For thirty years the thing that decided who published training was that it was hard to make. That constraint has gone — a subject-matter expert can now turn a process document into a finished course before lunch. Removing the difficulty also removed the accidental governance that came with it. This piece sets out a practical replacement: tier content by consequence rather than by author, so scarce review capacity goes where being wrong is expensive, and let everything else publish freely.
The constraint that was doing the governing
For thirty years, one thing quietly decided who published training at your organisation: it was hard to make. Building a course took an instructional designer, an authoring tool, a licence, and a fortnight. That difficulty was never a governance policy, but it worked like one. Only a few people could publish, so only a few people did.
That constraint has gone. A subject-matter expert can now turn a process document into a finished, interactive course before lunch. This is genuinely good — the person who knows how the escalation actually works is a better source than someone interviewing them about it. But removing the constraint also removed the accidental governance that came with it, and most organisations have not yet replaced it with a deliberate one.
So the question is live, and it is not rhetorical: who should be allowed to publish training?
The bottleneck moved; it did not disappear
The instinctive framing is that faster production means more training, sooner. What actually happens is that the constraint relocates.
Ten course requests used to mean ten courses, produced slowly, by people whose full-time job was producing them. Now those ten requests can become fifty assets — a course, a job aid, a video, a quick-reference card, a scenario — produced in a week by fifteen different people across four departments.
Every one of those assets still has to be correct. Every one still has to agree with the other forty-nine. And the people who can judge that are the same small group who were the bottleneck before.
Production got cheaper. Review did not. If you plan only for the first half of that sentence, you get a content library that grows faster than anyone can verify it.
Three ways it goes wrong
Contradiction
Two teams publish guidance on the same process six weeks apart. Both are plausible. Neither is marked as superseding the other. A new joiner finds whichever ranks higher in search and follows it. Nobody notices until something goes wrong — and then the investigation finds two official-looking answers and no way to tell which one the organisation stood behind.
Volume as noise
When publishing is free, publishing becomes the default response to every problem. The result is not a better-informed workforce — it is a workforce that has learned to ignore training notifications, because most of them are not for them. The cost of low-value content is not the cost of making it. It is the attention it takes from the content that matters.
Fluent inaccuracy
This is the one specific to AI-assisted authoring, and the one to take most seriously. Generated content is confident by construction. It reads as though it was written by someone who knew, because it was trained on writing by people who knew. What it cannot know is the exception that applies only at your Chennai site, or only to the regulated version of the workflow, or only since the policy changed in March. A human first-drafter who does not know something tends to write around it. A generator fills the gap smoothly, and the gap is invisible in the output.
None of these is an argument against letting subject-matter experts create training. They are arguments for deciding, in advance, what happens between "created" and "published".
The wrong correction
The reflex, once a bad asset gets through, is to route everything through L&D for approval.
This fails for a predictable reason: it recreates the bottleneck you just removed, and it does it at the new, higher volume. A three-person team cannot meaningfully review fifty assets a week. What happens instead is rubber-stamping — review that exists on the process diagram and not in practice — which is worse than no review, because it manufactures false confidence.
Blanket approval is not governance. It is a queue.
Tier by consequence, not by author
The useful question is not who wrote this but what happens if it is wrong. That single reframing does most of the work, because it lets you spend scarce review capacity where being wrong is expensive.
- 1Tier 1 — low consequence. Publish freely.
A walkthrough of the new expense tool. A refresher on booking meeting rooms. If this is wrong, someone is briefly confused and asks a colleague. Let the person closest to the work publish without a gate — requiring review here is how you lose the credibility to require it anywhere.
- 2Tier 2 — operational consequence. One named reviewer before publish.
Customer-facing process, sales positioning, anything that changes how someone does their job. If this is wrong, work gets done wrong at scale until somebody notices. One reviewer with real domain knowledge, named in advance so responsibility does not diffuse. Not a committee — a person.
- 3Tier 3 — regulatory, safety or contractual consequence. Specialist control.
Compliance, safety procedures, anything you may have to produce evidence for. Authoring can still start with a subject-matter expert — it usually should. But sign-off is a qualified specialist, the version is controlled, and the completion record has to be defensible to someone outside the organisation.
Tier 1 is bigger than you think
Most teams find that tier 1 covers far more of their content than they expected, which is the point. You are not adding process — you are concentrating the process you already had onto the third of your content that actually warrants it.
Key takeaways
- Use the roles you already administer. If authoring permissions live in a different system from your learner records, you now maintain two permission models and they will drift.
- Keep publish distinct from save. If building and publishing are the same action, tiering is unenforceable — there has to be a state where content exists but is not yet live.
- Keep one current version. Updating should replace what learners see, not add a second copy alongside it. A library with five versions of the same guidance has no governance, whatever the policy says.
- Give every asset a named owner and a review date. Content does not become wrong on the day it is published; it becomes wrong quietly, eighteen months later, when the process changes and nobody remembers the course exists.
- Keep one completion record. If different content is tracked in different systems, you cannot answer 'did this person receive the current guidance' without reconciling by hand — which is exactly what you cannot afford when someone is actually asking.
Frequently asked questions
Who should be allowed to publish training?
The useful question is not who wrote it but what happens if it is wrong. Tier content by consequence: low-consequence material publishes freely from whoever is closest to the work, customer-facing or process-critical material gets a named reviewer, and anything with legal, safety or regulatory weight goes through formal sign-off. Gating by author recreates the bottleneck you just removed.
What is tier-based training governance?
A model that sorts content by the cost of being wrong rather than by who produced it. Each tier gets a different level of review, so scarce reviewer time is spent where an error is expensive and everything else moves without a queue.
Does AI-generated training need more review than human-written training?
It needs a different kind of attention. Generated content is confident by construction and reads as though the author knew, so the gaps are invisible in the output — the exception that applies only to one site, or only since a policy changed. A human first-drafter who does not know something tends to write around it. That is an argument for checking against local exceptions, not for routing everything through one team.
Should all training go through L&D for approval?
Blanket approval is not governance, it is a queue. A small team cannot meaningfully review fifty assets a week, so what happens in practice is rubber-stamping — review that exists on the process diagram and not in reality, which is worse than no review because it manufactures false confidence.
How do you stop two teams publishing contradicting guidance?
Name an owner per process rather than per asset, and make superseding explicit so a new version visibly retires the old one. Contradiction is usually discovered during an investigation, when two official-looking answers exist and nothing records which one the organisation stood behind.
How do we start without writing a policy first?
Sort the last twenty things you published into the three tiers — it takes about an hour and surfaces the disagreement while it is still cheap. Then name the tier 2 reviewers by person rather than by team, and find your oldest published course and ask who owns it.
Where to start
You do not need a policy document to begin. You need three conversations.
- Sort the last twenty things you published into the three tiers. This takes an hour, and it is usually where the disagreement surfaces — which is the point of doing it before you write a policy rather than after.
- Name the tier 2 reviewers. By person, not by team. "L&D will review" is how nothing gets reviewed.
- Find your oldest published course and ask who owns it. If the answer is a pause, that is the gap — and it existed long before AI authoring did.
The organisations that will handle this well are not the ones that adopt the tooling most aggressively, nor the ones that resist it longest. They are the ones that decide what needs a gate before the volume arrives — while the question is still theoretical and the answer is still cheap.
The EdzLMS SCORM Builder is built around the mechanics above: authoring runs inside the LMS using the roles you already administer, publishing is a separate step scoped to a real course, and re-publishing updates one version in place. See how it works →
Want to see authoring, review and publishing run on the roles you already administer?
Book a Free Demo