Internal training: the videos and PDFs nobody finds (and the new hires who always ask the same things)

The “training” Drive: a graveyard

Open your company’s shared “Training” folder. What do you find? Almost certainly: a handful of videos recorded two years ago, a few PDFs with procedures you don’t know if they’re still valid, three versions of the same document with names like “procedure_v2_FINAL_ok_done”, slides from a course someone ran once, and a nested-folder structure that only made sense in the head of the person who built it — who then left. It’s a graveyard. Dead material nobody watches, where nobody finds anything, and that exists only so you can say “we have training”.

Meanwhile, in real life, training happens anyway — in the most expensive way possible: out loud, over and over, interrupting the people who are actually working. The new hire doesn’t open Drive (or they open it, find nothing, and give up): they ask the colleague next to them. That colleague, for the third time this month, explains the same thing. Whoever sits at the desk answers the same questions they’ve already answered a hundred times. The company’s knowledge lives in a few people’s heads, who get interrupted continuously to pour it out verbally, one chunk at a time, always the same things.

This has a precise cost, and it’s higher than you think. Every new hire who takes weeks to become autonomous — because they have to extract information by force of questions — is their unproductive time plus the time of colleagues who stop to explain. Do a rough count: if an experienced colleague gets interrupted even just five times a day for ten minutes explaining the usual things, that’s almost an hour a day lost — plus the new hire’s time sitting still waiting for the answer. An hour a day of a skilled person, over a year, is thousands of euros of work; multiply by how many “experts” get interrupted and how many hires go through. Spoken, repeated onboarding is a tax you pay on every person who walks in and every procedure that changes, and you don’t see it because it’s scattered across a thousand micro-interruptions.

This article is about how you turn that graveyard into something people actually use: a place where whoever walks in finds what they need on their own, where the company’s knowledge stops living only in heads, and where whoever is at the desk stops answering the same questions over and over. This isn’t an “HR” topic: it’s a topic of cost and of your best people’s time, the ones who today spend half the day explaining instead of working.

What a new hire should find on day 2

The test of internal training that works is brutal and concrete: what can a new hire do on their second day, on their own, without asking anyone? Day 1 you walk them through it, that’s normal. But day 2 is the moment of truth: if they have to turn and ask a colleague for every little thing, your training doesn’t exist; if they find what they need on their own, it works.

What they should be able to find, on day 2, without bothering anyone:

• How to do the thing they have to do today. Not “all the company’s training”: the specific procedure for the task they’ve been assigned right now. “I need to register an order — how do I do it?” must have an answer findable in thirty seconds, not in a phone call.
• Where things are. The tools, the logins, the places where information lives. The basic map so they’re not lost.
• The rules they can’t guess. The things that in your company are done in a precise, non-obvious way — the ones they’d get wrong if you didn’t tell them. Those are the most precious to put in writing, because they’re exactly the ones that today get passed on only by talking.
• Who to ask for what, when you really do need a person. Not everything goes in a document; but knowing who is the reference for what avoids the random round of questions.

The point is the task, not the “course”. A new hire on day 2 doesn’t want to follow an eight-hour learning path: they want to do the thing in front of them and not get it wrong. Useful internal training is organised around what people have to do, not around how the company is organised on the org chart. It has to answer “how do I do X?” when the person asks themselves the question, at the moment they ask it — not offer a catalogue of courses nobody has time to sit through.

This changes everything in the design: you don’t start from “which courses do we run”, you start from “what are the 20 things every new hire always asks” and from “which are the tasks that get messed up most often”. Those are the first things to make findable. The rest comes later.

Videos, PDFs, quizzes: the minimum you need (in the right order)

You don’t need a pharaonic platform to start. You need the right minimum, and above all used the right way. There are three basic ingredients, and each does a different job.

• Videos. They’re irreplaceable for showing how you do something that’s better seen than read: a walkthrough of a tool, an on-screen procedure, a practical gesture. A three-minute video “here’s how you register an order” is worth ten pages of text and you record it in an afternoon. The golden rule: short and specific, one per task, not a one-hour mega-course nobody finishes. A long video is a video that doesn’t get watched.
• PDFs and text. They’re irreplaceable for what you look up on the fly: a step-by-step procedure, a checklist, the rules to follow, the reference data. Text you can search, scan, find the point — which you can’t do in a video. That’s for “what’s exactly step 4?”.
• Quizzes. Not to run exams, but to check that the message landed on the things where getting it wrong costs. A couple of questions after a video on a critical procedure tell you whether the person understood, and force them to stop and think instead of scrolling passively. They matter most where a mistake has consequences (safety, money, quality).

