What Validated Comprehension Means for L&D Measurement

Consider this scenario: Sarah is finishing her second coffee when a notification arrives in Teams. She has thirty minutes before her next meeting and a compliance module on her queue she has been postponing for two weeks. The notification opens with a question about compliance.

She types a hurried answer. Instead of moving on, the system asks her why she thinks that. She pauses to think about her next reply.

Why has L&D measurement been stuck on completion rates for twenty years?

Learning management systems (LMS) were built to answer the question regulators asked in the early 2000s: did the employee complete the training? Course completion served as a usable proxy when nothing better existed, and the technology category that grew up around it was built to capture clicks, time on page, and quiz scores.

Two decades later, the question has changed. CFOs want to know whether the L&D investment produced capability rather than participation, and boards want confirmation that the workforce can execute the strategy that depends on new skills. However, the infrastructure has not. Twenty years of bolt-ons (learning experience platforms, skill graphs, microlearning catalogs) added new interfaces on top of the same activity data. Hours trained went up, yet whether anyone learned remained unknowable.

While the consensus among L&D leaders is that course completions fail as sufficient evidence of capability, the subsequent measurement stack — comprising post-training quizzes, behavioral inference, and manager attestations — is often propped up as proof despite remaining merely proxy data generated outside the learning event itself.

Regrettably, mere completion data answers a question no one is asking anymore.

What does validated comprehension actually mean?

Knowing whether someone has learned happens in one of three ways: inference (guessing at skill level from career history, content consumed, or behavioral signals from work), assessment (testing understanding through a separate event after the course, usually a quiz), or demonstration (scoring comprehension during the learning itself, against specific objectives, as the byproduct of building the capability).

Only the third produces direct evidence. Inference is informed guessing, and assessment captures short-term recall but rarely transfer. Demonstrated learning evidence can be elicited inside the conversation, scored in real time, with the supporting evidence excerpted from the transcript: a comprehension signal the learner sees as they progress, a defensible score the L&D team can take to the CFO, and aggregated workforce data the board can use to evaluate whether the strategy has the capability it requires.

Perceptyx’s Develop, which launched today, has been purpose-built to produce that evidence. It is the only product in the market that generates validated comprehension as the byproduct of instruction rather than as a postscript bolted onto course delivery.

How does Develop work inside a learning conversation?

Five specialized AI agents coordinate through every conversation, each with a defined role and each supporting and feeding information into the next.

The Content Architect (Content Conversion Agent) builds the course from source material the company already owns. PDFs, slide decks, videos, SCORM files, leadership frameworks, and licensed courseware from Coursera, Skillsoft, and LinkedIn Learning are ingested and structured around learning objectives. L&D administrators retain control of those objectives, and the conversational training recompiles when they are edited. Existing content investments become the starting point rather than a sunk cost.

The Tutor (Adaptive Learning Agent) delivers learning as a one-on-one Socratic conversation, modeled on the Oxford tutorial that has produced original thinkers for nine centuries. Sarah’s session begins with a question rather than a static module. When she answers, the Tutor asks her why she thinks what she just said rather than handing her the answer or advancing to the next item. Its refusal to merely give the answer is the mechanism. The pedagogical team behind the product consists of Ph.D. researchers and university educators, including Dr. Glenn Platt, originator of the “flipped classroom.” The AI was built to deliver a learning methodology that already worked.

The Evaluator (Learning Validation Agent) runs an ensemble of more than ten learning theories that operate in tandem and guide the Tutor through the conversation. These theories include Dunning-Kruger calibration, which compares the learner’s stated confidence against the accuracy of what they said, surfacing the gap between feeling competent and being competent. Transfer of learning, drawn from the Perkins and Salomon framework, distinguishes recall in the original context from the ability to apply a concept in a novel one. Gaming-behavior detection, from Baker’s research on intelligent tutoring systems, catches surface engagement where learners game the dialogue without engaging the material. Levels of processing, from Craik and Lockhart, evaluates whether the learner is encoding the material at a shallow level or processing it deeply enough for retention.

The Development Advisor (Learner Insight Agent) synthesizes the session into a transcript-grounded report. Each learning objective receives a comprehension score, with strengths and concerns quoted from the conversation. Scores aggregate to a percentage rating where 100% is absolute mastery, with a non-linear curve that gets harder to move as the learner nears the top. Sarah sees that rating as a live progress bar during her session, and her manager receives the full report with coaching recommendations that previously required an expert one-on-one coach.

