The Training Trap

Right now, a meaningful portion of enterprise AI training spend is being wasted. Not because the training is bad. Because it is solving the wrong problem.

Here is the number that should be making boardrooms uncomfortable. In a May 2025 survey of 506 CIOs and technology leaders, Gartner found that 72% reported their organisations are breaking even or losing money on their AI investments. Budgets are climbing. Returns are not. The gap between spend and result is not a technology problem. It is a sequencing problem, and training is at the centre of it.

The logic most leaders are working from goes roughly like this: we have an AI skills gap, so we need to train people, so we run workshops and roll out learning platforms, and then the gap closes. It is a tidy sequence. It is also largely wrong.

Training without redesigned work is the organisational equivalent of teaching someone to drive on a road that does not exist yet. The skills land nowhere. People return from the session, sit back down at the same process with the same inputs and the same accountabilities, and the training fades. Three months later, leadership wonders why adoption is flat.

Deloitte's 2026 enterprise survey is pointed about this. Education, not role or workflow redesign, was the number one way organisations adjusted their talent strategies in response to AI. Those ordering matters. It reveals what most leaders believe: that if people understand AI better, value will follow. The evidence says otherwise.

The organisations pulling genuine return from their AI investment are not starting with training. They are starting with the work. They identify a process that is slow, expensive, inconsistently executed, or chronically overloaded with manual effort. They redesign that process around what AI can do. Then they train the people who own that process in exactly what they need to know to operate it well. McKinsey's 2025 State of AI research, drawing on nearly 2,000 respondents across 105 countries, found that of all the organisational factors tested, fundamental workflow redesign has one of the strongest contributions to achieving meaningful business impact from AI. The difference is not enthusiasm or investment size. It is sequence.

There is also a subtler problem embedded in how most training is designed. The dominant format across most organisations is a combination of online modules and periodic workshops, often generic, rarely role-specific, and almost never connected to a workflow that has changed. That approach produces awareness. It does not produce capability. Awareness means someone can describe what a large language model does. Capability means they can identify where it applies in their specific role, use it reliably, recognise when it is wrong, and explain to their team why the new process works differently to the old one. Those are not the same outcome, and the gap between them does not close with a 90-minute webinar.

The Davos framing by People Management on this is worth stating plainly: the biggest challenge organisations face is not a skills gap or an AI gap. It is a work design gap. That reframe matters commercially. Because if the problem is work design, then the solution is not more training spend. It is a different decision about where to deploy the training you already have.

McKinsey found that AI high performers are nearly three times more likely than others to say their organisations have fundamentally redesigned individual workflows. Yet they represent only around 6% of all organisations surveyed. The rest are layering AI onto what already exists. That is the training trap in operational terms. You are building capability for a version of work that is not changing fast enough to use it.

Before you approve next quarter's AI training budget, run this test

Start with the work, not the curriculum. Pick three workflows you expect AI to improve. Write down, in plain terms, how each of those workflows currently operates and how it should operate once AI is embedded. If you cannot articulate that clearly, the training will have nowhere to land.

Ask who owns the outcome. Training a team is not the same as assigning accountability for the result. Someone needs to own the before-and-after on each workflow you are targeting. If that accountability is unclear, the training will produce interest, not change.

Audit your current training format. If your primary delivery is online modules and one-off workshops with no role-specific application built in, you are producing awareness, not capability. Ask your provider what percentage of the programme involves applied practice in real work contexts. If they cannot answer that, find a different provider.

Measure capability, not completion. Certificates issued and hours logged are not the same as capability built. Define what the person should be able to do differently after the training, then test for that specifically, not for whether they finished the module.

Close the loop with the business. Three months after any training cohort, someone should be reviewing whether the targeted workflows have actually changed, and if not, why not. That review should sit with a business leader, not in the L&D function alone.

The point is not to stop investing in AI capability. The point is that training spend only compounds when it is connected to work that is already being redesigned. Separated from that, it becomes a line item that looks responsible on paper and underperforms in practice.

Most organisations are currently doing the responsible-looking thing. The ones pulling ahead are doing the thing that works.

𝛀

If you enjoyed this Independent A.I. Think Tank release, consider subscribing to our D!sruptive newsletter. Come back for more in our next edition.

If you need the help discussed in this article, get in touch to see how we can work with your business 🇺🇸 +1 512 518 4196, 🇦🇪+971 585 629 975 or hello@byDisrupt.com