It feels like a lot of the early optimism around AI has quietened down. Not every sector, but definitely in training and education.
On paper, training courses should be a great use case. There’s repetition and a huge amount of informational support required to help prospects and students move through the system.
But what I suspect a lot of providers discovered quickly was that AI struggled to hold the nuance. It handled surface-level questions well, but lost its grip the moment prerequisites, progression, insurance, accreditation or scope of practice entered the picture. And when it didn’t know the answer, it didn’t pause or ask for clarity — it made things up.
My impression is that most providers are waiting — not dismissing AI, just waiting for the conditions to line up in a way that would make the next step feel worth taking.
Is that a fair read, or am I imagining it?
The tacit knowledge problem in training businesses
If you look closely at how training providers work, most of the operational knowledge sits in people’s heads rather than systems. It’s tacit. It moves through conversations, outdated documents, tutor judgement, and whoever happens to be around to answer a question at the time.
There’s a huge number of “it depends” scenarios. Eligibility, prerequisites, scope and insurance usually sit with tutors or with team members who’ve been around the longest. Even pricing has caveats that only make sense when someone talks you through them.
Sales guidance tends not to be documented either. Unless the business has scaled to having its own sales function, responses vary dramatically depending on who’s in the inbox that day.
Brand voice and tone follow a similar pattern. One person excels at responding to enquiries, another struggles, and another is technically correct but something about their responses feels off. It’s hard to pinpoint why because nothing has been formalised.
It’s no wonder new hires take so long to settle in and end up asking the same questions repeatedly.
My take is that the lack of meaningful AI adoption is less about the technology and more about this environment. Yes, AI tends to go straight for the fastest possible answer, meaning it struggles with nuance, and asking it to slow down or ask questions feels counter to what people expect. But the deeper challenge is that training providers tend not to have the structure that would allow AI to succeed in the first place.
Doing the unglamorous work first
If training providers want the upside of AI, they have to fix this. AI doesn’t magically tidy things up — it amplifies whatever structure already exists. If the structure is weak, the output becomes chaotic and unpredictable.
Most providers simply didn’t anticipate having to do this piece of work. Turning tacit knowledge into structure takes time and ownership, and most training businesses are already at capacity. Under those conditions, keeping the current system running feels easier than reengineering it — even if it’s not ideal.
Humans can improvise without structure. AI can’t. The bridge between the two is a knowledge base.
Documenting the business
A knowledge base acts as the operational memory of the business — the rules, caveats, pathways, tone and decisions that make things work, written down and structured in a way that AI can reliably follow.
If you’re familiar with SOPs, you might think of it as something similar; only instead of prescribing steps, it captures decisions, conditions and judgement calls.
In practice, most training providers would have the following files as a starting point:
Course FAQs & content — logistics, learning outcomes, assessment, structure and what to expect.
Eligibility & prerequisites — who qualifies, under what conditions, and when additional questions are needed.
Scope, insurance & compliance — what the qualification covers, what the boundaries are, and how awarding bodies interpret those boundaries.
Pricing, bundles & discounts — how pricing works, where exceptions are made, and what payment plans are available.
Sales guidance & positioning — how the course is framed, its main selling points, and how hesitations are handled.
Tone & voice — how the business sounds in email, WhatsApp and DMs, including punctuation and emojis.
When you look at those files together, you start to see how they map to the people you rely on most. The sales person who knows how to position things just right. The tutor who understands scope or prerequisites. The team member who writes emails that sound exactly like the business.
A knowledge base is where all of that lives.
Internal capability before external capability
With these files in place, the benefits start to show up even before involving AI. Simply having this knowledge in one place allows for:
Faster onboarding — new hires aren’t learning the business through a slow drip of questions.
Fewer bottlenecks — knowledge stops being held in one or two people, and senior staff aren’t constantly pulled into small decisions.
Fewer missed opportunities — learners are guided to the right course, bundle or pathway more reliably.
Better tone cohesion — WhatsApp, email and DMs sound like the same business.
This is based on nothing more than team members being able to access those files on the network.
Things step up again when the knowledge becomes queryable and can answer questions — through an internal model built on that knowledge (for example, a CustomGPT to start with).
“Does this lead meet the entry requirements?”
“Do we need more information before confirming eligibility?”
“If they’re unsure between two options, how should we guide them?”
“How do we position value without feeling salesy?”
“Should we recommend a bundle or a single course?”
“How do we help someone progress from interest to decision?”
“When should we check in again — and how?”
“How should we explain the course to someone who’s unsure?”
Once a model can answer these kinds of questions, the business changes. Senior staff stop acting as the bottleneck for every decision, and junior staff get the clarity they need to do their best work.
No one is being replaced. What’s being replaced is the need to constantly extract knowledge from the same small group of people.
Extending outward to prospects and students
Once the model can reliably answer team questions, extending it outward feels a lot less risky. Early versions support the team. Later versions can move closer to the prospect or student — whether through live chat, drafted replies or personalised follow-up.
It takes time and effort to get there, but what you end up with is a training business that’s ready for where AI is heading. Without that groundwork, there isn’t much for AI to build on.
Training can be served well by AI. I’ve done it. The foundations just need to be there, and the team needs to be ready to adapt.
Closing Thoughts
Training businesses run on nuance and “it depends”. There’s no getting away from that, but it’s not beyond the reach of AI. It just needs the knowledge to be structured, rather than tacit.
Most of that work isn’t glamorous. It’s building the operational memory of the business — the judgement calls, the pathways and the tone that hold everything together behind the scenes. With that in place, AI adoption can stop being speculative and start being practical.
Sooner or later, that framework has to be built. There isn’t going to be a model update that magically tidies things up. The foundations belong to the business, not the tool.
