A pharma distributor buys a forty-module video library for its field managers. Month one, 61 per cent of licences are opened. Month three, 12 per cent. The L&D head is asked in a review why the spend has not changed anything, and the honest answer is that nothing was designed to change. Forty videos were made available. Availability is not a blend.
The failure is not the videos. It is that no stage of the programme needed any other stage. A manager could skip module nine, attend the workshop, nod, and return to work exactly as before, and no artefact, no colleague and no report would ever show it. Variety without dependency is a catalogue. A blend is a sequence in which each step consumes the output of the previous one.
The evidence says the gain was never the medium
The largest review of this ground is the US Department of Education’s 2010 meta-analysis of online learning studies. It pooled 50 independent effect sizes from 45 studies. Instruction that combined online and face-to-face elements beat purely face-to-face instruction with a mean effect size of +0.35 (p < .001). Purely online instruction beat face-to-face by only +0.05, which was not significant (p = .46).
That is the number most vendors quote, and it is the wrong half of the finding. The same report notes that the blended conditions “often included additional learning time and instructional elements not received by students in control conditions”, and states plainly that the positive effects “should not be attributed to the media, per se”. The moderator analysis makes the point sharper. Where online learners spent more time than the classroom group, the mean effect size was +0.45. Where the classroom group spent as much or more time, it fell to +0.18.
Read that as a design instruction, not a media argument. The gain comes from the extra structure, the extra practice and the extra time that a well-built blend forces into the calendar. A video library that people open for eight minutes on a Tuesday adds none of those. It adds a page view.
Why skipping is invisible in most blends
Three design habits make every stage optional, and they are common enough to be a pattern.
The stages are interchangeable. Module, workshop, e-learning, coaching session: each one stands alone, each one can be consumed in any order, and none of them refers to what the learner did before. If you can shuffle the sequence without breaking anything, you do not have a blend. You have a menu.
The live session re-teaches instead of using. When pre-work is optional, the facilitator has to assume nobody did it, so the first ninety minutes repeat the module. Learners learn quickly that the pre-work is decorative. Skip it once, lose nothing, skip it always.
Nothing lands on the job. The programme ends at the end of the classroom day. There is no task with a due date, no artefact a manager looks at, no moment where the new behaviour is either there or absent. Without that, the only thing anyone can report is attendance, and attendance is the metric of a programme that has given up.
The chain: four cells, filled in order
The method is to design backwards from a single job task and then force each stage to hand something to the next one. Do it on one page before you buy or build anything.
| Stage | Learner produces | If skipped, this breaks | How a manager sees it |
|---|---|---|---|
| Self-paced module (20-30 min) | A completed one-page diagnosis of their own team using the module’s four categories | The live session has no cases to work on | Diagnosis sheet is the entry ticket |
| Live session (half day) | A rehearsed script for one real conversation, plus a peer’s written critique | The conversation never gets rehearsed | Script is shown to the reporting manager |
| On-job task (next 10 working days) | One completed conversation, logged with date, person and outcome | Nothing changes on the floor | Manager initials the log line |
| Retrieval checkpoints (day 21 and day 60) | Three answers to a scenario, scored against anchors | The skill decays quietly | Score is visible on the cohort sheet |
The first cell is the one people skip, and it is the one that carries the design. Write the job task as an observable sentence: “runs a fifteen-minute performance correction conversation with an operator who has missed two targets, and leaves with a written next step”. If you cannot write that sentence, you are not ready to commission content. Once it exists, every stage has an obvious output and an obvious consumer.
Then apply the skip rule. The live session opens by putting four learner diagnoses on the wall and working them. Anyone who did not bring one works on a blank sheet and cannot contribute, which is uncomfortable in a way that a reminder email is not. You do not need to punish latecomers. You need the session to be unusable without their input.
Put the retrieval where the forgetting is
The second design decision is spacing, and it is usually made by accident. The common pattern is a single refresher three weeks later, chosen because it fits a calendar.
The evidence is more specific than that. Cepeda and colleagues reviewed 839 assessments of distributed practice across 317 experiments in 184 articles. Two findings matter for programme design. First, spacing beat massing consistently. Second, the optimal gap between study episodes grows with how long you need the knowledge to last: at retention intervals of six months or more, the optimal interval was at least a month, and the authors concluded that for practical purposes the optimal interval “will likely be well in excess of 1 day”.
