As co-founder of Education Perfect, Shane Smith has spent nearly 20 years navigating the gap between what the research says and what works in real classrooms. His reflections on that journey offer a useful lens for any school evaluating the tools in front of them.
Key takeaways
- Building on learning science from the beginning produces a more defensible product over time, particularly as schools become more discerning about what they adopt.
- There is a meaningful difference between citing research in marketing and implementing it in a way that moves outcomes for students. Smith describes three distinct stages that EdTech companies typically move through.
- What feels good to learners in the short term is not always what produces durable learning. Holding that tension is one of the harder ongoing challenges in product design.
- AI has real potential in the feedback loop, but the design choices around how it is integrated matter enormously. Tools that do the thinking for students remove the learning opportunity.
- Every minute a teacher or student spends with a tool carries an opportunity cost. That framing is a useful test for any EdTech decision.
There is a version of evidence-based learning technology that most school leaders will recognise: a company’s website cites Hattie, Rosenshine, or Bloom’s Taxonomy, a reference to cognitive science in the sales pitch, and a claim that the product is grounded in research. The language has become so common that it has started to lose meaning.
Shane Smith, who co-founded Education Perfect (EP) with his brother Craig in 2007, has watched this dynamic develop from a unique vantage point. When EP initially launched as Language Perfect, a vocabulary learning tool built for Language classrooms, the science of learning was not yet a mainstream conversation in schools. The principles the Smith brothers built into the product, including the testing effect, spaced repetition, deliberate practice, and keeping students in their zone of proximal development, came not from a marketing brief, but from personal reading habits shaped by a scientist mother and a genuine preoccupation with how people learn.
Nearly two decades later, Smith has returned to EP to lead its AI initiatives, and he finds himself thinking about the same question the platform started with: what does it actually mean to build something that moves the dial for students, and how do you know you are on track?
The Three Stages of Evidence in EdTech
One of the most useful frameworks Smith offers in this episode is a description of the stages that EdTech companies tend to move through in their relationship with evidence.
The first stage is the most familiar.
“The marketing team cherry picks ideas from the research that sort of vaguely match up to concepts or mechanisms in your product,” Smith says. “Obviously this is the least valuable for students because there’s not necessarily alignment there.”
The second stage moves closer but still falls short: surface-level implementation that maps a product feature to a recognised framework and treats that mapping as sufficient.
“Building in surface level implementations to tick a box and say ‘we do Bloom’s Taxonomy’ because, look, we have mapped it out here, therefore it’s done.”
The third stage, which Smith describes as the one he hopes all EdTech companies eventually reach, is harder and slower. It requires not only an understanding of what the literature says, but crossing what he calls the chasm between a good idea and a successful implementation.
“Building an evidence base to demonstrate that not only have you understood the ideas in the literature, but you have crossed the chasm between a great idea and successfully implementing it in a way that moves the dial for students.”
This framing shifts the question schools should be asking of their EdTech providers. Citing evidence is not the same as building on it. The more meaningful question is whether a company can demonstrate that its specific implementation of a research principle, in real classrooms, with real students, actually produces the outcomes the underlying research predicts.
What the Classroom Teaches You That the Research Cannot
One of the clearer threads running through Smith’s account of building EP is the degree to which time spent in classrooms has shaped the product, sometimes by confirming what the research predicted and sometimes by revealing where the initial assumptions were wrong.
The expansion from Language Perfect to Education Perfect is a case in point. The original platform was built around vocabulary acquisition, a relatively atomic learning task that mapped well onto the retrieval and repetition mechanics the team had developed. When they extended those mechanics to History, Geography, Science and English subjects, the early results were instructive.
“We found that a History classroom and a Geography classroom and a Science classroom all have different needs. They are trying to teach different shapes of knowledge.”
The response was not to abandon the evidence base but to broaden the reading of it. Bloom’s Taxonomy offered a framework for understanding what had been missing: the original product catered well to one level of cognitive demand, but teachers were needing to teach across all of them. The product evolved accordingly, building in scaffolding, sequenced information, and a closer loop between assessment data and lesson assignment.
