Recurring training is accurate for about a month

Most organisations that run mandatory training on an annual cycle are delivering something that stopped being fully correct several months earlier, and everyone involved knows it.

The pattern is consistent across regulated and unregulated work alike. The material is produced in the autumn. A threshold changes in February, an internal process is reorganised in April, and a system used in one of the walkthroughs is replaced in June. The training continues to run, because the alternative is to rebuild it, and rebuilding it costs what building it cost.

So a note gets added. Then a second note. By the time the next cycle arrives, the training carries a set of corrections that people read before watching the thing being corrected, which is an unusual way to teach anything.

The constraint is the update, not the production

Discussion of training video almost always concerns production cost, and production cost is the part that was already solved.

The number determining whether training stays accurate is the cost of changing forty seconds of it. When that number is high, the material drifts, because nobody authorises a reshoot to fix one threshold. When it is low, the material stays current, and the same team that tolerated a year of drift will fix something the week it changes.

Recorded video sets that number high for a structural reason. A recording is not modular. Changing one passage means matching a room, a microphone, a presenter’s delivery and a filming day that has passed, which is why the practical minimum unit of change is the whole video.

What actually changes between cycles

The proportion is worth measuring before deciding anything, and it surprises people.

Across a typical annual refresh, the material that genuinely needs to change is usually a small fraction of the runtime. Specific figures, named systems, a step order, the escalation contact. The explanatory material around it, why this matters, what goes wrong, what the person should do when uncertain, is generally still correct from one year to the next.

That ratio is the argument. An update process proportional to what changed would cost a fraction of a rebuild. An update process that is all or nothing costs a rebuild every time, which is why it does not happen.

Generating from the source rather than performing it

Tools acting as a training video creator invert the economics by making the document the master and the video a derived output. Leadde takes Word, PDF, PowerPoint or plain text, or pasted text, and generates an outline, narration and a sequence of scenes from it.

The consequence that matters here is not that generation is fast. It is that a change to the source document regenerates the affected material without touching anything else, so the cost of a correction is roughly the cost of editing a paragraph plus a review pass.

That changes what an organisation can reasonably commit to. Fixing something in the month it changes becomes a routine task rather than a project requiring a budget line.

The settings that decide whether it lands

Three optional context fields do more to the output than any other control, and left blank they produce narration addressed to nobody in particular.

The audience field determines what gets assumed. Described specifically, for example as a first year employee in a customer facing role who has not seen the internal system names, the narration defines terms before using them. Described vaguely, it produces a reading of the document.

The speaker background field sets the position the narration speaks from, which separates material that sounds like a colleague explaining a requirement from material that sounds like a document being recited.

The objective field determines emphasis. Generated with the objective of covering the policy, the result covers the policy. Generated with the objective of helping someone recognise the situation when it occurs and know what to do next, it emphasises different sections entirely, and the second is what recurring training exists for.

Register is worth setting deliberately. Leadde offers Formal, Neutral, Objective, Structured, Analytical, Explanatory, Narrative and Persuasive, and this material is served by Explanatory or Structured rather than by anything more assertive.

The document remains the authoritative version

This needs to be a rule rather than an intention, because the failure mode is specific and expensive.

The video is derived. When the source changes, the video is wrong, and confidently delivered narration describing a superseded threshold is worse than a stale paragraph, because nothing in the delivery signals that it is out of date. Corrections belong in the document, with the video regenerated rather than edited, so the two never diverge.

A review interval should be agreed before a library exists rather than after. Organisations discover the need for one at the point where they have forty videos and no record of which were reviewed when.

Whether any of this satisfies a particular obligation is a question for whoever owns that obligation internally. Generation tooling produces training material; it does not produce a compliance position, and no software vendor can supply one. The broader framing of workplace training duties is set out in the material published by the International Labour Organization for anyone mapping their own requirements.

What should not become video

The reasoning is the same in each case below: the format cannot do the thing the material exists to do.

Anything requiring a recorded acknowledgement should not, at least not on its own. The acknowledgement is the artefact, and it belongs wherever the organisation records such things.

Anything with contested interpretation should not. Where a requirement is genuinely ambiguous and the answer depends on circumstances, narration flattens it into a rule, and someone will apply the rule.

Anything consulted rather than learned should not. Escalation contacts, thresholds and reference tables are looked up, and a video is the worst place to look something up.

Anything requiring practice should not. Where competence means doing the task under supervision, video can prepare someone for the session and cannot replace it.

Access is not a separate workstream

Captions belong on the first version rather than being added when someone asks. A significant share of this viewing happens at a desk without headphones, and captions for prerecorded video are a WCAG success criterion.

Where a workforce spans languages, a finished video can be translated with the narration and on screen text handled together, producing a parallel version rather than replacing the original. For training that carries specific requirements, a translated version should be checked by a fluent speaker before use, because machine translation handles ordinary prose well and handles conditional requirements less reliably, and a conditional rendered as an absolute changes what the training says.

Start with the module that already carries corrections

The useful first candidate is not the newest material. It is the module with a list of amendments attached to it, rebuilt from a corrected source document.

If the corrections disappear into the material rather than sitting alongside it, the update problem was the real problem. If a fresh set accumulates within three months, the source document is changing faster than any format can follow, and the honest answer is a document rather than a video.

This article describes content and communication practice. It is not legal or compliance advice.