The minimum you need is this: short videos to show, text to look up, quizzes where a mistake costs. But — and this is the point everyone gets wrong — it isn’t enough to have these materials. The Drive-graveyard is full of videos and PDFs: the problem isn’t that they’re missing, it’s that nobody finds them. Having the material is the easy 20%; making it findable at the moment of need is the 80% that decides whether training lives or dies. And findability is exactly the point of the next two sections.

Who updates when a procedure changes

Here’s the question that kills most internal training, and that nobody asks themselves when they start full of enthusiasm: when a procedure changes, who updates the material? Because procedures change — always, continuously. A tool changes, a rule changes, the way of doing a thing changes. And if the video and the PDF stay the old ones, the worst disaster happens: the training isn’t just useless, it’s actively harmful, because it teaches people to do things the wrong way.

A new hire who follows a documented, obsolete procedure does damage in good faith, and that’s the system’s fault, not theirs. At that point people learn not to trust the material (“it’s old anyway”) and go back to asking out loud — and you’re back at square one, with a graveyard on top. Material nobody trusts is worse than no material, because it costs effort to keep and it returns nothing.

The solution isn’t technological, it’s organisational, and it has to be decided before you fill any platform:

• Every piece has an owner. Not “the company updates the training”: that procedure has that person who keeps it current. Without a name, nobody updates it.
• The update is hooked to the change. When you decide to change how something is done, updating the material is part of the change, not a “we’ll do it later” that never arrives. Whoever changes the procedure updates the piece, immediately.
• You can see what’s old. The system has to show when a piece of material was last updated, so you can tell at a glance what’s fresh and what needs a look. A piece “stuck for two years” is a warning bell.

The structural point is that internal training isn’t a project, it’s a process: you don’t “do” it once and it stays, you keep it alive. And that’s why it has to be built easy to update from the start: if changing a video or a text is a technical nightmare, it won’t get done, and the graveyard reforms. Ease of updating matters more than initial richness — better little material that’s always fresh than a lot of material that rots.

AI: “where’s the piece” with the citation, not an invented course

This is where AI really helps, and it’s the thing that turns internal training from a graveyard into a living tool — but only if used the right way, which is the opposite of how they’ll sell it to you.

How they’ll sell it to you (and why you should be wary): “the AI that generates training courses”, “the AI that creates the content for you”, “the internal academy written by artificial intelligence”. The problem is huge and subtle: a language model that invents a course produces plausible content that is not necessarily true for your company. It teaches the “average” procedure of the sector, not your specific procedure — and your specific procedure is exactly the value, the part the new hire can’t guess. Generated training that looks right but isn’t is worse than the graveyard: it’s the obsolete material above, multiplied by AI speed.

How it should actually be used, and it’s powerful: AI as a search engine over your real material. The person asks a question in natural language — “how do you register a return?”, “what’s the procedure for a foreign client?” — and the AI finds the right piece in your existing material and answers citing the source: “watch video X at minute 4” or “from the Returns Procedure document, point 3”. The whole difference is here:

• It doesn’t invent: it cites. The answer points to a real, verifiable piece of yours, not to an explanation generated from nothing. The person can go to the source and check.
• If it isn’t there, it says so. If the material on that thing doesn’t exist, the AI has to say “I can’t find a documented procedure on this, ask [the reference]” — not invent one. An honest “I don’t know, ask so-and-so” is worth more than an invented answer that’s sure of itself.
• It finds what was there but couldn’t be found. The immediate value: all that graveyard material that existed but nobody fished out becomes finally reachable with a question instead of a treasure hunt through folders.

This is exactly the same engine I describe in finding information in company PDF documents: the AI that answers with a citation of the source, not with invention. It’s the non-negotiable boundary: AI lets you find the real piece, it doesn’t write you a fake course. Whoever sells you the second thing hasn’t understood that in internal training the precious thing is your specific truth, and AI can’t invent that — it can only make it findable.

Permissions by role: not everyone sees everything

As internal training grows, a practical issue shows up: not everything is for everyone. There are procedures that only concern one department, sensitive information (financial data, contracts, confidential stuff), material for managers that doesn’t need — or shouldn’t go — to everyone. Serious internal training has permissions by role.

