses may indicate that staffing shortages are tied to inadequate funding, compensation concerns, or high turnover. "Now the insights are right there for country office and regional personnel to see and act on. They have the supporting data that tells the whole story,” Cherian says.

It used to take weeks or even a month for staff to wrangle survey responses, gain country office approval, and share results. “Country office staff are swamped with things to do, and there was no time for them to react to comments,” Cherian says. Now each step is automated, and the AI solution analyzes open-ended comments in about half an hour. “Using AI to derive insights in a timely manner helps us address problems more proactively. We’re therefore more ready to address a disaster or can even work to prevent an emergency.”

Cherian estimates that the people responsible for the emergency readiness reports used to spend 20% of their time on collecting and synthesizing that data. “Now they can use that saved time to serve people rather than chasing data,” he adds.

CARE is working to apply these benefits to other areas of its work, too. For example, Cherian hopes to use the sentiment analysis tool to understand unstructured data collected from program participants. The speed with which AI can analyze vast quantities of qualitative data could enable country offices to quickly pivot, even during emergencies, to better meet needs on the ground.

“In a crisis, we don’t have the luxury of time and resources, and data will help us identify the biggest needs,” Cherian says. “AI can help us leverage untapped data from millions of people. The opportunity is there—this is just the beginning.”

Find out more about CARE on X (Twitter), Facebook, and LinkedIn.

All images ©CARE / used with permission.