By its very definition, innovation will always lead to some failed starts. And when that innovation involves educating children, it's especially important to learn from mistakes and adjust quickly.

The Walton Family Foundation has invested more than $385 million in creating new charter schools over more than two decades to seed educational innovation and improve U.S. education at scale. The foundation has allocated a small fraction of that investment—about $550,000—to virtual charter schools, which teach full-time students exclusively online.

We remain strong believers in creating educational options and opportunities. We have provided startup dollars to about a quarter of the charter schools in the United States, all with the goal of creating opportunity for high-needs students, and we recently committed to investing another $1 billion over the next five years to expand access to high-quality educational choices. In recent years, we have hoped that online charter schools could provide a lifeline for some students. But while we were enthusiastic about supporting online education entrepreneurs, our first priority is always making sure that students are served well.

Measuring impact is fundamental to responsible philanthropy. It is a responsibility we take seriously. The Walton Family Foundation spends about $10 million annually underwriting the nation's best researchers to investigate questions that will help us make smarter funding decisions to benefit high-needs students, develop promising new technologies and methods to fuel student learning, and help parents, educators, and policymakers improve outcomes for children.

As the largest private funder of charter schools and as strong believers in making fact-based decisions, we wanted to see the hard evidence on virtual charters: What would a dependable measure of the impact of these schools show about their students' academic growth? We funded three research studies—by the Center for Research on Education Outcomes (or CREDO), at Stanford University; the Center on Reinventing Public Education, at the University of Washington; and Mathematica Policy Research—to investigate this question. As with all of our research dollars, we committed to funding these research teams regardless of what their investigations revealed.

The results are, in a word, sobering. The CREDO study found that over the course of a school year, the students in virtual charters learned the equivalent of 180 fewer days in math and 72 fewer days in reading than their peers in traditional charter schools, on average.

This is stark evidence that most online charters have a negative impact on students' academic achievement. The results are particularly significant because of the reach and scope of online charters: They currently enroll some 200,000 children in 200 schools operating across 26 states. If virtual charters were grouped together and ranked as a single school district, it would be the ninth-largest in the country and among the worst-performing.

Funders, educators, policymakers, and parents cannot in good conscience ignore the fact that students are falling a full year behind their peers in math and nearly half a school year in reading, annually. For operators and authorizers of these schools to do nothing would constitute nothing short of educational malpractice.

As a result of these findings, we at the foundation will ask new, more rigorous questions of online charter operators when we review their funding proposals, in order to expose whether applicants are addressing the problems this research identified. In particular, we want to know: Are the operators suggesting innovative solutions to improve the quality of online learning? There is no magic formula here, but it is clear that what exists doesn't create the academic opportunities children need. Going forward, we'll probe deeply on applicants' answers to the following questions:

What does the proposed instructional program look like? Just as it is important in traditional brick-and-mortar schools for students to spend time with teachers, virtual schools must provide plenty of time for students to learn and interact in live, synchronous ways with their teachers. In today's online charters, students typically have less instructional time with their teachers in a whole week than students in brick-and-mortar schools have in a day. We don't presume to know what the best pedagogical approach is for all children, but clearly this formula needs to change.

What are the proposed teacher-student ratios? Today's online schools have much larger ratios than their brick-and-mortar counterparts. It's unclear what the right balance is, but it seems clear that schools must facilitate interactions between students and teachers.

What are the expectations of parents? Virtual schools must help students learn without requiring parents to be constantly present and monitoring progress. Parents, of course, must be involved in their children's education, but schools can't abdicate their responsibility in this equation. Virtual charter educational providers should be thinking through the right role for parents and how schools will meet their own educational remits.

We urge policymakers to make changes, too. Charter authorizers—government-sanctioned bodies responsible for reviewing and approving charter operators—must take action if schools are failing students. And the oversight of educational practice applies to authorizers, as well as to schools and educators. Authorizers should be graded on the performance of their portfolios: If schools fail students, authorizers must take action. If they don't, authorizers themselves should be put out of business.

Going forward, authorizers should create new accountability systems to ensure that no school fails, month after month and year after year. The review process must include observation of instruction and close review of student and parent expectations. We think a shorter review cycle, rather than waiting years, might catch problems earlier.

As states think about the future of online education, they should rethink their expectations and policies and test novel policy arrangements. For example, one policy that we think has potential would tie funding to performance. Four states—Florida, Minnesota, New Hampshire, and Utah—are currently testing performance-based funding systems. Perhaps we'll discover that providing virtual schools funding only after students demonstrate mastery is the right way to hold these publicly funded schools accountable.

That said, with this approach, it would be important to guard against perverse incentives that could arise. For example, policymakers would need to make sure that schools weren't pushing out or otherwise neglecting students who are so far behind that getting them to pass required courses or summative assessments would appear extremely resource-intensive.

Other funding reforms worth considering are tracking enrollments and providing associated student-based funding each month, rather than annually. This would help prevent students from being easily forgotten.

To be clear, our comments about online charter schools are not an indictment of instructional technology or online learning more generally, nor how these stand to help create more high-quality educational options. Nor is this the Walton Family Foundation abandoning its mission of creating more educational opportunities for American children. There are many examples of technology being used in conventional classrooms in ways that enhance learning. New blended learning models are showing promise, as is allowing students to customize learning through the use of online platforms.

But the data from this study do not lie: Online education must be reimagined. Ignoring the problem—or worse, replicating failures—serves nobody.

This article originally appeared in Education Week.