Tutoring Works. The Devil is in the (Implementation) Details.

This summer, The Hechinger Report ran a headline: “Tutoring was supposed to save American kids after the pandemic. The results? ‘Sobering.’” It grabbed attention but communicated the wrong message. Tutoring itself did not fail. What failed was how tutoring was implemented.  

The research is clear: tutoring improves student learning and helps students catch up. Two recent meta-analyses—one by Matthew Kraft and colleagues (2024) and another by Nickow and colleagues (2024)—reviewed hundreds of randomized controlled trials, the gold standard for identifying causal impact. From these researchers, we know that high-dosage tutoring produces larger gains in student achievement than any other intervention, with average improvements of about one-third of a grade level.

Yet both meta-analyses also underscore a critical point: not all tutoring programs are created equal. The challenge lies not in the concept of tutoring itself but in how it is implemented and scaled. Sustained, high-quality tutoring in small-group settings yields substantial learning gains, but many students never receive enough tutoring for those benefits to take hold. Just as a patient who takes only half of a prescribed medication is unlikely to fully recover, a student who receives only a fraction of the intended tutoring dosage is unlikely to realize its full academic benefits.

And students urgently need additional instructional support. In the 2024 NAEP results, less than one-third of eighth grade students nationally are proficient in math and reading; the results are even worse for the most economically disadvantaged students. The gap between the need for tutoring and the amount of tutoring students receive is staggering. As of October 2024, only 37 percent of schools nationally even offered high-dosage tutoring, and just 8 percent of students received it, according to the federal School Pulse Panel.

So why do so few kids receive the tutoring support they need? The Strategic Data Project partnered with Accelerate and five states—Ohio, Colorado, Louisiana, Delaware, and Arkansas—to better understand what needs to be true for tutoring to be effective. Here is what we learned:  

Tutoring is most effective when it happens during the school day. In many states, participation drops sharply when tutoring is scheduled before or after school. In Ohio, for example, only 62 percent of scheduled tutoring sessions actually took place. Districts that wove literacy tutoring into existing reading blocks saw far higher participation than those offering online math tutoring outside of core hours.

Recognizing this challenge, Arkansas took a policy approach: the state required that students scoring below proficiency in literacy receive tutoring embedded directly within the school schedule. This ensured tutoring wasn’t left to chance or dependent on after-school attendance. As a result, participation rates rose and schools were better able to reach the students most in need.

Tutoring works best when it’s measured. We know that tutoring frequency and duration matter, yet many states still lack the systems to track this data. In Colorado, education leaders had to dismantle system silos just to connect student records with program data. Only then could they see whether students were receiving the recommended three 35-minute sessions each week. Louisiana faced a similar challenge, spending more than a year negotiating data-sharing agreements before it could even begin to understand how often students were participating.  States and districts need better data systems and staff who can use them to track attendance, dosage, and delivery in real time. This requires data-sharing stipulations with tutoring providers and common norms around data labeling and collection. Without data that enables visibility into student participation, it is impossible to know if children are getting the tutoring dosage that they need.

Tutoring depends on people. Tutors must be trained, supported, and reliable. Delaware’s pilot shows what happens when they are not. Students were scheduled for three sessions a week but on average received only two. Why? When the state fellow reviewed the first data pull, she noticed a pattern of tutor absences. For students who are already behind, that inconsistency can mean the difference between catching up and falling further back. One way to support tutors is to differentiate support for novice tutors versus experienced educators. For example, recent research from Stanford’s National Student Support Accelerator found that AI-supported human tutoring can tailor a tutor’s instructional strategies to better focus on improving student learning.  

Tutoring programs do best when they’re held accountable. Tutoring requires infrastructure: coordinators in schools, spaces for students to focus, and mechanisms that ensure vendors deliver on their promises. Outcomes-Based Contracting, which ties vendor compensation to student attendance, dosage, and program impact, is one tool districts in Arkansas are leveraging to support program implementation and maximize the program’s impact on student learning.

America has already invested billions of dollars in tutoring.  But when we judge reforms by their design instead of their delivery, we end up declaring failure before programs ever had a chance to succeed, even those with decades of rigorous evidence behind them.

Tutoring remains one of the most powerful tools to help students recover and thrive. Its future depends not on whether the idea was sound, but on whether we have the discipline to make it work for the children who need it most. 

Miriam Greenberg is the Senior Director of the Strategic Data Project at the Center for Education Policy Research at Harvard University. Matthew Steinberg is the Managing Director of Research and Evaluation at Accelerate.