Contextual learning in Salesforce: training at the point of need

Most corporate training runs on the calendar's schedule, not the job's. The 45-minute session happens on a Tuesday, in a room or a video call, away from any real work. The question it was supposed to answer shows up on Wednesday, in front of a live record, with a customer waiting. By then the session is a fading memory, and the rep does what everyone does: guesses, or asks a colleague.

Contextual learning is the alternative: training embedded in the work and triggered by the work, sized to fit the gap it fills. We first made this argument in 2020, and both the evidence and the platform have moved since. This is the rebuilt case: what the term actually means, what the memory science supports (including one viral statistic it doesn't), and the two architectures for putting learning inside Salesforce.

This article is part of the complete guide to running an LMS inside Salesforce and feeds the sales & service readiness guide.

What contextual learning actually means

Contextual learning is training delivered inside the context where it will be used: the same system, the same record, the same task, at or near the moment of need. The context does two jobs at once. It decides when the learning appears, because the work itself signals the need. And it improves how well the learning sticks, because the learner encodes the knowledge alongside the situation they'll retrieve it in.

The best-known framing of the idea belongs to Josh Bersin, who coined "learning in the flow of work" in June 2018. His argument: a large share of professional learning, roughly half by his account, already happens at the point of need rather than in formal programs. The coinage named something practitioners had felt for years: the learning that changes behavior is rarely the learning on the training calendar.

Three neighboring terms get tangled here, so a quick untangling. Microlearning describes the format: short, single-topic units, typically 30 seconds to a few minutes. Just-in-time training describes the timing: content on call whenever a question comes up, usually meaning the learner can pull it up quickly. Contextual learning subsumes both and adds placement: the short unit, at the right time, delivered inside the work context itself. There is no separate destination the learner must remember to visit. A two-minute video is microlearning anywhere; it becomes contextual learning when it appears on the record that raised the question.

That placement requirement is what makes the concept an architecture problem, not a content problem, and it's why the rest of this article is about where the training system lives.

The science, honestly

The memory research genuinely supports contextual learning, and it does so without the invented statistic that dominates vendor content. Let's take the real numbers first, then the fake one.

Forgetting after a single exposure is real and steep. The primary modern evidence is Murre & Dros's 2015 replication of Ebbinghaus's forgetting curve, published in PLOS ONE. The savings method is Ebbinghaus's own measure: the time saved when relearning material compared to learning it fresh. On that measure, the replication found savings of about 58% at 20 minutes and about 33% at one day. In other words, most of the forgetting happens within hours, not weeks. A one-off training session leaves the learner with a fraction of the material by the next morning, before any of it has met a real customer.

Now the fake number. The claim that "employees forget 90% of training within a month" appears everywhere, and the primary research does not support it. Ebbinghaus's original data showed roughly 20% savings remaining at 31 days, meaningful retention, not a near-total wipe. The replication's own 31-day measurement came in lower, and its authors flagged that point as an outlier, likely a measurement artifact, rather than evidence of steeper long-term forgetting. Vendor content keeps repeating the 90% figure anyway, usually without a citation that survives a click. The honest curve is dramatic enough: you don't need to invent a cliff when the real slope already makes the case.

And the case the real curve makes is for spacing, in context. A 2021 meta-analysis by Latimier and colleagues covering 29 studies found that spaced retrieval practice outperforms massed practice with an effect size of g = 0.74. That is a large effect by any convention in the field. The pattern holds with expert professionals, not just students in labs. A 2024 study of practicing physicians in Academic Medicine found spaced learners retained 58.0% versus 43.2% for massed learners (d = 0.62). The advantage persisted in on-the-job transfer measured ten quarters later: 58.3% versus 52.4%.

Put the two findings together and the design conclusion writes itself. Forgetting after one exposure is fast, so a single session cannot carry a skill. Spaced, repeated retrieval is what durable learning looks like. And the cheapest way to space learning across weeks is to attach small pieces of it to work that naturally recurs. Contextual learning isn't a trend wearing science as decoration; it's the delivery model the science points at.

