The Netflix Problem in Corporate L&D: Too Much Choice, Too Little Signal
In 2000, a psychologist named Sheena Iyengar set up two jam-tasting booths at a grocery store. One booth offered 24 varieties. The other offered 6. The 24-variety booth drew more interest. But when it came to actually buying, shoppers at the 6-variety booth were 10 times more likely to make a purchase.
Barry Schwartz later named this phenomenon the "paradox of choice" — the counterintuitive finding that more options don't just fail to improve decisions, they actively prevent them. Decision paralysis sets in. People choose nothing.
Your company's learning platform has 20,000 courses. Your employees haven't started one this month.
This is not an engagement problem. It's a choice architecture problem. And Netflix solved it — not with less content, but with a better signal layer.
The 220 Million Title Problem
Netflix has somewhere north of 220 million subscribers globally. They offer thousands of titles. And yet, on any given Friday night, most users find something to watch within 90 seconds.
The reason is not that Netflix has better content. It's that Netflix's recommendation engine is so good that users never feel the size of the catalog. They see a curated surface of 20-30 titles that are, in aggregate, very likely to contain something they want to watch. The catalog's vastness is invisible.
Netflix spends approximately $1 billion per year on recommendation algorithm research. The algorithm processes over 300 behavioral signals per user: what you watched, when you paused, what you rewound, what you abandoned 12 minutes in, what you hovered over without clicking, what you watched at 11pm versus 8am, and hundreds more.
The result: a personalization engine so precise that different users browsing the same homepage see almost entirely different content surfaces, customized to their viewing psychology.
Now compare that to your LMS.
The Signal Poverty of Corporate L&D
Most corporate learning platforms run on two signals:
Signal 1: Job title. You're a Senior Software Engineer? Here are the courses for Senior Software Engineers. Never mind that you're three years in and deeply proficient in everything on that list, or that you've been asked to move into an engineering lead role and need management fundamentals, not technical refreshers.
Signal 2: Department. You're in Marketing? Here's the Marketing learning path. The fact that you spend 60% of your week in cross-functional meetings with engineers and product managers is not represented anywhere in your learning profile.
Some more modern platforms add a third signal — completed courses — but this is barely better. If you completed "Introduction to Data Analysis" last year, the system knows to stop recommending it. What it doesn't know is whether you retained any of it, whether you're applying it, or what adjacent skill would unlock the most value for you right now.
Netflix uses 300+ signals. Most LMS platforms use 2-3. The gap between those numbers is the gap between content discovery and content paralysis.
Callout: More content is not a learning strategy. More signal is.
What Barry Schwartz Would Say to Your L&D Team
The paradox of choice literature is useful for diagnosing exactly what happens when employees land on a 20,000-course catalog.
Choice overload occurs when the number of options exceeds the decision-maker's ability to evaluate them meaningfully. In consumer contexts, the threshold for paralysis is surprisingly low — research suggests meaningful degradation in decision quality starts around 10-15 options for unfamiliar categories.
Corporate learning is an extremely unfamiliar category for most employees. They don't know what they don't know. They can't evaluate whether "Advanced Negotiation Strategies" will be more useful to them than "Effective Stakeholder Management" — both could plausibly apply. So they defer. They close the tab. They come back when they have to.
Anticipated regret compounds the problem. When there are thousands of courses and you can only take a few, you worry that you're choosing the wrong ones. Netflix doesn't trigger this response because the recommendation surface is small enough that all the options feel vetted. The LMS catalog feels like a library where you have to do your own research — without knowing what subject you're looking for.
The status quo bias is the final nail. When the cost of choosing is high (cognitive load, uncertainty, fear of the wrong choice) and the benefit of choosing is diffuse and long-term (skill development, career growth), the rational move is to do nothing. The status quo wins.
This is not an employee motivation problem. This is a system design problem.
The Curation Gap: What L&D Should Actually Optimize
Most L&D teams spend their time on two things: content acquisition (licensing more courses, building more modules) and content delivery (platform administration, campaign emails, compliance tracking).
Neither of these addresses the real problem, which is signal quality — the accuracy and richness of the information used to determine what each employee should learn today.
The Netflix analogy is instructive here. Netflix's most valuable investment is not its content library. It's the behavioral data layer that tells the recommendation engine what each user actually responds to. The content is the raw material. The signal is the refinement.
For L&D, the equivalent investment is in:
Behavioral signals beyond completion. Does the employee apply the content? Do they return to reference it? Do they share it with colleagues? Engagement depth is more predictive of learning outcome than completion rate.
Skill gap inference from work behavior. A productivity platform, a communication tool, and a project management system all contain signals about where an employee struggles. Calendar data shows how they spend time. Document collaboration patterns show communication clarity. These signals exist — they're just not connected to the learning system.
