Onboarding new employees is done wrong. It should be done through forums rather than documentation

Right now, when new people join an organization, they have to be brought up to speed on a lot of institutional knowledge that was built up over a long time. This is done incredibly inefficiently and requires the person to deal with out of date documentation or to continually ask more experienced members about things.

This whole system is designed badly and should be revamped in favor of something that allows the person to quickly learn things while using other members’ time as efficiently as possible.

This is possible by setting up an internal version of a forum like stackoverflow and making it searchable with something like google.

What’s the problem with documentation?

In general? Nothing. But it’s terrible for institutional/tribal knowledge.

Standard documentation that someone is supposed to digest before they start is a classic example of Preemptive Learning. This is where everything a person might need to know about the topic is presented as a monolith that is usually learnt before it is put into practice. This is the sort of learning used in formal education, and it makes sense there. It is very well suited to learning universal systems that are (roughly) constant and present intellectual challenges in understanding them. Something like math, philosophy, or economics. But not tribal knowledge.

The knowledge base of a large organization generally encodes the processes people have to follow to be able to do their jobs. These are specific to the organization and change all the time as new systems and enterprises rise and integrate into the company. They are also rather intellectually trivial, with systems not being very challenging to understand.

Essentially, this type of knowledge is encoded as pairs of simple problems and solutions. And while in real life, the problem always comes first and motivates the solution, large monolithic documentation indexes the knowledge by the solution, so if you already know the answer, then it’s easy to find. But as people don’t know the answer, they have to start reading the whole thing from the beginning. This is very inefficient for answering the simple queries that most entry level institutional knowledge is made of.

This is stuff like “which table in the database has this value?” “do we have any usage limits on our client outreach tools?” “whats the endpoint and authentication to run an SQL query through python?” or “where do we keep the extra coffee? This one's out”. The more “advanced” topics that need “experts” or “tribal elders” to answer are simply complex combinations of these simple atomic trivia.

It is simply not feasible to maintain up to date documentation that covers all these topics. The details will change too frequently and the quantity of minor trivia will be too large. And even if the documentation is made properly, the material will not be challenging enough to need the deeper understanding and more time to ruminate that a monolithic documentation affords.

How can you learn these things without documentation?

On-demand Learning. It’s the other side of the coin to preemptive learning.

While preemptive learning forces the person to digest everything before they start applying it, on-demand learning has the person start performing the process they are learning about and keep doing it until their lack of knowledge leads to a roadblock. This blockage then motivates the person to seek help. They are then guided to the set of information that allows them to proceed.

This approach is ideal for learning processes that change quickly and are not particularly deep in their intellectual rigor. The workflows that are specific to an organization and are mostly encoded in their tribal knowledge are the perfect application of this learning approach.

Not only do I believe this approach is better suited to onboarding a new member of an organization, this is the approach that naturally ends up being followed in large organizations when preemptive learning inevitably fails. New hires are directed to a tediously long documentation, they do not absorb most of it because it is trivial and they don’t even know the problems the solutions they are learning are for, and once they actually start working, they still have to ask people when they get stuck because of something that wasn’t covered in the outdated and hard-to-maintain documentation.

The people in these organizations who have enough knowledge and experience to unblock the students quickly and efficiently are also necessarily the people whose time is most valuable and productive. Unless an organization wants to pay its most qualified people to be lore keepers who pass on this oral tradition, a better and more efficient way is needed.

What’s the alternative?

The problem has already been solved by open source communities online. There are a lot of jokes in the tech world about how all programmers do is google stuff and copy paste code from others, and how no one has seen the homepage of stackoverflow. This is because the most efficient way for programmers to gain trivial knowledge about a certain process is by searching through a catalog of questions that have been previously asked and finding their query on there, or creating a new one if theirs truly is novel.

This forum system is ideal because while most of these things are covered by documentation, it simply isn't the most natural way to get the answer for most people. You don’t really want to crack open the vim user manual just to figure out how to exit the application. This goes back to the query-response structure of these trivial pieces of knowledge. Forums essentially index this knowledge by the problem rather than the solution. This makes it far easier to search for a particular piece of knowledge, when you know what problem it’s supposed to solve.

This system obviously does not work when a person wants to get deeper into a topic and understand things better. But if trivial things like syntax or the name of the function designed for a particular task is what you’re looking for, then it’s perfect. And these are the exact types of things that organizations have to waste a lot of time teaching to members so as to bring them up to speed.

All organizations would benefit from maintaining an internal Q&A forum focused on organizational and procedural queries people have. Something that performs the same function that google+stackoverflow performs for the general public. There could even be an open source implementation using an open source search engine like elasticsearch plus a generic forum webapp.

Instead of having its “tribal elders” repeat the same information for every single new member, they should invest that time in answering questions on the forum that will be there for every new member to search for and find when they need it.

This will be a lot more efficient than the current onboarding paradigm

You’re making a post about knowledge systems in 2023 without mentioning LLMs?

Of course not.

LLMs are very powerful and the most useful function they can perform is as natural language interfaces. So the paradigm I prescribed above might quickly become obsolete in favor of LLMs that answer natural language queries about trivial problems, in context of large monolithic documentations. This is an ideal usecase as it solves the initial problem with using documentation: that you have to read through all of it to find obscure details. LLMs can parse the documentation in such detailed semantic manner that it can essentially index it by a query rather than a solution.

But this still does not solve the other problem with documenting these sorts of knowledge systems; that they are difficult to maintain. The LLM has to pull from something and if that something is not an acceptable representation of the truth then the LLM can not make up for it. Garbage in; garbage out.

The forum aspect of the system might still be necessary for cataloging the information in an easy-to-maintain manner. Maybe only the search aspect is replaceable, with LLMs taking the place of the traditional reverse indexing engine.

TL;DR

• People don’t learn systems through huge monolith documentation
• They learn pieces as and when they are needed
• Examples include stackoverflow and reddit
• Questions shouldn't be answered before they are asked
• New people need a platform to easily search old questions and ask new ones
• This search engine + forum platform could be replaced by LLMs soon