The 24-minute problem

The hardest constraint in corporate learning is not motivation or content quality; it's time. Bersin research from 2015 put the number memorably: the average employee has about 24 minutes a week available for formal learning. Whatever the precise figure is in your organization today, the shape of the constraint hasn't changed: formal training competes with the actual job, and the job wins.

What has changed is how much organizations say they care. In LinkedIn's 2025 Workplace Learning Report, learning and development ranks as the number one strategy companies cite for retaining talent, ahead of every other lever. And learners themselves called this preference years ago. In LinkedIn's 2018 report, 49% of employees said they'd rather learn in the moment the job requires it: the stat we cited in 2020. That preference has only become easier to serve since.

Hold those facts side by side and you get the tension that defines the field: learning is a top strategic priority that receives roughly half a lunch break per week. There are only two ways out. You can fight the calendar for more dedicated training time, a fight L&D has been losing for decades, or stop treating training as something that happens outside the work. Twenty-four minutes is a poor budget for a classroom program. It's a generous one for contextual learning: a two-minute module between calls, or a short video on the record that raised the question. This model doesn't ask for time the job won't give; it fits inside the time the job already contains.

Two ways to put learning in the flow of Salesforce

For teams that work in Salesforce, there are two architectures for this kind of training. Overlay tools sit on top of the screen; a native LMS keeps its training objects inside the org. They solve different problems, and each deserves a fair hearing.

Overlay tools, the digital adoption platform (DAP) category, inject guidance into the browser: tooltips on fields, and step-by-step walkthroughs that point at the button you're looking for. For what they're designed for, they are genuinely good. If the problem is "reps don't know which field to fill or where the new quote flow starts", an overlay answers it at the exact pixel where the confusion lives. No LMS of any architecture beats that for UI guidance.

The limitation is architectural, not qualitative. An overlay is a separate product with a separate backend: its walkthroughs, its usage data, and its completion events live in the vendor's cloud, outside your CRM. That has three consequences for structured training. You can't report on it with Salesforce reports next to your pipeline. You can't trigger it from your own automation based on record data, or feed its outcomes back into records. And you can't build certification on it, because "watched the walkthrough" never becomes a governed record in the org that runs your business. Overlay data can be synced across, but a synced copy is an integration to build and trust. That is the general problem with bolted-on training stacks, and the native versus integrated comparison unpacks it in full.

A native LMS inverts the placement. Courses, enrolments, completions, scores, and certifications are Salesforce records in the same database as the Opportunities and Cases the training is about. Training can then do things only data in the org can do. It can appear on record pages according to the data already on the record. It can launch from the same automation that runs the business, and land in the same reports as revenue. The trade is real: a native LMS won't draw an arrow at a button the way a DAP will. But for structured, tracked, consequential training, the work's own data model is the context that matters.

The practical summary: overlays put help on the screen; a native LMS puts training in the data. If your training problem is "which button", buy the overlay. If it's "is this team ready, and can I prove it", the rest of this article is for you.

What contextual learning looks like natively

Natively, training stops being a destination and becomes a layer of the CRM itself, and the clearest way to show it is a pattern running in production today.

One customer wired product pitch training directly into its sales pipeline. When an Opportunity moves into a specific stage, the sales pitch video for the product on that deal appears directly on the Opportunity record page. The video is selected by the deal's stage and the Opportunity's record type. The rep never opens an LMS and never searches a catalog. The pitch for the product they're about to sell arrives on the deal itself, at the pipeline moment where a pitch is exactly what happens next.

Mechanically, it's a learning component on the record page, driven by the record's own fields. Conceptually, it's the forgetting curve answered in kind: the refresher lands minutes before use, at the top of the curve, not weeks after a session at the bottom.

The same shape repeats across both audiences. On the sales side, it's pipeline-driven: content chosen by stage, product, or account attributes, surfaced where sales enablement actually happens, on the deal. On the service side, it's case-driven: a micro-module on handling a case type an agent hasn't seen before, or product-update refreshers surfaced in the console. These are the patterns that make support training continuous instead of quarterly.

