AI Upskilling Guide: Train the Skills That Actually Pay Off

55% of workers now use AI daily or weekly. Only 33% have received any employer-provided AI training in the past six months. That gap alone should worry anyone who owns learning and development, but new research from The Conference Board points to a bigger problem hiding underneath it: the AI upskilling that does exist is teaching the wrong things. This guide walks through what the latest data says companies get wrong, which skills employers say actually separate strong performers, and how to build a program that pays for itself, with one study showing a 258% return on exactly the kind of training most companies skip.

Why AI upskilling lags so far behind actual AI use

AI upskilling is losing the race because employees adopted the tools faster than employers built the training. In a global survey of nearly 1,300 workers released July 28, The Conference Board found that 55% of workers use generative AI or AI agents daily or weekly, while just one in three took part in any employer-provided AI training in the past six months. 28% said their employer offers no AI training at all.

The support around training is just as thin. Fewer than half of workers (48%) say they get sufficient time during work hours to develop AI skills, and only 47.6% say they have the tools and resources to do it.

So most AI skill-building today is self-taught, improvised, and invisible to the organization. People are learning on live work, without feedback, and whatever habits they pick up (good or bad) become the company's de facto AI capability.

This is not a new blind spot. The Conference Board's earlier research from December found 68% of leaders already admitting their organizations lack the employee skills to capitalize on AI, even as 91% of workers said AI had changed their tasks. The demand for training is there. The supply is not.

What most AI training teaches, and what it skips

Most corporate AI training stops at AI literacy: what the tools are, how to write a basic prompt, what not to paste into a chatbot. The Conference Board found organizations concentrating on these foundational skills while few teach the advanced ones that create value, such as managing AI agents, integrating AI into workflows, and applying it strategically.

Matt Rosenbaum, the report's principal researcher, put it bluntly: "Many organizations have made progress introducing employees to AI, but AI literacy alone will not create business value."

The distinction matters because how people work with AI changes what they get from it. As HR Dive noted in its coverage, workers who direct and refine AI output outperform those who simply delegate to it. Directing and refining are learned behaviors: knowing when the output is wrong, pushing back on it, iterating toward something usable. A one-hour literacy module does not build any of that.

Think of it like teaching someone the rules of chess and calling them a player. Literacy is the entry ticket. The value comes from judgment developed through practice, and practice is exactly what most programs never schedule.

Why human skills decide who gets value from AI

The skills that determine whether AI makes someone more valuable are mostly human ones: judgment, communication, and the willingness to keep learning. A Drexel University survey of more than 600 employers found that across industries and company sizes, hiring managers ranked interpersonal skills, willingness to learn, communication, and dependability above technical and AI abilities.

One HR manager in the study summed up the prevailing view: "Computer skills, excellent customer service and dependability are of utmost importance. The rest, we can train."

The same study carries a warning for the AI era. Employers said they usually noticed AI in applications only "because it's being used so poorly," and sloppy AI use raised doubts about a candidate's judgment and attention to detail. In other words, AI does not hide weak human skills. It amplifies them. A person with poor judgment now produces polished-looking bad work at scale.

For L&D teams, the implication is direct: an AI upskilling program that ignores communication, critical thinking, and judgment is training people to delegate to a machine they cannot evaluate.

What soft skills training returns in hard numbers

Soft skills training is one of the best-documented investments in workforce development, and the returns are large. A National Bureau of Economic Research study by Adhvaryu, Kala, and Nyshadham measured an on-the-job soft skills program for garment workers in India and found trained workers were 20% more productive than the control group. The program returned a net 258% ROI within eight months of ending.

Three details from that study are worth stealing:

• It was on the job. Training happened in the flow of real work, not in a classroom disconnected from it.
• It compounded. Communication training "spurred technical skill upgrading" too, because people who communicate better learn faster from everyone around them.
• The firm captured the gains. Productivity rose 20% while wages rose 0.5%, meaning nearly all the value landed on the employer's side of the ledger.

Pair that with the Drexel finding and the picture is consistent: the skills employers say they cannot hire for, and the skills that produce measurable returns when trained, are the same skills most AI upskilling programs leave out.

How to build AI upskilling that actually works

The fix is to train applied capability, not awareness, and the research points to five concrete moves:

1. Put practice time on the clock. Fewer than half of workers get work-hours time to build AI skills. If skill-building only happens after hours, it only happens for volunteers.
2. Train on real workflows, not generic demos. The Conference Board's leaders said effective skill-building requires "dedicated time, hands-on experience, and managerial support." Teach the claims analyst on claims, the recruiter on requisitions.
3. Go beyond prompting to directing and refining. Build exercises where learners must catch AI mistakes, push back, and iterate, because that is the behavior that separates high performers.
4. Train the human skills alongside the tool skills. Judgment, communication, and learning agility are what make AI output usable. Budget for them like the 258%-ROI investments they are.
5. Make it interactive, and measure behavior. Passive video is where learning goes to die. People build skill by doing, being questioned, and getting feedback in the moment, which is why interactive training videos with an AI tutor that asks questions and adapts to each learner outperform watch-and-forget content. Then measure changed behavior on the job, not completions.

None of this requires a bigger course catalog. It requires fewer, deeper experiences where people practice with the tools, make mistakes safely, and get corrected while the stakes are low.

There is also a trust dividend. The Conference Board found employees are far more optimistic about AI when they believe their organization will help them adapt as the technology evolves. A visible, serious upskilling program is not just capability-building. It is the difference between a workforce that leans into AI and one that quietly resists it.

FAQ

What is AI upskilling and how is it different from AI literacy?

AI literacy is knowing what AI tools are and how to use them at a basic level, like writing a simple prompt. AI upskilling is the broader work of building applied capability: integrating AI into real workflows, directing and refining its output, and strengthening the judgment and communication skills that make the output useful. Literacy is the first step, but on its own it does not change business results.

How much time should employees get for AI upskilling during work hours?

There is no universal number, but the data shows the current answer is "not enough": fewer than half of workers say they get sufficient time at work to build AI skills. A practical starting point is a protected weekly block of 60 to 90 minutes spent practicing on the employee's own real tasks, plus short feedback moments built into normal work, which beats occasional all-day workshops for retention.

Why do soft skills matter more in the age of AI, not less?

Because AI raises the ceiling on output volume while doing nothing for judgment. Employers in the Drexel survey ranked communication, dependability, and willingness to learn above technical skills precisely because those qualities determine whether someone can evaluate, correct, and apply what AI produces. Weak human skills plus powerful AI just means bad work gets produced faster.

The takeaway: your people are already using AI, and the gap between the half who use it weekly and the third who have been trained is where value quietly leaks out. Closing it takes practice-based, interactive learning that builds tool skills and human skills together, not another literacy module. Companies that train for judgment, not just prompts, will be the ones whose AI investment actually shows up in the numbers.