ASSISTments is rated Strong by Evidence for ESSA, based on two qualifying randomized studies: a Maine study finding +0.18 SD on seventh-grade math, and a North Carolina replication finding +0.10 SD on eighth-grade math, measured one year after the intervention period ended, averaging +0.13 SD across the two studies. Both were conducted under real school conditions.
For context, a 2021 meta-analysis of 33 ALEKS studies found a negligible overall effect of g = 0.05 when ALEKS replaced instruction entirely, versus a meaningfully stronger g = 0.43 when used as a supplement to teacher-led instruction. This reinforces that implementation mode rather than the platform alone drives outcomes. Alpha School reports 2.6 times faster student growth on nationally normed MAP tests, based on internal analysis interpreted by AI tools rather than independent review, and a charter application by Alpha-affiliated Unbound Academy was denied by Pennsylvania state education authorities. As a contrast case, Alpha School is high-visibility and high-ambition, but not currently comparable on evidence.
A consistent pattern holds across the sector. Access does not equal use, and use does not equal impact. Khan Academy’s own reporting acknowledges that only around 15% of students with access to Khanmigo engage with it, despite more than 108 million cumulative interactions since launch. Michigan State University implementation data from its UIA Learning Innovation Project found that undergraduate students had nearly twice as many Khanmigo interactions when the tool was actively integrated into coursework compared to sections where it remained optional. In both cases, the platforms' value is contingent on deliberate teacher involvement.
ASSISTments’ evaluated implementation included two days of in-person professional development before each school year and three coaching visits during the year. The platform’s own teacher guidance notes students typically need four to six weeks before engaging productively with its feedback and reports. The evidence base for ASSISTments was generated under those conditions. It does not transfer to plug-and-play deployment.
The ASSISTments North Carolina replication enrolled nearly 6,000 students across 63 schools. The final analytic sample was approximately 20% Black, 22% Hispanic, and 61% economically disadvantaged. Statistically significant effects were found. For funders whose capital targets under-resourced populations, no finding in the evidence base carries more direct relevance. Most AI math tools have not been evaluated in these populations at all, let alone replicated across them.
ASSISTments is free to schools. An independent cost-effectiveness analysis by WestEd found that ASSISTments achieved a +0.10 SD long-term math gain at $46 to $70 per student. High-dosage tutoring achieving a comparable long-term effect size cost $3,200 to $4,800 per student, per WestEd's analysis. ASSISTments delivers the same outcome at roughly 1-2% of that cost. The primary cost to a funder is the ongoing professional development the evidence base requires, not the platform license. The North Carolina replication finding of +0.10 SD represents approximately a 31% improvement over the annual gain in math otherwise expected for eighth graders, per Arnold Ventures’ analysis. The 0.8-point gain also closes almost 30% of the achievement gap between the statewide average and Black students’ average scores, per WestEd’s initial results analysis.
ASSISTments has operated across multiple states, with the Maryland state partnership serving as the most documented scaling example. In 2024, Arnold Ventures and the State of Maryland co-invested $20 million to scale ASSISTments across Baltimore County, Charles County, and Prince George’s County. The teacher-integrated model is simultaneously the platform’s strongest evidential asset and its primary scaling constraint, as every new implementation requires investment in teacher readiness that cannot be bypassed without degrading the conditions that produced the gains. Khanmigo’s trajectory offers a relevant parallel. It has significant reach and philanthropic backing yet shows only 15% engagement in the absence of deliberate instructional integration. Scale of access and scale of impact are not the same thing.