Just-in-time training answers "how do I do this?" at the moment the user needs it, on the screen where they need it, instead of storing the answer in a course they took in January. It works because timing is the strongest variable in software learning: instruction at the moment of application gets used, and used knowledge sticks.
Why the moment of need beats the scheduled course
A course teaches forty things in the order the curriculum chose; the learner needs one of them, three weeks later, and by then it has evaporated, skill decay is brutal for procedures learned but not used. The "how do I" moment inverts every disadvantage: motivation is maximal (the user is mid-task and blocked), context is exact (the real account, the real data, the actual state), and application is immediate, which is precisely the condition under which learning consolidates. Scheduled training builds foundations; the moment of need is where fluency actually accretes, one unblocked task at a time.
What can answer a "how do I" today?
Four mechanisms, in ascending order of finish. Documentation and search: essential reference, and honestly good when the user knows the right words; the failure mode is translation, matching their situation to the article's. Help chat over the knowledge base: faster retrieval, but blind, it answers a description of the screen, not the screen. Guidance overlays: can point at the right control if someone authored a flow for that exact question in advance. A live training agent: sees the user's actual screen and state, answers by voice or text, shows the path on their interface, and can walk it with them, then, if the user wants, turns the answer into a sixty-second micro-lesson with one supervised repetition so the question doesn't come back tomorrow.
From answered questions to a curriculum
Here's the compounding move most teams miss: every "how do I" is a data point about what your training didn't cover, or didn't stick. Aggregated, the question stream is the most honest needs analysis you'll ever get, ranked by frequency and cost. Route repeat questions into the curriculum: the workflow asked about weekly becomes a practice-mode scenario; the question that precedes churny behavior becomes part of onboarding's agenda. Just-in-time and just-in-case training stop being rivals and become one loop: sessions build the base, moments patch it, and the patch log rewrites the sessions.
Getting the balance right
Don't dissolve all training into moments; some skills need structured sequences before the moment of need ever arrives (a rep can't just-in-time their way through a compliance workflow's first run). The design rule: agendas for foundations, moments for everything the agenda didn't predict, one system remembering both. (Disclosure: Skippr's Skippr AI training does both jobs, agenda-driven sessions plus always-on "how do I" answers in the product, with one memory per learner, measured on users trained.)
Questions buyers actually ask
Is just-in-time training the same as in-app help?
In-app help answers the question. Just-in-time training answers it, shows it on the user's own screen, and lets them perform it once supervised, which is what stops the same question recurring.
Does just-in-time replace onboarding and courses?
No. Foundations still need structured sequences. Just-in-time covers what the curriculum couldn't predict and reveals what it should teach next.
Watch it train someone
Completion is not competence, and the difference shows up in the product rather than the dashboard. Fifteen minutes is enough to tell them apart.