How it works, in practice:

• Each person sees what they need. The new warehouse hire sees warehouse procedures, they aren’t drowned in everything else. Less noise, easier to find the right thing: permissions aren’t only there to protect, they’re there to not scatter people into material that doesn’t concern them.
• Confidential stays confidential. The things that have to stay in a circle — sensitive data, delicate procedures — are visible only to whoever should see them. Internal training often contains information you don’t want within anyone’s reach, and permissions are what let you put the important things in there without fear.
• Managers have their own part. Management material, supervision procedures, things for whoever coordinates: visible at the right level.

Permissions by role aren’t a big-company complication: they’re what lets you put into internal training even the things that matter, not only the generic ones, because you know each person only sees theirs. Without permissions, either you put everything within everyone’s reach (and then the sensitive things stay out, and training stays shallow) or you put nothing. With permissions, internal training can become the real place where the company’s knowledge lives, confidential included. It’s the same principle of views by role that comes back in every serious company system — knowledge is an asset, and like every asset it has to be given to the right people.

The knowledge that walks out with people

There’s a cost of spoken onboarding we’ve already talked about — the time lost re-explaining. But there’s a bigger, more insidious one, that you only see when it’s too late: the company knowledge that walks out with whoever leaves.

Think of the person who “knows how everything works”: they know the unwritten procedures, they know why certain things are done that way, they have in their head the thousand details that keep the department running. As long as they’re there, it’s wonderfully convenient — just ask them. The day they retire, change jobs, or go on long sick leave, their knowledge walks out the door with them, and you suddenly realise how much you depended on it. Nobody knows how to do certain things any more, or they do them worse, and you spend months painfully reconstructing what was already there — in one person’s head.

This is the hidden risk of the company where knowledge lives only in heads: you’re fragile in a way you don’t see. Every key person is a breaking point. And the paradox is that the better and more irreplaceable that person is, the more exposed you are — because the more things they know that nobody else knows and that aren’t written down anywhere.

Internal training done well is, before it’s even an onboarding tool, a way of pulling knowledge out of heads and putting it where it stays: a video where the key person shows how they do a thing, a procedure they write before they leave, the non-obvious rules put in black and white while the person who knows them is still there. Not to “control” people: so you’re not hostage to one person’s absence. It’s insurance against a risk that, when it shows up, costs far more than any platform — and that today, if all your knowledge is spoken, you have completely uncovered. It’s worth starting to capture it before the door closes, not after.

The metric that matters: repeated questions at the desk

How do you know if your internal training works? Not from the number of videos uploaded, not from the hours of “courses” delivered — those are vanity metrics, they measure effort, not result. The real metric is one, and it’s wonderfully concrete: how many times people still ask the same things out loud.

The cost we want to cut is repeated onboarding, the continuous interruptions, whoever at the desk always answers the same questions. So success is measured there: do the repeated questions go down? If after putting internal training in place people ask out loud less often, if new hires become autonomous faster, if whoever is at the desk gets interrupted less for the usual things — then it works. If they keep asking like before, you’ve built another graveyard, however pretty.

How you measure it, in practice, without going mad:

• Look at what people search in the system. The questions they ask the AI, the terms they search, are a goldmine: they tell you what people need to know. If they search a lot for something and don’t find it, you’ve discovered a hole to fill — with precision, not by guessing.
• Ask whoever sits at the desk which questions they still always get. Those are the holes in the training: the things that should be findable and aren’t. Every repeated question is a missing or unfindable piece of training.
• Look at how long a new hire takes to become autonomous. If the time shortens, the training is working.

This metric closes the loop with the starting cost: we said spoken, repeated onboarding is an invisible tax. Measuring repeated questions makes it visible, and tells you whether you’re lowering it. It’s the same spirit I bring to all the work on document flows in the documents and workflows cluster: what counts isn’t the archives you accumulate, it’s that at the moment of need the information is found — and the proof it’s found is that people stop having to ask for it.

When Notion is enough (and when it isn’t)

Now the honest part, the one people who sell platforms don’t tell you: often you don’t need to build anything custom. For a great many companies, a ready-made tool — Notion, a wiki, a well-organised document space, or a small LMS on subscription — is the right answer. I’m telling you this as someone who builds custom: starting by buying is almost always wise.

The ready-made tool is enough when:

• Your need is organising and making findable material you already have, and your procedures are relatively standard.
• You’re willing to keep the order yourself, with a clear structure and the discipline of updating (which, remember, is the real secret, and no platform gives it to you).
• You don’t need very fine-grained permissions or integrations with your systems.