Social learning signals. What are high-performers in this role learning this quarter? What does your manager think you need to develop? What is your team currently prioritizing? Peer behavior is one of Netflix's strongest signals — "people like you watched this" is genuinely predictive, even in learning contexts.
Goal-proximate content selection. If you've told the system you're working toward a leadership transition, content relevant to that transition should surface more prominently than content that optimizes for your current role. Most systems invert this: they over-index on what you've done, under-index on where you're going.
The One-Thing-Per-Day Constraint as a Feature
There's a counterintuitive design decision at the center of Omie: we show you one learning nugget per day. Not ten. Not a catalog. One.
This is often perceived as a limitation. It's not. It's an explicit application of the choice architecture research.
When you open Omie, there's no decision to make about what to learn. There's one thing. It's been selected based on your skill gaps, your goals, your recent behavior, and where you are in your mastery trajectory. The cognitive cost of starting is zero. The question is not "which of these 47 things should I do?" It's "yes or not today?"
The research on behavior change is unambiguous on this: reducing friction on desired behaviors increases follow-through. BJ Fogg's Tiny Habits work, Katy Milkman's Fresh Start Effect research, and the entire behavioral economics literature on commitment devices all point to the same principle: the best way to build a habit is to make the first decision the only decision.
One nugget per day is not a content scarcity constraint. It's a system that has already made the selection decision for you, based on better signal than you have about your own learning needs. Your only job is to show up.
Callout: The curation decision is as valuable as the content itself. Netflix's real product isn't movies — it's the 90-second path to finding one you'll love. Omie's real product isn't nuggets — it's knowing which one is right for you today.
The Recommendation Engine Gap in the LMS Market
Let's be specific about where existing LMS platforms fall short on the recommendation problem.
Catalog-based recommendation. Most enterprise LMS platforms (Cornerstone, SAP SuccessFactors, Docebo) have added recommendation features in recent years. Most of these are collaborative filtering systems similar to "customers also bought" — they surface content that was completed by similar users. This is a meaningful step above pure segmentation, but it optimizes for what's popular among your peer group, not what's specifically right for you.
Skills-graph systems. A newer generation of platforms (Degreed, 360Learning) builds skills graphs that map content to competencies and allows employees to self-report skill levels. This is better. But self-reported skill levels are notoriously unreliable (the Dunning-Kruger effect is well-documented in professional skill assessment), and skills-graph systems still don't capture the behavioral and contextual signals that make a recommendation truly predictive.
Assessment-driven systems. Some platforms build in pre-assessments to set personalized paths. This is closer to right, but assessments are snapshots, not continuous models. They degrade in accuracy as soon as the learner's state changes — which begins immediately after assessment.
What none of these systems has — yet — is the combination of continuous behavioral signal ingestion, a persistent mastery model that updates in real time, and a recommendation layer that incorporates current context as well as longitudinal history. That's the gap between them and what's possible.
The Signal Investment HR Should Make Right Now
If you're running an L&D function and you want to move toward better signal quality without rebuilding your entire stack, here's the prioritized list:
1. Pre-learning skill assessment with validated instruments. Replace self-reported skill levels with structured behavioral assessments. The assessment itself is signal. It tells you where someone actually is, not where they think they are. This feeds more accurate recommendations from day one.
2. Post-learning application tracking. Add a one-question prompt 7 days after any learning module: "Did you apply anything from this in the past week?" That answer is signal about transfer, which predicts whether the learning actually worked. Feed it back to the recommendation engine.
3. Manager input on development priorities. A quarterly 5-minute manager form: "What's the one skill you most want this person to develop in the next 90 days?" That's the highest-quality signal you can get about goal-proximate learning needs. It's also almost universally uncaptured.
4. Engagement depth over completion. Instrument your learning platform to track time-on-content, replay events, and save/share actions. These are better proxies for genuine engagement than binary completion.
None of this requires a new platform. It requires treating signal as a first-class investment — not a byproduct of your current system.
What a Better System Looks Like
The L&D system that actually solves the Netflix problem has three properties:
One clear recommendation per day. Not a catalog, not a suggested learning path with 40 modules. One thing, selected by an algorithm that knows you better than you know your own skill gaps.
A recommendation engine that earns trust through precision. The first time an employee thinks "wow, that was exactly what I needed right now," the system has won. Getting to that moment quickly is the product design challenge.
Measurement that closes the loop. If the recommendation engine is working, it should show up in behavioral change data, not just completion rates. The feedback loop from outcome to signal to recommendation is what makes the system improve over time.
The good news: the technology to build this exists. The content to fuel it is abundant. What's been missing is the architecture to connect them.
That's what we built. Sign up and see your first recommendation — the one thing you should learn today, based on who you are and where you're going.