Why Soft Skills Training Is Hard to Scale in Universities

For higher education leaders and educators responsible for employability outcomes

Picture a final-year business cohort of 240 students. The module leader wants every one of them to practise a difficult client conversation before their placement starts. In a class of twelve, that's an afternoon of paired roleplay with feedback from a tutor who can watch every exchange. In a class of 240, it's twenty afternoons, a room booked for a fortnight, and a marking load no one has budget for. Most programme teams don't choose the twenty afternoons. They choose a case study, a short video, and a single reflective essay instead.

This comes down to arithmetic, not a lack of imagination on the part of academic staff. Roleplay-based skills training has always worked best in small groups, with a person watching and responding in real time. Universities have grown, budgets have tightened, and the maths hasn't kept up.

The same scene plays out across nursing programmes preparing students for difficult patient conversations, engineering courses trying to build client-facing confidence, and business schools running mock negotiations. Whatever the discipline, the pattern is identical: a skill that genuinely needs practice and feedback gets replaced with something that can be delivered to a lecture hall instead, because a lecture hall is what the budget and the timetable allow.

This article looks at why soft skills training is so hard to scale in higher education, what the data says about the resulting gap between what employers need and what graduates can do, and how AI-powered roleplay is starting to close it without asking already-stretched academic teams to do more with less.

The gap employers keep reporting

Employers have been consistent for years about which skills they want from new graduates. They've been just as consistent about not getting enough of them.

The World Economic Forum's Future of Jobs Report 2025 surveyed employers representing more than 14 million workers across 55 economies. Analytical thinking topped the list of core skills, rated essential by seven in ten companies, followed by resilience, flexibility and agility (67%), leadership and social influence (61%), and creative thinking (57%). None of these are technical skills. All of them are built through practice and feedback, not through lectures. The same report names skills gaps as the single biggest barrier to business transformation.

LinkedIn's Workplace Learning research found that communication was the single most in-demand skill among employers in 2024, and nine in ten global executives say human skills matter more now than they used to. Meanwhile, the National Association of Colleges and Employers (NACE) has tracked a persistent mismatch between how new graduates rate their own proficiency and how the employers hiring them see it.

NACE's most recent analysis, drawn from its 2024 Student Survey of over 20,000 students and its Job Outlook 2025 employer survey, found that the gap in perceived proficiency runs to roughly 25 percentage points for communication and critical thinking, and exceeds 30 points for leadership and professionalism. Teamwork is one of the few areas where students and employers broadly agree.

The AAC&U's 2023 employer survey, conducted by Morning Consult with just over 1,000 executives and hiring managers, put a sharper point on it: 64% of employers call oral communication a very important skill for new hires, but only 34% think recent graduates are very well prepared in it. Oral communication showed the largest gap of any skill measured. It's also one of the hardest skills to teach without giving someone a chance to actually speak, be interrupted, and recover.

The Cengage Group's 2025 Graduate Employability Report, which surveyed graduates, employers, and for the first time educators, found that nearly nine in ten educators believe their students are ready for the workforce. Just 48% of graduates agree, a gap wide enough that these sound like two different graduating classes rather than two groups with a minor difference of opinion.

Employers and graduates agree on what matters. They don't agree on whether graduates can actually do it, and the gap is widest in exactly the skills that require rehearsal rather than reading.

Why universities can't just add more roleplay

None of this is news to programme leaders. Most already know their students need more practice at difficult conversations, presentations, and interviews. What's missing is capacity, not awareness.

Class sizes are moving in the wrong direction for this kind of teaching. Cardiff University's restructuring, reported by Times Higher Education in February 2025, will see some schools move to a ratio of one academic for every 24 students, and vice-chancellors quoted in the same piece described further class-size increases as inevitable given the sector's finances.

Those finances are genuinely strained: the Office for Students' November 2025 update projects that 45% of English higher education providers will run a deficit in 2025-26, up from 34% forecast earlier the same year. Academic staff, meanwhile, are already stretched. UCU's workload survey, cited by Times Higher Education, put the average working week for UK academics at close to 51 hours, nearly two full days more than a standard contracted week.

Put a rising ratio of students per academic together with an already overloaded working week, and the arithmetic from the opening example gets worse every year, not better.

The traditional answer to this problem, standardised simulation, is itself expensive to run at scale. Wolters Kluwer's cost analysis of clinical simulation centres puts start-up costs at $200,000 to $1.6 million, with at least $15,000 a year in ongoing maintenance, and individual patient simulators can run to $100,000 each.

