In Minnesota’s Southwest Intermediate District 288, Superintendent Jeff Horton is piloting an AI-powered classroom observation system with the intent of changing that timeline. The system, called Evelyn, analyzes audio and video recordings of lessons to generate rubric-aligned feedback for administrators.
According to Horton, the concept emerged from an internal districtwide needs assessment, and aims to support faster, more detailed and intentional coaching conversations, in addition to reducing administrative workload and ultimately improving student outcomes.
“We started actually by doing listening sessions with our staff to understand what they were looking for and what we could do to help improve the system,” he said. “What we found out was that 69 percent of our staff were in their first three years in the district. ... We had historical turnover ranging from 25 percent and near 30 percent year after year for our staff. And we found that our administrators were having similar levels of turnover.”
That combination created what Horton saw as a coaching bottleneck: a huge need for supporting, coaching and guiding newer staff who, because they were in the probationary phases of their jobs, required multiple observations each year by state law — unmet because few administrators had time to consistently provide worthwhile feedback.
Traditional evaluation structures, he said, often fall short.
“Those who make the biggest impact in a child's life are those who are working directly on the ground in the classroom,” Horton said. “Unfortunately, we see that evaluations are nothing more than a compliance, check the list and move it on, for both the teacher and the evaluator ... they don’t truly become coaching conversations.”
According to Horton, delays in coaching sessions following an observation can take days, if not weeks, to facilitate, which he said compounds the issue.
“Say it’s a week before you get feedback, and now you’re trying to go back and have a detailed, reflective conversation," he said. "You know, that’s not fair to anybody involved in the process.”
Thus, Horton began exploring whether emerging AI tools could help close that gap. He said he partnered with tech collaborators outside the district to align the AI with internal evaluation needs. He said the district tested several options before settling on the Owl Labs meeting camera, which, according to Horton, has a panoramic 360-degree view that can see and hear everything happening in the classroom space.
The result of their collaboration was Evelyn, an AI-powered video analysis tool designed specifically to analyze classroom instruction and generate structured feedback. Once a video is uploaded to the system, Horton said, Evelyn processes the recording and compares what it sees against the Danielson Framework for Teaching, a research-based model for teacher evaluation. Horton said Evelyn can generate timestamped notes, direct quotes and suggestions for improvements to teacher practice.
The goal, Horton emphasized, is not to replace human evaluators but better support them.
“This is why I suggest that the administrator uses this as a tool rather than a replacement," he said. "Having the trained administrator in there who ultimately is going to do the final evaluation — they need to be there, too.”
During the current beta phase, the district has been comparing Evelyn’s AI-generated analyses with those of trained administrators. The results, on average, aligned extremely closely.
“When we ran the data across all evaluations, when averaged out, the difference was only seven one-hundredths of one point [0.07 percent] ... it was almost identical,” he said.
Horton also noted that the system could capture multiple student conversations at once — something a single observer cannot realistically do in real time.
“I can’t listen to 20 students talking at one time in collaborative groups,” he said. “This can, and it can see all of them.”
Still, the district encountered hiccups in some of Evelyn’s analyses. For example, Horton said early testing showed the tool scoring some special education classrooms more critically because it did not account for instructional accommodations. Those results prompted the district to do more work to account for those types of differences, he said.
Beta testing also surfaced questions about privacy and surveillance, which led him to seek legal counsel throughout Evelyn’s development process.
“We’ve had ongoing conversations with them to understand where the lines of privacy exist ... and what to do with that data,” he said, adding that classroom videos are treated as protected educational records.
Horton also pointed out that recording is common in other settings, though he acknowledged schools require a different level of sensitivity.
“If you’re an employee at Target or Best Buy ... I mean, you’re on camera all day long, right?” he said. “This isn’t new to the private sector ... but yeah, it has a different feel in education.”
Even so, Horton emphasized the importance of trust and communication with educators.
“This isn't an ‘I got you’ tool,” he said. “This is a, ‘Hey, I sincerely want to help you get better at what you do.’”
Looking ahead, Horton believes AI-assisted coaching reflects a broader shift in how schools support educators.
“If we're truly committed to putting our students first, we have to be willing to consider that maybe the way we've done things in the past isn't the way we need to do things in the future,” he said. “AI doesn't need to replace people. But we need to learn how to work in partnership with it.”