A well-kept ready-made tool beats a badly-kept custom platform, every day. The graveyard isn’t created by the lack of a powerful tool: it’s created by the lack of discipline in updating and of an organisation by task. Those you don’t buy, you have to put in yourself — and if you put them in, Notion or a wiki is more than enough.

You need something more built when:

• You need the AI search engine with citation on your material, integrated with your real documents (videos, PDFs, procedures), because the volume is such that “searching through folders” no longer scales.
• You need fine-grained permissions by role and the training contains confidential material that a generic tool doesn’t protect the way you want.
• You want to integrate it with your systems — that onboarding starts on its own when a hire comes in, that procedures are hooked to the real tools people use.
• The volume and turnover are high enough that every week of faster onboarding is worth more than the cost of building.

The rule is the usual one, honest: ready-made covers 80% of cases, custom is for the 20% where ready-made doesn’t reach — here, typically, AI search on your material and fine-grained permissions. If you’re in the 80%, take Notion, organise it by task, put in the discipline of updating, and you’re set. If your problem is volume and findability on a mountain of confidential material, then it’s worth building. Don’t let anyone sell you a pharaonic platform if a well-kept wiki is enough; but don’t kid yourself that a messy wiki solves a real volume problem.

If you have a “training” graveyard and colleagues who always answer the same questions, the first step is understanding which of the two cases you’re in. See how I work or write to me and we’ll reason it through — even if only to tell you “for you a well-kept Notion is enough”, if that’s the case.

It’s for you if / it’s not for you if

It’s for you if:

• you have a “training” folder that’s a graveyard and new hires always ask the same things;
• your best people spend time re-explaining out loud things that should be findable;
• onboarding is slow because knowledge lives only in a few heads;
• you have volume of material and of turnover such that searching through folders no longer scales;
• you need training to contain confidential things as well, with permissions by role.

It’s not for you if:

• you’re a very small team and knowledge transmits well in person: here a platform is more weight than help, a wiki is more than enough;
• you’re not willing to put in the discipline of updating: without that, any tool becomes a graveyard, and building custom would be money thrown away;
• your material is little and standard: Notion or a well-kept wiki is the right answer, and I’m telling you that honestly.

8 questions from someone about to sign

1. Don’t I already have everything on Drive? Why isn’t that enough? Because the problem isn’t having the material, it’s finding it at the moment of need. Drive is a graveyard not because files are missing, but because nobody fishes out the right piece when it’s needed, and so they go back to asking out loud. Having the material is the easy part; making it findable is what counts.

2. Will AI write the training courses for me? No, and be wary of whoever proposes that: an AI that invents courses produces the average procedure of the sector, not yours, and teaches things that aren’t true for your company. AI should be used to find the right piece in your real material and answer citing the source, not to generate content from nothing.

3. What happens when a procedure changes? The material has to be updated, immediately, as part of the change — and every piece must have an owner with a name. It’s the point that kills internal training: obsolete material is worse than no material, because it teaches people to get it wrong and it burns trust. Ease of updating matters more than initial richness.

4. How do I measure if it works? From repeated questions at the desk: if people ask out loud less often, if new hires become autonomous faster, it works. Also look at what they search in the system and don’t find: those are the holes to fill. Hours of courses delivered and videos uploaded are vanity metrics.

5. Do I need a platform or is Notion enough? Often a well-kept Notion (or a wiki) is enough: if procedures are standard and the volume is manageable, ready-made is the right choice. You need something more built when you need AI search with citation on a mountain of material, fine-grained permissions by role, or integrations with your systems.

6. Videos, PDFs or quizzes: what do I need? All three, for different jobs: short videos to show how it’s done, text to look up the exact step on the fly, quizzes to check that the message landed where a mistake costs. That’s the minimum, with the rule of short, specific videos, one per task.

7. Can I keep confidential things separate? Yes, with permissions by role: each person sees what they need, confidential stays with whoever should see it. It isn’t a big-company complication: it’s what lets you put into training even the things that matter, not only the generic ones, without fear.

8. How long does it take to start? Not long, if you start from the 20 things every hire always asks and from the tasks that get messed up most: make those findable first, the rest comes later. Internal training isn’t a project to finish, it’s a process to keep alive — better to start small and always fresh than big and destined to rot.