Insights / Learning and development
Learning in the flow of work has a ceiling
Learning in the flow of work is the most useful idea the field has adopted in a decade. It is also being asked to carry topics it structurally cannot hold.
By McKinley Malbrough III, J.D., MS-HRM. Published September 28, 2026.
Short answer
It handles procedural knowledge and recall better than any session can, and it cannot reach capability where the difficulty is another person. Sort by failure mode: if failure looks like the person did not know, use flow-of-work support; if failure looks like the person knew and could not do it, use a live session. Applied honestly that moves most of a catalogue into the cheaper column, which is the argument for the approach, and frees the live budget for the few topics that require rehearsal with correction.
What does learning in the flow of work actually mean?
Support delivered at the point of need rather than scheduled in advance. A three minute explainer surfaced inside the tool somebody is already using. A job aid that appears when a form is opened. A search result that answers the question the person has right now.
It works because it removes the two largest sources of decay in traditional training: the gap between learning and application, and the fact that scheduled training arrives when the calendar allows rather than when the need exists.
Where does it genuinely outperform a session?
Procedural knowledge with a correct answer. How to file the expense. Which field means what. What the approval threshold is this quarter. Nobody should sit in a room for any of that, and a room is measurably worse at it than a well-placed job aid.
It also outperforms on refresh. Capability decays, and a system that surfaces the reminder at the moment of use beats an annual session at precisely the moment the annual session is least present in memory.
What is the ceiling?
Anything where the difficulty is another person. Delivering news somebody does not want. Holding a room when two participants start arguing. Receiving a disclosure and not promising confidentiality you cannot deliver.
These are not knowledge gaps and cannot be closed by information arriving at the right moment. The person already knows what they should do. The gap is between knowing it and doing it while their heart rate is elevated, and that gap closes through rehearsal with correction or it does not close.
Is this just the 70-20-10 argument again?
It is adjacent to it and the distinction is worth holding. The 70-20-10 framing describes where learning happens across a career. Learning in the flow of work is a design approach for the seventy, and a good one.
The error is reading the seventy as evidence that formal delivery matters less. It does not follow. Experience teaches efficiently when the person has a model to hang it on and somebody occasionally correcting them, and produces confidently repeated bad habits when they do not.
What does the spending data suggest?
That the shift is already well underway and the quality question has not caught up. ATD's figures put direct learning spend at $846 per employee against $1,254 the previous year, with formal learning hours rising from 13.7 to 16.7. More hours, far less money per hour.
Some of that is genuine efficiency: content that should never have been a session became a resource. Some of it is behavioral topics quietly routed into cheaper formats that cannot deliver them, which shows up two years later as a manager capability problem that nobody traces back to a format decision. The HR guide on building manager capability is where that bill eventually arrives.
How should a learning and development team split the portfolio?
Sort by failure mode rather than by topic. If somebody failing at this looks like they did not know, that is flow-of-work territory. If somebody failing at this looks like they knew and could not do it, that needs a live session.
Applied honestly, the sort usually moves more of the catalogue into the flow-of-work column than people expect, which frees the live budget for the handful of topics that actually require it. That is the argument for the approach, not against it.
What is the strongest version of the combination?
Live session for the behavior, flow-of-work support for the recall. The manager practices the conversation in a room with other managers, then the structure they practiced is one search away on the day they need it eight months later.
That sequencing is also what makes the live session defensible on cost, because the expensive hour is doing only the work that nothing cheaper can do. How to measure whether training worked covers proving that once it is running.
Where these numbers come from
| ATD State of the Industry, 2026 | Direct learning expenditure of $846 per employee against $1,254 the prior year, with formal learning hours rising from 13.7 to 16.7. |
| TalentLMS 2026 L&D Report | Continued movement toward microlearning, self-paced modules and interactive group sessions as parallel modalities rather than substitutes. |
| 70-20-10 framework, Center for Creative Leadership | Descriptive model of where workplace learning occurs across a career, originating with the Center for Creative Leadership. |
Data current as of September 2026. Sources are linked where each figure appears.
Keep going
Why recorded training does not finish
The completion ceiling on the asynchronous half.
How to measure whether it worked
Proving the split was the right one.
Difficult conversations training
The clearest example of a topic the flow of work cannot reach.
Building a learning and development plan this quarter?
Book twenty minutes and bring the capability you are trying to move.