From Data to Decisions: AI Innovation Streamlines the Assessment Process

Assessing property values isn’t easy. Indiana has detailed guidelines for county assessors, but there’s something that still makes the process potentially inconsistent: it’s subjective.

Nexus LTD helps Indiana counties assess properties. With assistance from Rose-Hulman Ventures, Nexus found a way to improve assessment consistency with the help of artificial intelligence, and the state of Indiana awarded Nexus LTD an innovation voucher to support the partnership.

Assessors give real estate properties letter grades based on construction quality, with E-1 at the bottom, AAA at the top and two dozen other grades in between. Better quality means higher grades, which increases the assessed property value.

But Nexus LTD president Frank Kelly says assessors often disagree on a property’s grade. “Everybody has got an opinion,” he says. “The assessor’s office usually has multiple people out looking at properties, so it is not as standardized as one might like.”

And Kelly says assessment errors can be costly and confusing for property owners. “If one person calls a house a C and one calls it a B, that would have tax assessment ramifications,” he says.

Kelly thought if there were a way to make this process consistent, it would lead to more accurate assessments across the board. Then he thought about facial recognition software, and realized machine-learning technology might help with consistency. The idea, he says: “Here is a picture of a house — what other houses does this match up with?”

He contacted Ventures for help making the idea a reality. The result is a machine-learning model that receives pictures of a house being assessed, compares them with photos of already-graded houses, then outputs a suggested grade. Kelly says the model does this much faster than humans.

Ventures’ lead software developer and project manager Payden Beyer notes that there’s still an opportunity for human input. “They can override the AI,” he says. “A human can step in, since models aren’t perfect.”

The main challenge of creating these AI models was getting pictures of graded houses. “We trained on thousands of images,” Beyer says. “To take those images and figure out what the grades are is the hardest part.”

The property assessment process uses numerous criteria, Kelly says, including building material quality, ceiling height and level of upkeep. But he says this kind of computer model can be adapted to changes in criteria and standards. “If standards are different from county to county, software allows you to change that,” Kelly says. “If at some point this product goes to another state and their criteria are different, you can rename your database, and it will produce that type of result.”

Beyer says work on creating the models started in summer 2024 and was completed by February 2025. The next step is to create a website allowing access for Nexus employees.

Using Ventures for this project helped keep costs low. “We have hardware on campus we can leverage,” Beyer says. “Other projects where we have used non-onsite resources cost two to three times as much to train.”

“They are really easy to work with, with readily available staff,” Kelly says of Rose-Hulman Ventures. “Me going out to hire one programmer to do this might produce something great or might not produce anything. Working with a group, I am trusting the project manager to determine who is best suited for this kind of analysis.”

As project manager, Beyer is there to teach the interns, but he tries to let them lead as much as possible. After they’ve gotten used to the project, he only steers them when necessary. “The students have become really independent, good at working together, looking at the project, seeing the whole project, seeing what the next task is,” he says.

Kelly says the project is nearing the end of development and hopes to have the website available within six to nine months.