Public vs in-house AI training: how to choose

On-Off Group

September 2026

Send two people to an open course when you are still testing whether AI helps your work. Run it in-house when a whole team shares one process and you want them to change it together. The deciding factor is whether the work is individual or shared.

Two seats on an open AI course, or a trainer in your own meeting room with the whole department. Here is how to pick, and what to write down before you spend anything.

If you are the one who has to book the AI training and defend the spend, this is for you. When each one works, and what it costs you if you get it the wrong way round.

Send individuals when you are still finding out whether AI helps your work at all. Run it in-house when several people share one process and that process has to change.

Most AI training starts without one real task named

A department gets AI licences, or a day of training, and a few weeks later nobody can say what changed. Attendance was good, feedback scores were good, and Monday looked the same as before.

That happens in both formats. It happens because the session covered tools in general and nobody decided, out loud, which piece of real work the tools would be used on first.

So before you compare an open course with an in-house workshop, write down one task. Drafting the branch memo. Summarising complaint emails. The first check on onboarding documents. Something a named person does every week and can time.

We wrote about the licence version of this problem here: Instead of buying AI licences for everyone, train one real task first. The same logic decides your training format. If you cannot name the task, an open course for two people is the cheaper way to find one.

Send two people when you are still testing

An open course is the right call when you are still working out where AI helps. You think AI could help somewhere in operations, but you do not know where yet, and you cannot ask for training for the whole department on a hunch.

Pick two people who already want it. Not the two with free diaries. Someone who has already been messing about with prompts on their own will get more out of it than anyone who was told to attend.

The other reason to use an open course is who else is sitting there. An open course has people from other banks and insurers, and a few founders. You hear how someone else's operations team got past the thing you are stuck on. That does not happen when everyone in the room reports to the same person.

Ask the two to come back with one specific thing: a prompt that worked on real work, and how long the task took before and after. That is what you show your boss when you ask for the in-house day. A certificate tells you nothing.

If you want to see what is typically covered at this level, What is corporate AI training and what does it cover lays it out.

Run it in-house when the work is shared

Run it in-house when the whole team has to agree on the new way of working. A claims team on one queue. Branch staff following the same onboarding steps. An underwriting team working from one set of guidelines.

In those cases, one trained person cannot change much. They will write better prompts and then hit the fact that the process, the templates and the checks around them have not moved. Sending two people from a team of twenty tends to produce two frustrated people.

In-house also lets you use your own material. Real customer emails, your policy wording, the forms branch staff fill in, the queue as it stands this month. That is what makes a difference on Monday. Practising on made-up examples does not.

The cost of getting this wrong is a day of everyone's time. So be blunt about who is in the room. If half the group is there because their manager volunteered them, expect half the group to ignore it.

Write down the before number either way

This is the part most teams skip, and it is the reason the question "did it work?" gets answered with a shrug.

Pick a number that already exists in your reporting. Turnaround time on an onboarding case. Days to close a complaint. Hours to produce the weekly pack. Write it down before anyone is trained. Check it again a few weeks after.

It does not have to be clever. It has to exist before the training, because you cannot go back and work it out afterwards.

Here is what a before number makes possible. On a series of five-day design sprints with a large insurer, customer-facing web journeys that used to take six months or more were designed, tested with customers and live in four weeks. The share of customers who finished the new journeys rose 80 percent, and the old drop-off points disappeared. Those numbers only mean anything because the client had measured the old journeys first.

Same goes for training. Without a before number, you have had a good day out and nothing to show for it when you ask for the next one.

The sequence that usually works

Most teams do not have to pick one. They can do them in order.

Two seats on an open course. Those two come back with one task, one prompt that worked, and a rough time saving. If the saving holds up, you have the argument for bringing it in-house, and you also have two people inside the team who can help run the session and answer questions afterwards.

If the saving does not hold up, you have spent two seats finding that out instead of a department day.

There is a third option: no training at all. Sometimes what is missing is a decision. If four managers disagree on what AI is for, get those managers in a room for a working session instead. Custom workshop or strategy session: which one you need covers that distinction.

What to do this week

Name one task. A real, weekly piece of work that a specific person does, not a category like "reporting".
Find the before number for that task and write it down somewhere your boss can see it.
Count how many people share the process. Two or three, send individuals. A whole queue or a whole branch process, run it in-house.
Pick your two people by interest, not availability.
Book the check-in now, a few weeks out, and put the before number in the invite so nobody has to go hunting for it.

Questions people ask

How many people make in-house training worth it?
When enough of the team share the same process that they would have to agree on a new way of doing it, in-house makes sense. Below that, two seats on an open course gets you the same knowledge without booking a room.

What should we measure before the training?
One number from the work itself, such as turnaround time on an onboarding case or how long a report takes to draft. Write it down before anyone is trained, then check it again a few weeks later.

Why does in-house training often fail to change anything?
Because people are sent rather than chosen, and nobody names the task they will use AI on. If the first real task after the session is not decided in the room, Monday looks the same as it did before.

On-Off Group trains teams, tests products with real customers, finds where a transformation has stalled and builds what gets it moving, for banks, insurers and enterprises in the Philippines, since 2015. How we help with ai training.