Off-the-shelf vs bespoke AI training: how to choose

AI training that works shows the following Monday. Not in the monthly survey.

It’s the only yardstick that counts, and hardly any training passes it. The team comes out happy, with a folder of examples and three new accounts set up. Two weeks later it’s working exactly as it did before.

The difference between those two things isn’t in the syllabus or in how good the trainer is. It’s in a decision made before you hire anyone: whether the course existed before anyone met your team, or whether it’s built from what your team does every week.

What bespoke AI training for businesses is

AI training for a business is bespoke when the syllabus is built from the team’s real tasks, mapped out before the first session, and each block ends with one of those tasks done a different way. It’s off-the-shelf when the syllabus already existed before anyone met the business: same script, same examples and same exercises for everyone who buys that course.

In the training world the first kind is also called in-house training. The name doesn’t matter. What changes is where the exercises come from.

Both can be well delivered. Both can have a good trainer and a good room. What they leave behind in the business when the trainer goes is nothing alike.

Off-the-shelf training and bespoke training: the difference in practice

The right-hand column isn’t more expensive per hour. It’s more uncomfortable: it forces someone inside to put on the table how things are really done, with their shortcuts and their exceptions. That’s where all the work is, and that’s why hardly anyone sells it that way.

Six months later: two very different stories

On the day of the training, the two look very much alike. Attentive people, impressive examples, a good feeling on the way out. Nobody can tell them apart.

The difference shows half a year later, when someone in management asks what has changed since then. At that point the two answers are nothing alike.

If it was off-the-shelf

Nobody can answer with a concrete example. Two or three people are still using something on their own, each in their own way, without the others knowing. The licences bought for the whole workforce are half used. There isn’t a single written procedure, so what the person who left in March learned left with them. The invoice is paid and there isn’t a single task done differently.

And there’s damage worse than the money: the conclusion people draw. We tried it and AI wasn’t for us. That closes the door for another year. The door wasn’t closed; they just knocked on the wrong one.

If it was bespoke

The answer is boring, which is exactly the sign. There are three or four tasks done a different way and nobody calls them “using AI” any more: they call it working.

There are written procedures, half a page each, that say what to ask the system and what always has to be checked. The person who joined in May read them and got up to speed without anyone having to train them again. And the time those tasks take can be compared with before, because it was measured before.

Something else appears that wasn’t in the plan: a list of tasks that weren’t a knowledge problem, but a process problem. Those aren’t solved by training more people. That’s where the work continues, and it’s usually where the real money is.

Six months isn’t a shelf life. It’s the point at which you can see what you actually bought.

How bespoke training is structured

By stages, not by weeks. The team sets the timetable: a twelve-person office and a business with three branches don’t move at the same pace, and selling both the same fixed package is off-the-shelf training through the back door.

What doesn’t change is the order.

1. First, the real work is mapped out. Nothing is taught. A short conversation with each key person: what do you do every week, how long does it take you, which part do you hate. Out of that comes a list of tasks ordered by time spent. That’s the syllabus, and the trainer didn’t write it.

2. The session is delivered on those tasks. No generic examples. The real email gets opened, the real report, the real tender document. The team doesn’t learn to use a tool: it learns to do its job with the tool in front of it. It’s the same classroom time as an off-the-shelf course, spent on something else.

3. The team applies it alone, with someone to ask. It’s the stage that decides everything and the one almost no programme includes. Everyone works as normal, with a channel open for questions. This is where what doesn’t come up in the classroom comes up: the odd case, the format that doesn’t fit, the question nobody asks in front of their boss.

4. It gets corrected and written down. Second session: what worked, what didn’t and why. And it gets written down. Three or four half-page procedures, with the task, what to ask the system and what always has to be checked. That document is the programme’s real deliverable. Without it, six months later there’s nothing left.

There are few classroom hours, and it isn’t an oversight: it’s the design. A team doesn’t change the way it works by spending more time sitting and listening.

Not everyone needs the same thing

Training the whole workforce on the same content is the fastest way to make sure nobody uses it. There are three audiences and three different programmes:

• Management needs to know how to decide: what can be delegated to a system, what should never be delegated and what to ask a provider before signing. They don’t need to touch anything.
• Middle managers need to see where AI fits into the process they already manage, and above all where it doesn’t. They’re the ones who will sustain the change or bury it.
• The people who do the work need three specific tasks from their day done a different way. Not one more. Three is enough for them to go looking for the rest on their own.

When management hands AI over “to the youngsters on the team”, the programme dies on its own. Not for lack of talent below, but for lack of a decision above about what changes.

What you learn from training people who can’t afford a mistake

We’ve delivered AI training to the European Committee of the Regions, the Spanish Chamber of Commerce in Belgium and Luxembourg and the Delegation of the European Union to Venezuela, as well as to private companies. They’re very different audiences with one thing in common: a mistake of theirs can’t be fixed with an apology email.