Together, the live learning agents — the Tutor, the Evaluator, and the Development Advisor — form a closed evidence loop: the Tutor poses questions, the Evaluator scores each response against learning objectives, the Tutor adapts to the score, and the Development Advisor synthesizes the session into evidence the moment it ends, running continuously as the conversation unfolds. Multi-agent coordination and Socratic delivery have become familiar market patterns, but only the closed loop produces validated comprehension as the byproduct of instruction.

The Skills Strategist (Workforce Insight Agent) aggregates patterns across the workforce. Where the Development Advisor produces individual evidence, the Skills Strategist surfaces course-level and organization-level patterns: which objectives are landing across populations, where cohorts are collectively stuck, and where capability gaps cluster across functions and business units. Leaders see not only what employees did and did not understand, but what is preventing learning from transferring, which points to where the workplace or the learning experience needs to change.

Develop’s job ends at validated comprehension. Turning that understanding into durable skill requires reinforcement at the moment of application, and often a change in how the work itself gets done.

What does this evidence mean for learners, L&D leaders, and executives?

Back to Sarah, our learner. She has finished a session that engages her critical thinking instead of testing her ability to memorize or cut and paste information. Every session produces an evidence-backed report for her L&D leader, with strengths and concerns drawn from the transcript and coaching recommendations that previously required an expert one-on-one coach. At the executive level, the data maps capability across functions, identifies where gaps are widening, and connects comprehension scores to the business outcomes the strategy depends on.

The architecture produces evidence at three altitudes from the same conversations. The workforce intelligence that the CHRO brings to the board has the transcript evidence underneath it.

How does learning become durable behavior change?

Cognitive science has documented for more than a century that unreinforced learning decays fast. Ebbinghaus’s forgetting curve, consistently replicated since 1885, finds that roughly 70% of new learning is lost within twenty-four hours if nothing pulls it back. Develop closes the understanding gap inside the learning session, but the transfer gap that opens once the learner returns to work requires Activate, the behavior engine of the People Activation System: in-flow nudges, AI coaching, and contextual reinforcement in the tools where work happens.

When Activate runs alongside Develop, it draws on the same content the Content Architect prepared, the same objectives the Evaluator scored against, and the same strengths and concerns the Development Advisor surfaced. Reinforcement targets what each learner struggled with, not generic prompts disconnected from the original learning.

Develop’s percentage comprehension score is paired with a percentage-based application score from Activate, which measures engagement with course-specific reinforcement in the days and weeks after the session. Together they show where the learning landed and where it is translating into behavior on the job. The Haunstrup and Jensen (2024) randomized controlled trial, conducted with 226 managers and 4,442 employees, found that training combined with just-in-time nudges produced durable behavior change still measurable at eight months and beyond, while training alone did not.

A few days later, another message arrives in Sarah’s Teams. This one is from Activate, connecting the concept she learned to a task she is working on that week. She can start a coaching session in the thread.

Why does proving that learning worked matter so much right now?

Every executive is making capability bets in 2026: AI adoption, transformation programs, mergers and acquisitions, new operating models, all of which depend on whether the workforce can build the skills and behaviors the strategy requires. In spite of this, the evidence available to evaluate progress is weak. The industry spends more than $400 billion per year on corporate learning, and whether the spend produces capability has never been answerable from LMS data.

The CFO and the board want an answer no L&D team has so far been able to give with complete confidence: did it work? Validated comprehension produces a category of evidence that did not exist before, with the audit trail to defend it.

What happens to Sarah, our learner?

Sarah’s session ends. She is unlikely to remember the specific clause of the compliance regulation a month from now, but the question of why do you think that? will stay with her, because the conversation made her work through her reasoning rather than pattern-match to the right answer. A week later, Activate pulls the concept back to help her apply it to something on her actual project. For the first time in twenty years of corporate learning, the system on the other side wanted her to think, stayed with her until the thinking changed, and connected what she learned to the work she does.

How can you validate comprehension across your entire organization?

Learn how validated comprehension works across the workforce on the Develop product page, walk through how Develop connects to Discover and Activate inside the People Activation System, and then talk to an expert to see what how this could be implemented in your organization.

The Power of Your People — Activated.