Around the record-page moments, the rest of the layer fills in. A Learn tab sits in the app navigation next to Opportunities, an always-available surface for the full catalog when the learner goes looking rather than waiting to be found. Micro-modules sized for calendar gaps let a learner clear a two-minute unit between calls, the same slot-sized approach that makes onboarding inside the CRM survive a new hire's first chaotic weeks. On mobile, those same modules travel with field teams, so a rep can review a product refresher in the parking lot before walking into the account. And because every trigger is a record change, assignment itself can be event-driven. A stage change, a new case type, a product launch flag: each can enrol the right people automatically. The mechanics are covered step by step in the Salesforce training automation guide.

We designed for this deliberately: the learner-facing pieces are components that drop onto record pages and read the host record, because the goal was never a better training destination. It was training the work summons by itself.

Closing the loop: did it work?

The final advantage of contextual learning in the CRM is that "did it work?" becomes a report instead of a research project. Every completion, score, and certification in a native architecture is a Salesforce record, related to the same people and the same pipeline the training was meant to move. The measurement principle follows directly: measure training where you measure revenue.

Concretely, the questions that are unanswerable in a separate training silo become queries. Did the reps who watched the stage-triggered pitch progress those deals differently from the reps who skipped it? Which case types still generate escalations after agents complete the module, a content-gap signal no satisfaction survey will hand you? Is certification coverage on the new product line keeping pace with the deals being opened on it? None of these require an integration, an export, or a data team. They're reports joining training records to business records that were never in different systems to begin with, on the same learning management platform the training runs on.

This is also the real test of the whole approach. Contextual learning promises to change the learner's very next action; an architecture that can't observe the behavior can only promise. One that shares a database with the work can check.

FAQ

What is contextual learning?

Contextual learning is training delivered inside the context where it will be used: the same system, task, or record, at or near the moment of need. The placement does double duty, timing the learning to a real trigger in the work and improving retention by encoding knowledge alongside the situation it will be retrieved in. In a CRM, that means content surfaced on the records and moments of the job itself.

What is learning in the flow of work?

"Learning in the flow of work" is Josh Bersin's June 2018 term for learning that happens inside the workday and the work tools rather than in separate formal programs. It names the observation that much professional learning already occurs mid-task, and it argues that delivery should follow the learner into the tools the job already lives in.

What is just-in-time training?

Just-in-time training is training made available exactly when a question comes up, rather than scheduled in advance of it. It's a timing concept: the learner gets the answer while the question exists. Contextual learning goes one step further by also fixing the place: the same content, delivered on the record or task that raised the question.

Does microlearning actually work?

Short-format learning works when it's spaced and retrieved, which is precisely what the research measures. A 29-study meta-analysis found spaced retrieval practice beats massed practice with a large effect (g = 0.74), and the result holds for practicing professionals, not just students. Microlearning succeeds not because short is magic but because short units are what make spacing practical inside a working week.

What's the difference between contextual learning and microlearning?

Microlearning describes the format, short single-topic units, while contextual learning describes the placement and timing: learning delivered inside the work context, timed to the task at hand. Most contextual learning uses microlearning formats, but a short video sitting in a catalog nobody visits is microlearning without context. The automation patterns that trigger content from record changes are what turn one into the other.

Conclusion

Training that arrives weeks before it's used loses to the forgetting curve every time. The real curve, the replicated one, is steep enough that no single session survives it. Contextual learning wins by changing the delivery contract: small units, spaced by the natural rhythm of the job, embedded in the tools the job runs on. In Salesforce that's an architecture choice. Overlay tools guide the click; a native LMS makes training part of the CRM's data. A deal can then summon the pitch it needs, as one customer's stage-triggered videos already do. And the results land next to the revenue they're meant to move.

About the author. Brice Mbouani is the founder of Daniwoo and a Salesforce engineer. He designed and built the training engine described in this article, which runs in production Salesforce orgs at organizations including ENG Group. The deployment story and product-design reasoning recounted here are from his own work.