Workflow learning is the latest buzzword, but what does it mean? We’re seeing the term pop up everywhere, yet definitions are vague. Is there any there there, and if so, what is it?

The buzz is significant. Josh Bersin made a statement in 2018 that said learning in the workflow is a new paradigm (he actually registered the phrase “Learning in the flow of work”!). Ironically, Jay Cross was talking about workflow learning in 2004! We’ve seen a research report from The eLearning Guild on it this year, and we’re even seeing books opportunistically coming out. The important thing to do, then, is to define it, evaluate what the offer is, and then figure out if and how we might capitalize it.


A literal interpretation of the term is a start. It’s about learning in the workflow. We can compare that to the traditional model of learning, which is in the classroom. Here, instead of going away and simulating a work environment for practice, we are leveraging the real performance environment. And there are some important distinctions.

The advantages should be the lack of need to create the environment, nor makeup tasks, and build in motivation. These are all a given. What do we need to make sure it’s about learning?

One of the common interpretations of workflow learning, as discussed in The Guild report, is that it’s about support in the moment to succeed. Which is, of course, a good thing. It’s also typically not learning! It’s performance support, and works perfectly fine if they need it the next time. However, it’s not learning unless you deliberately make it so. Just seeing it again and again doesn’t ensure it’s learned.

From Friston’s Free Energy Principle, we see that our learning is driven by trying to match what we expect with what actually happens. To do that, we use models. Thus, if it’s learning, there’re models that explain why to do it this way, and provide a basis to predict outcomes and make better decisions.

Making workflow learning thus includes layering on this explanation of why we’re doing it this way. There can be a model that’s referenced (that, really, has to be already known, as that’s likely a longer task). It can be in the form of a video that shows how to do it (and explains something as well), text, annotated image, the medium doesn’t matter. What matters is that there’s additional depth being delivered about why you do it this way, not just what to do.

The reason to do this, of course, is to support performing more flexibly at a later time. If you’re helped now, and understand why, you are likely to be able to respond more flexibly later. And, importantly, in situations where the content hasn’t been developed to cover. As we face increasingly changing times (echoing the concept of VUCA: volatile, uncertain, chaotic, and ambiguous), becoming more agile over time is an increasingly essential ability.

A conflict is with the term ‘microlearning’. Which, too, has several different meanings. One is performance support just like workflow learning, and again it’s useful, but it’s not learning. The other is really ‘spaced learning’, learning distributed over time. But microlearning isn’t workflow learning, as it’s formal learning that comes in small chunks, but isn’t typically tied to the particular context.

There is one sense in which microlearning and workflow learning can interact, and that’s when it’s really about a small amount of content around a task that makes it a learning moment. Overall, however, the problem is that there’s still not a strong ‘learning’ component in most of the promotions around workflow learning. To do it right, we start to think about the mode of delivery. The easy solution is ‘pull’, but there are increasing opportunities for ‘push’.


How can we support workflow learning? One option is to have resources available ‘to hand’ for performers to access. What we’re doing here is curating resources that individuals can access when they have a problem. It could be in any form, but it’s about meeting a learning need in the moment.

So, resources could include little bits of prose, video, an animation, even a voice-over (particularly if the task is directionally critical, that is, you can’t take your eyes away). It could also be a person. In a discussion, Bob Mosher and Conrad Gottfredson opine that coaching typically can’t count because a coach can’t always be there in the moment. Still, being able to call upon help as needed is a potential mechanism as well.

Here, workflow learning’s based upon either searching or browsing. A critical element then becomes that it’s findable. Too often, people know a resource exists, but still can’t find it. To remedy this, it helps to have resources oriented around tasks and roles, not around silos. If someone’s selling a product, for instance, they want the product info in their resource spot, not off somewhere in engineering’s docs. Federated search is one solution, where the search covers all resources, not just one. Redoing portals is another approach. The point is that you can either browse in a user-centered way, or you can do a search.

Another element is what I call ‘the least assistance principle’. When in the flow of work, people don’t want everything, they want just enough to get past this current need. Understanding the task and need is important in ensuring that the content is no more than required to help the performer succeed, and understand better to succeed faster and more adaptively in the future.


Another opportunity is if you know not only the task, but also the performer’s knowledge about the task. Here you could customize what’s presented to be just the right thing to further their development. So, it’s ‘spaced learning’, customized by leveraging the current context and learner model. Which requires a content model. This is the ‘push’ option.

Here, we’re being proactive about providing assistance (whether support or learning). We’ve coded up ‘triggers’ that let us know when someone’s doing something (either their access of particular software, or via their calendar, or assignments like To Dos). And we also have a model of their abilities in a variety of tasks (against competency standards, for instance). Then, we can assess how much help they might need in this task, and push appropriate content.

This ventures into the realm of AI (artificial intelligence) as a backend, but is really an application of IA (intelligence augmentation). (Of course, once you know how to do it, it’s no longer AI.). The point being that we can, with tags and rules, develop recommendations and act on them.

Ultimately, the goal is to couple content and systems with people to help them perform better in the moment and develop them over time. Workflow learning takes advantage of a meaningful context to minimize the overhead necessary to support learning. However, it takes considerable work to do it appropriately, instead of missing the mark with content that is mismatched to need on a variety of dimensions. I’m all for ‘workflow learning’, but it has to be smart, otherwise it becomes another hype bubble that misleads and wastes money. Here’s to conceptual clarity, and continual improvement!