Three things those rooms teach that apply equally to a fifteen-person SME.

The first thing to explain is the limits, not the possibilities. Where something real is at stake, the first question is never what AI can do. It’s what information can’t leave the building and who answers if the system gets it wrong. Starting there doesn’t slow adoption down: it speeds it up, because it removes the fear that was blocking it.

The starting level is always more uneven than the person who hires you says. In the same room there’s someone who has been using this every day for two years and someone who has never opened a chat in their life. A single syllabus loses both of them: one to boredom, the other to embarrassment. An off-the-shelf syllabus has no way of knowing, because it was written before anyone saw the room.

A demonstration is worth more than an hour of arguments, and a demonstration with the client’s own data is worth more than ten. It’s the most awkward part to prepare and the only one people remember three months later.

There’s a fourth that came from the private sector. At Grupo Platea the relationship started with training and ended up with four processes automated. Not because anything was sold in the classroom, but because when the team’s tasks were mapped out, some appeared that weren’t a knowledge problem: they were a process problem.

When training isn’t the answer

There are three cases in which training is throwing money away. Saying so early works out cheaper than charging for it.

When the task happens two hundred times a month. That doesn’t get trained, it gets automated. Teaching someone to do faster what a system should be doing is paying twice for the same problem.

When the process is broken. Putting AI on top of a process that no longer works multiplies the mess, and faster. First you need to look at which business processes to automate and which aren’t ready.

When nobody is going to change anything afterwards. If there’s no one with the authority to say “from Monday, this is how it’s done”, the programme is a team-building activity, not an investment.

If cases of the first two kinds appear when the team’s tasks are mapped out, that isn’t the programme failing: it’s having found where the money is. It’s exactly what we look at when someone describes a process to us in the free AI assessment, to see whether there’s a case and what kind it is.

What to measure to know whether it worked

Four indicators. They’re checked a month after finishing, and checked again six months later, which is when they tell the truth.

• The time a specific task takes, before and after. Pick two and measure them properly. Without a starting figure there’s no way to defend the investment to anyone.
• How many written procedures exist that didn’t before. It’s the asset that stays in the business the day the trained person leaves.
• How many people use the tools without anyone reminding them. Use that needs a reminder isn’t adoption.
• How many tasks have changed owner, from a person to a system. It’s the indicator that connects training with everything else.

The one that doesn’t work: the satisfaction survey on the last day. It measures whether people liked the trainer.

Frequently asked questions

What’s the difference between off-the-shelf and bespoke AI training?

The syllabus. In an off-the-shelf course it exists before anyone meets your team: same script and same exercises for everyone who buys that course. In a bespoke one, what the industry also calls in-house training, the syllabus comes from the tasks your team does every week, mapped out before the first session. The practical consequence shows in what’s written down at the end: slides in one case, procedures for your own tasks in the other.

How long does my team need to be away from their work?

Less than anyone who has sat through a twenty-hour course fears. The useful classroom hours are few and split across two sessions. The part that really changes things happens while the team works as normal, applying it to their own work with someone to ask. A programme that needs to take your people away from their work for days on end is usually an off-the-shelf syllabus looking for somewhere to fit.

How much does an AI course for businesses cost?

It depends on how many people there are, how many different types of work there are within the team and whether the tasks have to be mapped out from scratch. What’s worth knowing: anyone who gives you a fixed price without having seen a single one of your team’s tasks is selling you off-the-shelf training, and it’s cheap for that reason. Ask for the quote to state what your team takes away in writing on the last day, and compare that.

How many people should attend?

Groups of six to twelve people, grouped by type of work rather than by department. Two people who do similar tasks in different areas get more out of it together than a whole department with five different roles inside it.

Is there any point in training only management?

It helps them decide better, and it doesn’t change any task. It’s a good first step when the business doesn’t yet know what it wants to do. If it stops there, the effect wears off by the next meeting.

How can I tell, before signing, whether a programme will actually be applied?

Ask one thing only: what does my team take away in writing on the last day. If the answer is slides and the recording, it’s a talk. If the answer is procedures for their own tasks, it’s bespoke training. And if they show you the finished syllabus before they’ve spoken to your team, you already know which of the two you’re being sold.

What to ask before you sign

AI theory has become cheap. It’s in any video, free and well explained. Nobody should pay for it.

What still takes work is translating it into the specific tasks of a specific team and writing it down, so it survives the enthusiasm of the first month. That doesn’t fit in a catalogue, because a catalogue doesn’t know how your people work. It’s how we design our AI training for businesses: from your team’s tasks, never from a ready-made syllabus.

If you’re looking at AI training for your business, or you’re not sure whether what you need is training or something else, tell us what your team does every week. We’ll tell you which part can be solved in a classroom and which part can’t.