Case Study 1: The School That Gave Every Student an AI Tutor

The Vision

In the fall of 2023, a mid-sized school district in a southeastern U.S. state — let us call it the Greenfield Unified School District — launched an ambitious initiative. With funding from a combination of federal pandemic-recovery grants and a private foundation, Greenfield deployed an adaptive AI tutoring platform across all 12 of its elementary and middle schools. Every student in grades 3 through 8 would have access to an AI-powered math tutor, available 24 hours a day, seven days a week, on school-issued tablets.

The superintendent, Dr. Angela Torres, described the initiative as "the most significant investment in personalized learning this district has ever made." The reasoning was compelling: Greenfield's students had experienced significant learning loss during the pandemic, with math proficiency rates dropping 18 percentage points between 2019 and 2022. The district faced a teacher shortage — nine unfilled math teaching positions across the district — and the remaining teachers were stretched thin.

The AI platform, developed by an educational technology company, offered adaptive math instruction: it assessed each student's current level, generated practice problems at the appropriate difficulty, provided immediate feedback, and adjusted the learning path based on performance. The company's marketing materials cited a peer-reviewed study showing "a 27% improvement in math scores among students using the platform for 30 or more minutes per day over one school year."

The Implementation

Greenfield rolled out the platform over a single semester. Every student received a school-issued tablet. Training sessions for teachers were offered — two 90-minute workshops during pre-service week. The platform was designed to be used for 30 minutes per day during math class, with teachers circulating to help students who were stuck.

The implementation plan sounded reasonable. The reality was more complicated.

What Went Right

Engagement in early weeks. Students were initially excited about the tablets and the gamified elements of the platform. Usage rates were high in the first month, averaging 28 minutes per day across the district.

Data for teachers. Teachers who actively used the platform's dashboard — which showed each student's progress, common errors, and time spent — reported that it helped them identify struggling students more quickly. "I could see that seven kids were stuck on the same concept before they even raised their hands," said one fourth-grade teacher.

Access outside school hours. Some students used the platform at home to practice additional problems. For students with engaged parents, this extended the learning time. One mother reported that her daughter "actually asked to do math practice before bed."

What Went Wrong

The training gap. Two 90-minute workshops were not sufficient to prepare teachers for a fundamental shift in their instructional approach. Many teachers reported feeling uncertain about how to integrate the platform into their lesson plans. Some simply gave students 30 minutes on the tablets and then taught their usual lesson as if the platform did not exist — undermining the intended synergy between AI-assisted practice and teacher-led instruction.

The engagement cliff. After the initial novelty wore off (approximately 4–6 weeks), daily usage dropped to an average of 19 minutes, then 14, then 11. By mid-semester, many students were going through the motions — clicking through problems quickly, guessing rather than thinking. The platform's adaptive algorithm struggled to distinguish between "the student does not understand this concept" and "the student is clicking randomly to pass the time."

The equity problem. The students who benefited most from the platform were students who were already performing near grade level — students who had strong enough foundational skills to engage meaningfully with adaptive practice. Students who were two or more grade levels behind found the platform frustrating: even at its lowest difficulty setting, the problems assumed knowledge they did not have. These students needed fundamentally different instruction — not just easier problems from the same sequence.

At home, the pattern was even starker. Students in households with reliable internet, quiet spaces, and supportive parents extended their learning time. Students without these advantages did not — and some reported that the take-home tablets became a source of conflict (siblings competing for device access, parents concerned about screen time, spotty internet dropping connections mid-session).

The teacher shortage was not solved. The platform was partly intended to compensate for the nine unfilled teaching positions. But AI-assisted instruction does not replace a teacher. In classrooms with substitute teachers or large class sizes, the platform was used as a babysitter — students sat with their tablets while an overwhelmed adult tried to maintain order. The AI provided practice problems. It did not provide motivation, discipline, relationship, or the dozens of other things a teacher does that have nothing to do with content delivery.

The assessment results. At the end of the school year, Greenfield administered its standardized math assessments. The results were mixed:

• Students in grades 3–5 showed a modest improvement (approximately 4 percentage points) compared to the previous year. However, this improvement was not significantly different from the improvement seen in a neighboring district that did not use the platform but had filled its teaching positions.
• Students in grades 6–8 showed no significant improvement.
• The achievement gap between high-performing and low-performing students widened by 3 percentage points.

The Aftermath

Dr. Torres described the results as "humbling." In a candid interview with the local newspaper, she said: "We invested $2.3 million in this platform. I believe technology has a role to play in education. But I have learned that technology without adequate teacher support, teacher training, and attention to equity can widen the very gaps it is supposed to close."

The district decided to continue using the platform but with significant changes:

1. Teachers would receive 40 hours of training (up from 3 hours) spread across the school year, including classroom coaching.
2. The platform would be used as a supplement to teacher-led instruction, not a substitute for it — and never in classrooms without a certified math teacher.
3. Students more than two grade levels behind would receive small-group human tutoring rather than AI-assisted independent practice.
4. Usage data would be reviewed monthly by a team that includes teachers, administrators, and a data analyst.

Discussion Questions

1. The Promise vs. the Evidence: The platform company cited a peer-reviewed study showing a 27% improvement in math scores. Greenfield's results were much more modest. What factors might explain the gap between the company's study and Greenfield's experience?

2. The Training Problem: Teachers received only 3 hours of training. How might insufficient teacher training undermine even the best educational technology? What should effective training for AI-assisted instruction include?

3. The Equity Paradox: The platform was intended to help all students, but the students who benefited most were those already near grade level. Why did this happen? How could the implementation have been designed to better serve struggling students?

4. The Engagement Cliff: Usage dropped dramatically after the initial novelty wore off. What does this tell us about the limits of gamification and technology-driven motivation? How does intrinsic motivation differ from the kind of engagement that technology can generate?

5. The Teacher Shortage Question: Greenfield partly adopted the platform to compensate for unfilled teaching positions. Is it ethical to use AI as a response to teacher shortages rather than addressing the root causes (salary, working conditions, respect for the profession)? Who benefits from framing AI as a solution to staffing problems?

6. **The $2.3 Million Question:** If you were Dr. Torres, how would you have spent $2.3 million to improve math outcomes? Would the AI platform be part of your plan? What other investments would you prioritize?

7. Connection to Your Experience: Have you ever used an adaptive learning platform? If so, did your experience resemble Student A's or Student B's from Section 16.5? What made the difference?