A 2024 study in the Journal of Nursing Education compared the cost per unit of learning benefit across three formats and found manikin-based simulation cost more than four times as much per hour as screen-based virtual simulation. Actor-based mock interviews and standardised-patient programmes carry similar per-session costs, because they depend on paying and scheduling real people.

Feedback consistency suffers too. Writing in The Conversation, academics reviewing university communication teaching noted that large classes make it hard to maintain consistency across the teaching teams needed to staff them, and that shrinking budgets push assessment towards peer and group work rather than individual expert feedback.

That's the quiet cost of scale: not that practice disappears entirely, but that the quality and consistency of feedback drops as cohorts grow, right when employers are telling us that feedback-dependent skills are the ones graduates need most.

There's a fairness dimension here too, one that rarely gets discussed alongside the budget conversation. When roleplay capacity is limited, someone has to decide who gets it. In practice, that's often the smaller cohort, the specialist pathway, or the students who happen to be timetabled into a seminar group with a particularly committed tutor.

Widening participation students, commuting students juggling part-time work, and those on placement-heavy courses with less flexible timetables are frequently the ones squeezed out first when practice sessions are the first thing to be cut from an overstretched module. A scaling problem quietly becomes an access problem.

Students, for their part, are asking for more of this kind of practice, not less. Inside Higher Ed's 2025 Student Voice survey of more than 5,000 students found strong demand for hands-on, work-based learning, and noted that more than three students now compete for every high-quality internship placement.

NACE's research on early-career graduates found that those who took part in experiential learning report higher career satisfaction and higher salaries than those who didn't, tracked through its Gen Z early-career survey. Students know they need the reps. Institutions know it too. The constraint is delivery, not desire.

What scalable practice actually looks like

This is where AI-powered roleplay changes the calculation, because it attacks the specific bottleneck described above: the need for a person to be present, watching, and giving feedback every single time a student practises.

The most detailed evidence on cost and scale comes from PwC's study of VR soft-skills training, which followed new managers across twelve locations trained through classroom sessions, e-learning, or VR roleplay. Learners trained through immersive roleplay were up to 275% more confident to act on what they'd learned, a 40% improvement over classroom training, and completed training four times faster.

The part that matters most for a scaling argument is the cost curve: VR training reached cost parity with classroom delivery at 375 learners, reached parity with e-learning at 1,950 learners, and became 52% cheaper than classroom training once cohorts hit 3,000. Traditional roleplay gets more expensive as it scales. AI-powered roleplay gets cheaper.

Early studies inside higher education point the same way, though the evidence base here is still young. A 2025 study published in the European Journal of Investigation in Health, Psychology and Education had 30 medical students practise patient consultations with an AI-driven virtual patient and automated feedback.

Average communication scores rose by nearly 37 points on an eight-domain assessment rubric, and the proportion of students rated high-performing went from zero to 70%. A separate 2025 study in Cureus had 27 medical students across three UK universities roleplay difficult conversations with an AI patient and found significant, large gains in self-reported confidence for breaking bad news and counselling anxious patients.

These are small, single-institution studies, and they measure confidence and immediate performance rather than long-term retention, so they should be read as promising early signals rather than settled proof. More rigorous designs are starting to appear, including a controlled crossover trial across two UK universities comparing AI-driven communication training against actor-based training, which found both approaches effective, with the AI-based version substantially cheaper to run. That's the pattern worth watching: comparable outcomes to the gold standard, at a fraction of the delivery cost, without asking one more hour of anyone's already-full week.

The skills employers say graduates lack most are the ones that can only be built through repetition and feedback, and repetition and feedback are exactly what shrinking staff-to-student ratios make hardest to deliver.

What this means for programme leaders

None of the above suggests scrapping human-led roleplay. Actor-based simulation and live mock interviews remain valuable, particularly for high-stakes, nuanced scenarios where a skilled facilitator adds something a model can't.

For most programme teams, the real challenge is giving every student in a 240-person cohort the same number of practice attempts that the twelve-person seminar group already gets, without recreating the staffing and scheduling problem that made this hard in the first place. That's a different question from whether to replace people with AI.

A few things follow from the evidence above. First, look for tools that let students fail safely and repeat a scenario as many times as they need, since the confidence gains reported above come from repetition, not a single pass. Second, prioritise consistency of feedback over volume of content: the NACE and AAC&U gaps are largest exactly where feedback has traditionally been most person-dependent, so a structured, repeatable rubric matters more than adding another lecture on communication theory.

