The Missing Accountability Layer in Every L&D System You've Ever Used

The enterprise learning market has had a remarkable few years. AI can now generate a full training curriculum in minutes. Simulation platforms let team members practice difficult conversations before they ever take a live call. Coaching tools surface real-time feedback from actual customer interactions. By almost every measure of content quality and delivery sophistication, the state of the art in 2025 is dramatically better than it was in 2020.

And yet the fundamental question at the center of the enterprise—will this training actually work?—remains almost entirely unanswered.

This is not a technology problem. The tools exist. It’s a framing problem. The L&D industry has spent decades optimizing for the production and delivery of learning content, and has largely neglected the discipline of predicting and measuring its impact on real-world skill. Every prior wave of innovation—eLearning in the early 2000s, LXPs, video-based microlearning, and now AI-generated content and simulation—has made the same structural mistake. Each one asked “how do we build better training?” without first asking “how do we know if training works?”

AI is now eating L&D at remarkable speed, and we are about to make the same mistake at greater scale, with greater confidence, and greater cost.

What Learning Science Already Knows

The field of cognitive science has understood for well over a century how human beings actually acquire and retain skills, and the core findings are not subtle.

Ebbinghaus’s forgetting curve established in the 1880s that people forget the majority of new information within days without reinforcement.

Spaced repetition—returning to material at increasing intervals as it starts to fade—is among the most replicated findings in learning research, and consistently outperforms massed practice by a wide margin.

Desirable difficulty, Robert Bjork’s term for the counterintuitive finding that harder learning conditions produce stronger long-term retention, explains why retrieval practice beats re-reading, and why active simulation beats passive video even when the video feels more satisfying in the moment.

Cognitive load theory adds a further constraint: working memory is limited, and programs that attempt to transfer too many skills in too short a window—regardless of the quality of individual components—exceed what learners can meaningfully absorb and retain. A polished video lesson scores well on completion rates and learner NPS. It produces, on average, poor real-world skill transfer. A simulation run once at onboarding still only checks a box.

None of this is new or contested. It is textbook cognitive psychology. And almost none of it is systematically applied in the way enterprise L&D programs are designed, delivered, or evaluated.

The Accountability Gap

The metrics that dominate L&D reporting—completion rates, assessment scores, learner satisfaction, time-in-platform—are all measures of training activity, not training impact. They answer “did learning happen?” rather than “did skill develop?” The consequences are predictable: organizations invest in modalities with low real-world ROI because they’re easy to produce and easy to track. If outcomes don’t improve, the diagnosis is almost always “we need more training” rather than “our training program is poorly designed and delivered.”

We believe the missing layer is a principled methodology for evaluating the likely skills impact of a training program before it runs and then validating that prediction against real outcome data once it does.

Credit Where It’s Due and Where It Runs Out

To be fair, the industry has made some good attempts on this topic before.

A few platforms have built spaced repetition features into their platforms to varying degrees. Each of them provide ways to deliver recurring microlearning nudges based on the forgetting curve.

But there is a difference between implementing a feature that applies a learning science principle and having a methodology that evaluates whether your program applies it sufficiently.

That distinction is where the industry is falling short.

They enable an organization to run spaced repetition nudges on one topic while covering a dozen others exclusively through video and classroom pull-offs, with no mechanism to surface that imbalance or score its likely impact on outcomes. This doesn’t evaluate/predict the program as a whole.

More fundamentally, none of these platforms ask the question that actually matters: was it enough? Not “did we schedule reminders?” but “given the complexity of this skill, the modality used to introduce it, the number of practice repetitions, and how that practice was distributed over time, does this program design have a realistic chance of producing durable skill transfer?”

This is a program-level evaluation question, and answering it requires a framework that sits above any individual feature. No platform currently provides it.

Introducing a Different Kind of Measurement

Zenarate’s approach to date has been to connect training inputs directly to longitudinal skill and KPI data—building evidence over time about what program designs actually produce skill movement, and what skill movement actually shifts outcomes. That makes it possible to say something meaningful about training ROI rather than inferring it from completion proxies.

But this approach has also revealed a problem.

Longitudinal outcome data is, by definition, retrospective. It tells you that a program worked or didn’t after the associates have been trained. What we’ve been missing is an earlier signal: a way to evaluate the likely skills impact of a program at the design stage, before deployment, so that organizations can make informed decisions about program structure rather than discovering its flaws six months later in the outcome data.

That is the gap the Zenarate Skills Index™was built to address. The core idea is to apply learning science to the way training programs are designed and delivered — producing a predictive score for the likely skills impact of a program on real-world outcomes, grounded in the same principles of spaced repetition, retrieval practice, and desirable difficulty described above, and calibrated against Zenarate’s longitudinal outcome data. Then close the loop again: update the index as new deployment data comes in, so the predictions sharpen over time.

The Index evaluates programs both at the topic level and in aggregate, where cognitive overload enters as a meaningful dimension. A program that attempts to cover fifteen skills across a two-week onboarding window, with each topic visited once and no spaced reinforcement, has a structural problem that individual topic quality cannot fix. Even if every simulation is well-designed, every lesson well-written, and every coaching conversation well-executed, the aggregate cognitive load is too high and the reinforcement too sparse for durable skill transfer to occur at scale. In this example, the Index makes this visible before it becomes an outcome problem.

The practical outputs are what make the concept concrete. A high-index skill with poor utilization is a different problem than a low-index skill with strong utilization and they require different interventions. A topic covered primarily through video and eLearning carries a structural disadvantage that no amount of content quality will overcome. Similarly, a few simulations run once at onboarding is not the same investment as recurring, personalized practice run multiple times across ninety days, or better yet weekly based on ‘yesterday’s skills gaps.’

What this enables, ultimately, is accountability—a word that has been too absent from L&D for much of its history. Not accountability as a threat, but accountability as a feedback loop: a way to evaluate program design against evidence before deployment, confirm that the skills covered are meaningful enough to justify the investment, and then close the loop against real outcomes to validate and sharpen the prediction over time.

The Same Mistake at Greater Scale

If our past is any indication of our future, its entirely possible that L&D will absorb everything AI has to offer—faster content, personalized delivery, simulation at scale—and still find a way to avoid the one question that actually matters:

Did it work?

Every prior wave of L&D innovation deferred that question. AI won’t be any different unless the field makes a deliberate choice to treat it differently and the window to make that choice is narrowing.

AI is accelerating the production of training fast enough that the gap between content volume and actual skills impact could widen dramatically before anyone notices. The Zenarate Skills Index™ – alongside our existing ROI and impact tools is a bet that the field is ready to close that gap.