Translate that into a calendar. If a supervisor needs the conversation structure to still be intact in six months, one refresher at week three is not the design. Three retrieval points, widening: roughly day 3 (recall the structure), day 21 (apply it to a new scenario), day 60 (apply it under pressure, with a curveball). Each is ten minutes, not a module. Each is scored, and the score is what you show the business instead of a completion percentage.
This is also where an e-learning build earns its cost. A short scored scenario at day 21 is cheap to deliver digitally and impossible to deliver in a classroom without pulling forty people off the floor. Keep the digital part small and punctual, and spend the expensive human hours on the rehearsal.
A worked example, illustrative
Numbers below are illustrative, built to show the arithmetic rather than to report a client result.
A 180-supervisor programme across six plant sites, delivered through a blended programme design with two trainers rotating. The old version was twelve videos and a two-day workshop. Reported completion was 94 per cent, because completion meant a video had played to the end.
The rebuilt version: one 25-minute module, one half-day session per cohort of 20, one logged conversation in ten working days, two scored retrieval points.
- Pre-work return rate: 156 of 180 brought a completed diagnosis (87 per cent). The 24 who did not were moved to the next cohort rather than absorbed into this one.
- Time per supervisor, total: 25 minutes plus 4 hours plus 10 minutes plus 20 minutes, about 5 hours across ten weeks. The old version was 16 hours in two blocks, so the new one is shorter and harder.
- Artefact return: 141 of 156 logged conversations (90 per cent). Nine of those logs recorded a conversation that did not go as planned, which is the useful data.
- Manager verification: 118 of 141 logs were initialled by a reporting manager. The gap of 23 is the honest number in the whole programme, and it is the one to take to the review meeting.
- Retrieval scores rose from a mean of 4.1 of 10 at day 3 to 6.8 at day 21 and 7.4 at day 60, on the same anchors. The rise, not the absolute score, is the claim.
Compare that with the old reporting. Both programmes can claim a completion number. Only one of them can tell you how many supervisors had the conversation and how many of those conversations were seen by anyone senior.
The honest limit
This design costs manager minutes, and manager minutes are the scarcest thing in the building. If reporting managers will not initial a log line, the chain breaks at the third cell and you are back to a video library with better paperwork. In that case, do not pretend. Substitute peer review: pairs exchange scripts and logged outcomes, and the pair attests. It is weaker evidence, and it is still evidence.
It is also the wrong design for pure compliance content, where the requirement is that everyone has been told, and a well-made video plus an acknowledgement is a legitimate answer. Applying a dependency chain to a POSH refresher wastes everybody’s time. The test is whether the programme is trying to change what someone does on a Tuesday, or to record that they were informed. Two different products.
One more limit worth saying out loud. The effect sizes above come from education research, mostly on students, not from Indian corporate cohorts. They describe a direction of effect, not a promise of a number. Anyone quoting +0.35 at you as an expected return on a training budget is overselling a meta-analysis.
What to do on Monday
Take the blend you are currently running and draw the four-cell table on one page. Fill in what the learner produces at each stage, what breaks if it is skipped, and how a manager would see it. Most programmes have one filled cell and three blanks.
Then fix the cheapest blank first. If the live session has no input from the module, change the opening twenty minutes this week: ask for the diagnoses to be brought, and work them. That single edit converts a library into a sequence, and it costs one facilitator briefing. For the wider build, a training needs conversation is usually where the job task sentence gets written properly, and where the multi-site consistency rules decide what stays fixed when the chain runs across six plants.
Sources
- US Department of Education, Evaluation of Evidence-Based Practices in Online Learning: A Meta-Analysis and Review of Online Learning Studies (revised September 2010): https://www.ed.gov/media/document/evaluation-of-evidence-based-practices-online-learning-meta-analysis-and-review-of-online-learning-studies-revised-september-2010-107159.pdf
- Cepeda, Pashler, Vul, Wixted and Rohrer, Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis, Psychological Bulletin (2006): https://augmentingcognition.com/assets/Cepeda2006.pdf