Third, keep assessment and reflection built into the same flow as the practice itself, rather than bolting a quiz on afterwards, since the research above measured competence gains within an integrated practice-and-feedback loop, not from practice alone.

It's also worth being honest with academic teams about what this changes and what it doesn't. AI roleplay doesn't remove the need for a module leader to design a good scenario, set a clear rubric, or interpret what the data is telling them about a cohort's progress.

What it removes is the requirement that a member of staff be physically present for every single attempt a student makes. That distinction matters when planning workload models and when reassuring staff that this is a tool for extending their reach, not a replacement for their expertise.

Bodyswaps' pedagogy and AI coaching approach was built around exactly this loop of repeatable practice, structured feedback, and integrated reflection. Our work with higher education institutions has focused on giving every student, not just those in small seminar groups or on well-resourced pathways, the same repeatable access to practice and feedback, drawing on a growing library of scenarios that programme teams can adapt to their own discipline.

Book a demo →

The bottom line

Universities aren't failing to prepare graduates in communication, leadership, and critical thinking because they don't see the problem. Class sizes, staff workloads, and institutional budgets have made the traditional fix, more roleplay with more people watching, structurally harder to deliver every year.

AI-powered roleplay doesn't remove the need for good pedagogy or skilled facilitation. It removes the constraint that used to tie feedback-rich practice to the number of staff available to give it, which is the constraint that was making soft skills training so hard to scale in the first place.

Further reading

References

1. World Economic Forum (2025). Future of Jobs Report 2025. weforum.org
2. LinkedIn (2024). The Most In-Demand Skills of 2024. linkedin.com
3. National Association of Colleges and Employers (2025). The Gap in Perceptions of New Grads' Competency Proficiency and Resources to Shrink It. naceweb.org
4. American Association of Colleges and Universities (2023). The Career-Ready Graduate: What Employers Say about the Difference College Makes. aacu.org
5. Cengage Group (2025). 2025 Graduate Employability Report. cengagegroup.com
6. Times Higher Education (2025). Class size increases 'inevitable' amid financial crisis, say v-cs. timeshighereducation.com
7. Office for Students (2025). Significant challenges continue to face higher education finances — with nearly half facing deficits in 2025-26. officeforstudents.org.uk
8. University and College Union workload survey, as reported by Times Higher Education / AcademicJobs. academicjobs.com
9. Wolters Kluwer (n.d.). Simulation Center Finances: Calculating Fees and Costs. wolterskluwer.com
10. Journal of Nursing Education (2024). A Cost-Utility Analysis Comparing Traditional Clinical, Manikin-Based Simulation, and Screen-Based Virtual Simulation Activities. pubmed.ncbi.nlm.nih.gov
11. The Conversation (2021). Communication is changing — and most universities haven't kept up. theconversation.com
12. Inside Higher Ed (2025). College Students Want Work-Based Learning Experiences. insidehighered.com
13. National Association of Colleges and Employers (2025). Impacts of Experiential Learning on the Gen Z Early-Career Experience. naceweb.org
14. PwC (2020). Understanding the Effectiveness of VR Soft Skills Training in the Enterprise. pwc.com
15. Suárez-García et al. (2025). DIALOGUE Study. European Journal of Investigation in Health, Psychology and Education, 15(8), 152. mdpi.com
16. Mukadam, Suresh & Jacobs (2025). Cureus, 17(5):e84381. pmc.ncbi.nlm.nih.gov
17. JMIR Formative Research (2025). Multicentre crossover trial comparing AI-driven and actor-based communication training. formative.jmir.org

FAQ

Why is soft skills training hard to scale in universities?

Because it has traditionally depended on a person watching and giving feedback in real time. As class sizes grow and academic workloads rise, there aren't enough staff hours to give every student the same amount of practice a small seminar group used to get.

What soft skills do employers say graduates lack most?

Communication, critical thinking, leadership, and professionalism show the biggest gaps between how graduates rate themselves and how employers rate them, according to NACE and AAC&U research.

Can AI roleplay really replace traditional practice methods?

Early evidence suggests AI-powered roleplay can match or come close to actor-based training on outcomes, while costing far less to scale, though most higher education studies so far are small and researchers are still building a fuller evidence base.

Does AI-powered practice work as well as human-led roleplay?

Studies comparing the two report similar effectiveness for many communication scenarios, with AI training becoming markedly more cost-effective as the number of students grows.

Try Bodyswaps →