Every year the Association for Talent Development publishes a report whose data pages every learning director turns to first: average learning hours and training spend per head. The 2026 edition, covering 2025, has two numbers moving in opposite directions1.
Average formal learning hours per employee rose to 16.7, after five consecutive years of decline. In 2024 the figure had fallen to 13.7 hours, a five-year low1. Over the same period, direct learning spend per employee fell from 1,254 dollars to 846, a drop of about a third1.
ATD gives three reasons for the gap. Virtual instructor-led classes replaced classroom sessions and cut travel costs. Learning teams got smaller, so salary costs fell. And more administrative work was automated1.
Read closely, the second reason deserves a pause. Part of the saving came from technology. Another part came from cutting the very people who run training.
Walmart and 17,000 headsets
To picture an organisation raising learning hours while lowering spend per head, Walmart's 2018 story is still the most visible example.
Until then, Walmart's most structured training took place at its Walmart Academies and was largely reserved for managers and department managers. Associates on the tills, in the stockroom and on the shop floor, more than a million people, had little access2.
In summer 2018 Walmart ran a pilot in ten stores: associates learned to operate a new piece of equipment, the automated pickup tower, entirely through virtual reality headsets, with no instructor beside them. The results were good enough to scale2.
In September 2018 Walmart announced it would put more than 17,000 Oculus Go headsets into its US stores by the end of the year: four for every supercenter and two for every Neighborhood Market and discount store2,3. More than 45 modules developed by STRIVR covered new technology, compliance, and soft skills such as empathy and customer service. Walmart reported post-training test scores rising by 10 to 15 percent2.
Seen through the ATD report, this is exactly the mechanism that raises hours while lowering spend per person: content built once, delivered to millions, with no travel and no instructor in every store.
What neither story answers yet
Walmart measured test scores. That is level 2 in the Kirkpatrick model: did learners acquire the knowledge? The next question, whether associates actually behave differently at the till, in the stockroom or in front of an irritated customer, is much harder to measure and rarely published.
The same goes for the ATD report. Learning hours are an input measure. They tell you how much time people spent, not what that time bought. When spend per person falls by a third and learning teams thin out, the biggest risk is that organisations keep reporting rising hours as an achievement while nobody has the time left to check whether learning is turning into behaviour.
For learning teams in Vietnam
ATD's two numbers are a ready-made argument for anyone defending next year's budget: spending less does not mean training was cut. But they are also a reminder that savings from technology are different from savings from cutting people, and the two lead to very different places a year later.
A small calculation from ATD's two numbers
Put the two figures together and you get a measure the report does not print: direct cost per learning hour. In 2024, 1,254 dollars divided by 13.7 hours is about 91.50 dollars an hour. In 2025, 846 dollars divided by 16.7 hours is about 50.70. Cost per learning hour fell by nearly 45 percent in a single year, on ATD's own figures1.
Few functions cut their unit cost by almost half in twelve months. That says something about the power of new formats, and about the pressure learning teams are under.
Any company can work this out for itself: total direct training spend for the year divided by total employee learning hours. Do it for two consecutive years and ask where the difference came from. If cost per hour fell because learning moved online, that is efficiency. If it fell because the hardest and most expensive programmes were dropped, such as middle-manager development or one-to-one coaching, that is a decision leadership should hear about plainly.
