Using AI to Free Up Time for Property Oversight: Summer Internship Report
This June, we welcomed our first-ever summer intern, Brenda Hernandez Rangel. A Portland State University finance major who is minoring in property management and working toward her real estate license, Brenda came to us through a partnership with PSU’s Center for Real Estate which aims to expose students to career paths in affordable housing. Brenda’s summer project sits at the intersection of her interests: training an AI agent to make life easier for asset managers who oversee the financial health of affordable housing properties.
The problem: manual data translation eats up time for real property oversight
Every month, asset managers receive hefty financial-report packets, sometimes more than 150 pages per property, from property management teams that handle day-to-day property operations. Asset managers need these reports to understand their properties’ current performance and future needs. But the information isn’t usable as-is. To unlock its value, an asset manager must comb through each packet, extract the data they want, manipulate it as needed, and manually key it into internal tracking and analysis tools.
This routine chore eats into the time asset managers have available to do what’s more important: use the data to catch property problems early and plan ahead. But translating data demands skilled judgment and attention, and automating the work is easier said than done. Our asset management consulting team started exploring potential technology solutions about four years ago, pulling together a cohort of nonprofit owners to advise our investigation and securing a capacity-building grant from Oregon Housing and Community Services to fund research and implementation. First step: testing financial-tracking software built for property managers. We found that these platforms tend to be costly, difficult to customize, and challenging to learn and hand off to new, untrained staff—deal-breakers for many owners.
Could a trained AI agent be a cheaper, better solution? We thought it could, and our advisory cohort agreed. Brenda set out to test this hypothesis.
“A human asset manager uses context clues and knowledge accumulated through job experience to make accurate inferences and judgements. Brenda’s task was to train an AI agent to do the same.”
Teaching an agent to think like an asset manager
Financial mapping is the process of designating exactly how each data point in one financial-tracking framework links to a different framework. At every organization, the map is a little different. Take rent concessions: depending on the organization and its accounting methods, funds used to entice new tenants to move in might be treated as property income or as a property expense. Payroll costs are sometimes split between maintenance and administrative costs, sometimes lumped together. A human asset manager uses context clues and knowledge accumulated through job experience to make accurate inferences and judgements. Brenda’s task was to train an AI agent to do the same.
She started small, building a mini-agent while getting hands-on experience with the work she was training the agent to do. She gave the agent prompts, telling it where to grab information and how to do quality-assurance checks. Working with teammate Maryam Alshaiji Hill, who is managing the project, she then moved on to training an agent built by Maryam to populate a full financial tracking workbook. She has since shifted her focus to customizing the agent's judgment to a specific organization. “The goal is to give the agent a prompt, upload the financial packet and tracking workbook, and let the agent do the rest,” Brenda said.
The team is now investigating which performs better, a general-purpose agent or organization-specific agents. Meantime, the project has grown beyond financial mapping. The team’s Occupancy Trend agent, a related tool, reviews rent rolls and vacancy reports across multiple months. Its job is to spot patterns, such as whether vacancies are clustering in a particular floor of a building or during a specific season, and detect whether the pattern is short-lived or ongoing. One organization in our advisory cohort is already using it to track how effective its rent concessions actually are.
“The agent is evolving with me.”
Brenda came into her internship having learned about balance sheets and income statements in the classroom. She didn’t anticipate the sheer density of real property financial reports or the prevalence of manual data entry in the field. "I was surprised at how tedious the work was," she said. "There was no technology to use instead of manual input."
Working closely with property financials deepened Brenda’s understanding of how to read financial reports and what to look for when doing property-level financial analysis. Training an AI agent to perform a specific task upped her general technology skills, while proving the adage, “The best way to learn is to teach.”
“Now that I’ve learned the foundations of property analysis, I can use the agent to understand the packets better,” Brenda said. “The agent is evolving with me.”
The promise of AI-assisted asset management
Every hour an asset manager doesn’t spend manually transcribing financial data is an hour they can use for oversight work that protects property health and resident wellbeing. But using AI to assist with asset management tasks isn’t just about giving time back to overstretched affordable housing professionals. AI agents like the ones Brenda and Maryam are training will ultimately be used to improve the quality of the information and insights asset managers use to monitor, forecast, and plan for their properties’ health. We're proud of Brenda and Maryam’s successful work to move this project forward and excited about where it goes from here.