When we talk about tutoring, we tend to focus on impact. But impact is only one part of the equation. To understand how tutoring actually works, and how to make it work better, we also need to understand cost. Not just sticker price, but the full set of resources required to deliver the program. Who is involved? What are they doing? How much time are they investing?
That’s why Accelerate built a cost analysis tool. Now, we’re starting to see what happens when it is used as part of rigorous program evaluations.
A recent randomized controlled trial from Harvard’s Center for Education Policy Research used our cost analysis tool, which is based on the ingredients method of cost analysis, to evaluate a tutoring program in Arkansas. The overall effects were null. In education research, that is often where the story ends, if the study gets written up at all. Too many null findings disappear from public view, leaving the field with little to learn about whether the intervention failed, the implementation failed, or the design never gave the intervention much of a chance. This study gives us something we rarely see in education research — a clearer view into how cost, implementation, and dosage relate to each other. Even with null effects, implementation details can help explain what was actually tested, what students actually received, and what would need to change before trying again. That kind of learning is essential if we want evidence-based interventions to scale successfully.
The same program was implemented in two districts within the same state, under the same policy and funding regime. However, implementation looked very different.
Same Program, Very Different Costs
The study examined BookNook, a virtual tutoring provider implemented in two districts under Arkansas’s statewide high-impact tutoring initiative.
Both district sites used the same vendor, under the same policy conditions, with the same funding mechanism.
And yet, in practice, the full cost of implementing the program varied dramatically:
- District A cost: ~$1,600 per pupil
- District B cost: ~$475 per pupil
That’s more than 3x difference in cost for the same program. How does that happen?
Some of District B’s lower per-pupil cost likely reflects economies of scale with fixed costs being spread across more students, not just differences in implementation intensity. However, implementation decisions at the school and district level were also part of the story, which meant the districts used different resources to support implementation.
Cost Isn’t Just Sticker Price, It’s Implementation
The power of the ingredients method, and the reason we built the tool this way, is that it doesn’t just estimate total cost. It shows what’s driving those costs.
The ingredients method provides a structured way to capture all resources required to implement a program (personnel, facilities, materials) and assign them value. It’s been used in research for decades, but often isn’t applied in ways that are accessible to practitioners.
In this case, the major cost difference was almost entirely personnel. And that personnel was based on two very different approaches to implementation.
District A invested significant staff time in making the tutoring program happen:
- School leaders contributed a small portion of their time (around 1% FTE), signaling that the program was viewed as a priority for leadership.
- Four dedicated instructional collaborators helped students log in and stay engaged.
- District and school-level project managers (25% FTE each) ensured the program actually ran.
- An additional staff member (1% FTE) processed data, likely feeding it to project managers and school leaders
District B relied on a lighter local staffing model, with smaller time investments from school-based staff. Notably, District B did not have dedicated instructional collaborators or a school-based champion dedicating 25% of their time to implementation.
Both districts chose the same program. But they made very different decisions and paid for them, mostly in staff time. Cost variation isn’t random noise. It reflects implementation decisions, and those decisions drive dosage.
You Get What You Pay For (At Least in Implementation)
Those differences in implementation were associated with differences in how much tutoring students actually received.
| District | Cost Per Pupil | Hours of Tutoring on Average |
|---|---|---|
| A | ~$1,600 | 11.7 |
| B | ~$475 | ~5 |
That’s more than double the dosage.
This matters because prior research suggests dosage is an important driver of tutoring effectiveness. If kids aren’t getting the tutoring, they aren’t going to get the additional learning. The program was designed to deliver up to 18 hours, and neither district got there, but one got much closer than the other.
The overall study found no statistically significant impact on ELA outcomes. That does not mean the cost difference is irrelevant. District A’s 11.7 hours came closer to the intended 18 hours than District B’s 5 hours, though still fell short. The right conclusion is not that the more expensive version would necessarily have worked. It is that the higher-cost implementation delivered substantially more of the intervention, while the lower-cost version likely never gave the model a real chance.
Importantly, the cost per session delivered was relatively similar ($136 per session for District A, $95 for District B), which suggests the higher total cost in District A was not simply inefficiency. It reflected more tutoring actually reaching students.
This Should Be Standard Practice
One of the biggest takeaways here isn’t about BookNook specifically. It’s about research design.
We should not be running evaluations of tutoring, or any educational intervention, without collecting detailed cost data alongside outcome data. Right now, that’s still the exception. But it shouldn’t be.
You don’t need our tool specifically. But you do need:
- A commitment to the ingredients method
- Transparent reporting of the resources required to implement a program (including school side costs)
- Alignment between cost data and implementation reality
- Transparent reporting of null findings with enough implementation detail to learn from them
Without that, we’re missing an important part of the picture.
What This Means for Schools and Providers
For schools, this is a reminder that sticker price is not the same as cost. The vendor fee is only part of the costs schools face to implement a program. An understanding of the requirements for staff time, on demand problem solving and coordination, and leadership attention makes an enormous difference to whether a program is implemented successfully. These things often do not show up in procurement decisions or budget line items, but they matter to the people who have to make programs happen.
For providers, the lesson is a bit different. Implementation happens inside schools. And variation in that implementation can drive huge differences in both cost and results. Understanding what strong implementation looks like, and helping partners get there, is part of achieving success and should be part of the sales process.
What Happens Next
This is just one study. We need more like this. If we continue including high-quality cost analysis as part of rigorous impact evaluations, we’ll start to see patterns.
Some patterns are obvious. Understanding how much tutoring programs actually cost, including school-side costs, and comparing across models is the beginning of improved decision-making transparency.
But the answers to new questions may emerge:
- How much does successful implementation cost?
- Which implementation choices consistently drive better implementation and better outcomes?
- How much implementation and cost variation exists within the same program?
Tutoring can lead here. The field is already built around rigorous impact evaluation. It should also be built around transparent cost and implementation analysis. As tutoring as a field generates more and better data about true costs, other education interventions should be held to a similar standard. Clearly establishing evidence of effectiveness and costs is a floor not a nice-to-have. And when results are null, the field should still learn from them rather than letting them disappear.
Luke Kohlmoos is Managing Director of Strategy at Accelerate.
