‍

At Disco, we hosted Google DeepMind's Senior Product Manager to reveal what separates transformational learning programs from those stuck at 5% engagement. Here's how the world's leading AI research validates why training businesses using Disco's human-first, AI-powered approach are seeing 4x higher completion rates.

In a recent Disco Academy workshop, we invited Kaiz Alarakyia (Senior Product Manager at Google DeepMind) to share something training business leaders need to hear: when his team set out to build the best AI tutor possible, they discovered that standard Gen AI models were missing five fundamental learning science principles that every good human tutor intuitively uses.

This matters for your training business. Most AI tutoring tools end up making the same mistake: they force-fit powerful technology into education without actually understanding how people learn. Dismal 5% engagement rates kill program completion, customer retention, and revenue growth.

At Disco, we've built our platform on a different approach. One that DeepMind's research validates: learning must be human-first, AI-assisted. Not AI-only.

The fork in the road: technology-first or learning-first?

When large language models emerged, DeepMind's education team faced a critical decision. They could take the technology and force it into education, or they could step back and ask: what problems are we actually trying to solve? What behaviors do we care about? Where can AI genuinely help?

"We decided to take that second path," Kaiz explained. The team brought together technologists, pedagogy experts, and professors to tackle a deceptively complex question: how do you build the best AI tutor for every learner?

But first, they needed to understand what makes a good tutor in the first place.

What students and teachers revealed about AI tutoring

The DeepMind team conducted semi-structured interviews and workshops with organizations across the world. They talked to students, teachers, pedagogy experts, and institutions, asking: what makes a good AI tutor? What are the risks and benefits of AI in learning?

From these conversations, they distilled five learning science principles that were absolutely necessary for an AI tutor to demonstrate, and that weren't present in today's Gen AI models.

The 5 principles every transformational training program needs

DeepMind's research identified five learning science principles that were absolutely necessary for effective learning, and that weren't present in today's Gen AI models. At Disco, we've embedded these principles into our platform architecture. Here's what they mean for your training business:

1. Inspire active learning
Good tutors don't just deliver information. They engage learners in the learning process itself, encouraging them to construct knowledge rather than passively receive it.

What this means for training businesses: If your program relies on video lectures and quizzes, you're leaving money on the table. Active learning through projects, discussions, and peer collaboration drives the completion rates that translate to customer renewals and referrals. This is why Disco's platform is built around cohort-based programs with discussion spaces, not just content delivery.

2. Manage cognitive load
Effective tutors understand when a learner is overwhelmed and can break down complex ideas into digestible chunks that match the learner's current capacity.

What this means for training businesses: Your learners are busy professionals. If your program dumps too much at once, they drop out. AI can help identify when learners are struggling and surface that to instructors, but the program design must support progressive skill building. Disco's modular structure lets you sequence learning in ways that prevent overwhelm.

3. Deepen metacognition
The best tutors help learners think about their own thinking. They develop awareness of how they learn, what strategies work for them, and where they need more support.

What this means for training businesses: Metacognition is what creates lasting behavior change, not just knowledge transfer. When learners reflect on their progress and process, they're more likely to apply what they've learned and see it as transformational. This drives testimonials, case studies, and word-of-mouth growth. Disco enables reflective practices through journaling prompts, peer feedback, and progress tracking.

4. Stimulate curiosity
Great tutors spark genuine interest and intrinsic motivation. They help learners connect new knowledge to what they already care about.

What this means for training businesses: Curiosity is what keeps learners coming back. If your program feels like a chore, completion rates tank. The most successful training businesses on Disco create "aha moments" by connecting curriculum to real-world challenges their learners face, using community discussions to surface diverse perspectives that spark curiosity.

5. Adapt to learner goals and needs
One-size-fits-all doesn't work. Effective tutors constantly adjust their approach based on individual learner context, background, and objectives.

What this means for training businesses: Your learners have different starting points, different goals, and different learning speeds. Programs that force everyone through the same rigid path see higher dropout rates. Disco's AI can help personalize pathways while maintaining the cohort structure that drives accountability.

The 95% problem: why your training program might be hemorrhaging revenue

During the workshop, Kaiz addressed what he calls "the 95% problem." Only about 5% of learners typically engage with AI tutoring tools in meaningful ways. But here's what matters for training business leaders: this isn't just an AI problem. It's a program design problem.

"If adult learners are not engaged, part of that comes from the environment they're in. Learning on the side of their desks, not in a full-time capacity," Kaiz explained. "My recommendation is to figure out ways to embed AI into existing learner workflows."

For training businesses, low engagement means low completion rates. Low completion rates mean poor outcomes. Poor outcomes mean no testimonials, no referrals, and no renewals. It's a revenue problem.

This is where Disco's approach diverges from AI-only solutions. AI can identify where interventions are needed by tracking not just completion but the actual learning process: how long someone spent on a task, how many back-and-forth exchanges they had, what questions they asked. But AI alone can't create the accountability structures that keep learners engaged.