What made that evolution possible, Smith argues, was staying genuinely close to the people using the product. “If you’re spending a lot of time with teachers and students, they’re probably going to tell you if it’s not working. And that’s going to set you back on track.”
The Tension Between What Feels Good and What Works
One of the more candid observations Smith makes in this episode is about the persistent tension between learner experience and learning outcomes, a tension that sits at the heart of a great deal of EdTech design.
“One of the challenges in the classroom is that there is sometimes a disjunct between the things that feel good to the learner and the things that really move the dial,” he notes.
Passive video consumption, for example, tends to feel engaging and accessible. The evidence on its contribution to long-term retention is considerably less encouraging. “We know that a lesson that focussed just on those could never really open up higher order thinking opportunities for students.”
Holding that line requires ongoing effort, particularly in a market where engagement metrics are frequently used as proxies for learning impact. Smith is direct about the distinction: a tool can generate high rates of interaction and still leave students no better off. The question he keeps returning to is whether the time a student spends with a tool has genuinely moved their understanding forward, not whether they found it enjoyable in the moment.
AI in the Feedback Loop: Design Choices Matter
The development of EP’s AI-Powered Feedback Tool, which Smith has guided over the past several years, brings many of these questions into sharp focus. The problem the tool was designed to solve is one that will be familiar to most secondary teachers: open-ended student responses, particularly in subjects like English, Science and Humanities, have always been the hardest part of the feedback loop to close. Teacher capacity limits how often meaningful feedback reaches students on extended responses, and when it does not, the incentive for students to invest effort in those responses diminishes.
The AI feedback tool aims to address that gap by providing immediate, personalised feedback on open-ended responses, along with what EP calls a learning loop, where responses that fall short of a reasonable standard are returned to students for revision before they can proceed.
Critically, the design of the tool reflects a deliberate set of choices about what role the AI should and should not play.
“It’s not a chatbot,” Smith says. “And it doesn’t do the heavy lifting. It refuses to do the heavy lifting for students.”
That distinction matters because, as a previous interview episode with Professor Rose Luckin explored in depth, the design choices around AI feedback determine whether the tool prompts students to think harder or simply removes the need for them to think at all.
Smith is also candid about where the evidence currently stands. Early data suggests the tool is producing more effortful responses and higher quality answers in the moment. Whether that translates to deeper long-term retention is the question the ongoing impact study with Professor Luckin’s team is designed to answer.
“What we’d really like to see is to track this from implementation all the way through to long-term recall and prove, end to end, that having implemented this, students have learned more and are better off in the process.”
The Opportunity Cost of Every Lesson
Perhaps the most useful framing Smith offers for educators thinking about their technology choices is also the simplest. Every time a teacher directs students to a tool, there is an opportunity cost. The same time could have been spent in direct instruction, small group work, or any number of other approaches. That is not an argument against using tools, but it is an argument for being deliberate about which ones earn their place.
“Teachers and students have limited time in their day,” Smith observes. “Every time a teacher decides to use EP for a task to assist their teaching, there is a real opportunity cost for them choosing this rather than another style of teaching. I think we have a duty of care to ensure that they are spending this precious time in a way that is really going to move the dial for their students.”
That framing, which Smith traces back to EP’s founding principles, offers school leaders and educators a practical lens for evaluating any tool they are considering. The question is not whether a product references research, or whether it generates engagement, but whether the time students and teachers spend with it produces outcomes that could not have been achieved to the same level another way. In a period when schools are under pressure to make better use of fewer resources, that question is worth asking more carefully than ever.
Hear the full conversation: This article draws on Episode 4 of Well-MeasurED, Education Perfect’s interview series exploring what genuinely works in education and EdTech with researchers and practitioners.
Watch or listen to Dr Jill McGuire and James Santure’s conversation with Shane Smith, Founded for Impact: Why an Evidence-First Approach Is More Urgent Than Ever, to hear the story in his own words.
