Lutheran Social Service of Minnesota unlocks the value of historical records using Power Automate and Azure OpenAI Service

Lutheran Social Service of Minnesota (LSS) is a nonprofit organization that provides human support services to a wide range of individuals, including youth and families, people with identified needs, and older adults. More than 2,400 team members support 91,000+ people annually, making LSS one of the state’s largest nonprofit social service organizations. The organization serves with a vision that all people should have the opportunity to live and work in a community with full and abundant lives. 

Established in 1865, LSS has a long and rich history in Minnesota. During this time, LSS has accumulated a huge archive of historical documents – including almost 100,000 files comprising millions of pages and spanning more than seven decades that this project covered. 

The challenge was that staff could not find the information they needed because files were not searchable or structured in any way. “People had to open many files to find just one answer. It took a long time, and often we still came up empty,” says Phao Thor, application support specialist at LSS. Microsoft Copilot estimated that it would take one person 28.5 years just to read through the entire archive. 

Intelligent text extraction and summarization 

To transform the data into a usable archive, LSS leveraged multiple Microsoft technologies, including Microsoft Power Automate, part of Power Platform, and Azure OpenAI Service. 

The work began with a document digitization solution, developed by the organization’s technology partner, RSM. To extract text from image files, the RSM team leveraged the pre-built, text recognition model in AI Builder, a capability that is directly integrated into Power Automate. 

The solution extracts readable text from a variety of scanned documents—handwritten notes, faded forms, even old microfilm. “The text recognition model in AI Builder worked better than anything we’d used before,” says Heidi Leach, senior director of Business Technology at LSS. 

Once the text extraction process is complete, a separate Power Automate script sends the raw text to a predict action in AI Builder, using a custom GPT-4.1 prompt, configured in an Azure AI Hub in Azure AI Foundry. This prompt corrects spelling errors, improves sentence readability, and converts charts into a readable format. The prompt also searches for predefined keywords in a file - with an emphasis on commonly-used fields such as name and parent names - and then links those keywords to the file. 

Along with creating and formatting the digitized files, RSM used Azure OpenAI Service to develop a model that creates clean, structured summaries from the millions of pages on file. “With the AI summarization layer, we not only extract text but capture the context, tone, and intent of the content. The AI model is also able to translate foreign-language documents, and surface key details we would’ve missed entirely,” says Leach. 

To optimize search even further, LSS wants to use Microsoft Copilot Studio to build an agent on top of SharePoint to enable natural language queries. “It took us just a few hours to get up to speed on Copilot Studio and refine our prompts to return exactly the data we needed,” says Thor. A prototype is currently in development. 

Reclaiming lost stories and gaining greater productivity 

With their new, AI-driven records management solution, LSS can now fulfill record requests much faster, reducing information discovery time and service request times from weeks to days. In addition, the organization's team members can generate accurate reports for government compliance purposes with minimal effort, transforming regulatory processes from very challenging to automatic. 

“We were finally able to start reclaiming the stories, decisions, and context that had been hidden for years,” says Leach. “This wasn’t just about preserving history. These files could include details about the people we support, decisions made around services provided, and program documentation.” 

LSS continues to expand and improve its archiving solution by refining prompts to improve accuracy. Also, along with the new agent in development in Copilot Studio, there are also plans for additional agents. This includes a document quality agent to further review scanned documents for legibility, completeness, and relevance before processing. The agent will be able to flag poor scans, missing pages, and irrelevant content and suggest re-scanning or routing to staff. 

In the meantime, team members are returning high praise for the new system. As Leach says, “Our AI-driven records system has changed the way our teams work, allowing them to spend less time scanning files or doing manual data entry and focus more on supporting people. That’s the real win.” 

Learn more about LSS on LinkedIn, Facebook, Instagram and YouTube