Human instructors, peer cohorts, and community connection create the environment where learning sticks. At Disco, we combine AI intelligence with social learning architecture because we know this is what separates 70%+ completion rates from the industry average of 15%.

What winning training businesses do differently

The insights from DeepMind validate what we've learned from hundreds of training businesses on Disco: AI is powerful, but it's not a silver bullet. The training businesses seeing 4x higher engagement and revenue growth are the ones combining AI efficiency with human-centered program design.

Here's what they're doing:

They embed AI into cohort experiences, not replace them. The most successful programs on Disco use AI to eliminate operational busy work (grading, scheduling, reminders) so instructors can focus on what humans do best: coaching, connection, and real-time adaptation. AI handles the scalable parts. Humans handle the transformational parts.

They use AI insights to improve programs, not just track completion. When you can see how long learners struggle with specific concepts, what questions come up repeatedly, and where people get stuck, you can iterate on your curriculum in real time. This is how training businesses move from good to exceptional without adding headcount.

They design for social accountability from day one. Group-based cohorts create peer pressure that AI alone cannot replicate. When learners know their cohort is counting on them to show up, complete assignments, and contribute to discussions, engagement skyrockets. This is the secret to 70%+ completion rates.

They focus on outcomes, not features. The training businesses winning with Disco aren't the ones with the fanciest tech stack. They're the ones who understand that learners pay for transformation, not content. AI can personalize the journey, but the program must be designed around real behavior change and measurable results.

The future of training businesses: beyond chatbots

Kaiz also addressed the assumption that chat will remain the primary way learners interact with AI. "Chat was just the first modality," he noted. "Everyone was like, okay, chatbot is the way to go, but there's so much more potential than that."

DeepMind's investments in immersive world models like Genie 2, combined with advances in voice interfaces and multimodal learning, point toward far richer possibilities. Practice-based learning environments where learners can interact with virtual scenarios. AI coaches that can see and respond to their work in real time. Voice-based feedback on presentations and communication skills.

At Disco, we're already integrating voice capabilities into our platform because we see the same future. Training businesses focused on leadership development, communication skills, and soft skills will especially benefit from these advances. But the principle remains: AI enhances human learning, it doesn't replace it.

The question for training business leaders is: are you building your programs on a platform that can evolve with these capabilities? Or are you locked into legacy LMS systems that can't adapt?

The takeaway: human-first, AI-assisted is how you scale without sacrificing transformation

What's remarkable about DeepMind's approach is how it reinforces the principle we've built Disco around: technology should enhance human connection, not replace it.

The five learning science principles (active learning, cognitive load management, metacognition, curiosity, and adaptation) aren't just academic theory. They're the foundation of programs that actually work. Programs that drive completion rates above 70%. Programs that generate testimonials, referrals, and revenue growth.

AI makes it possible to provide personalized support at scale. But only when it's built on solid learning science foundations and integrated into social, human-centered learning environments.

As Kaiz put it: "AI came about so quickly that everyone was like, okay, let's do AI. But we needed to step back and ask: what are the problems we're trying to solve? What are the behaviors we really care about?"

For training business leaders, the question is the same. What behavior change are you trying to create? What outcomes do your learners need to achieve? How can you design programs that combine AI efficiency with the human accountability that makes transformation stick?

At Disco, we've answered these questions by building a platform that combines cohort-based learning, community connection, and intelligent automation. The result? Training businesses that scale from six figures to seven without adding headcount. Customer academies that reduce churn by 40%. Leadership programs with completion rates 4x higher than industry average.

That's what happens when you get the learning science right. That's what happens when AI assists humans instead of replacing them.

Key takeaways

For training business leaders: Google DeepMind's research validates what Disco has built into its platform: transformational learning requires five core principles (active learning, cognitive load management, metacognition, curiosity, and adaptation) that most AI-only solutions miss. The training businesses achieving 4x higher completion rates are those combining AI efficiency with human-centered cohort experiences.

The 95% engagement problem: Only 5% of learners engage meaningfully with AI-only tutoring tools. The solution isn't more AI, it's embedding AI into social learning environments where peer accountability and instructor support create the conditions for transformation.

Beyond chatbots: Voice interfaces and immersive learning environments represent the next frontier for training businesses, especially those focused on leadership development and soft skills. The platforms that will win are those built on solid learning science foundations, not just impressive technology.

Why human-first, AI-assisted wins: Training businesses on Disco achieve 76% average engagement rates and 68% completion rates by using AI to eliminate operational busy work while preserving the human connection that drives behavior change. This is how you scale from six to seven figures without adding headcount.

About the author: Candice Faktor is Co-founder and Co-CEO of Disco. She previously scaled Wattpad to 120M+ users and is passionate about building the future of human-centered, AI-powered learning.t

About Disco: Disco is the purpose-built learning platform that unlocks human potential through AI-powered, social learning experiences. We help training businesses and customer academies create transformational programs that scale by combining cohort-based learning, community connection, and intelligent automation. Because the future of learning is human-first, AI-assisted.