Training soft skills was always the hard part

You can teach the theory of a difficult conversation in an afternoon. Being able to have one, calmly, when someone pushes back, is a different thing, and it’s the part most training never gets around to providing.

AI changes what’s possible here. Instead of multiple-choice training that checks whether someone remembers the theory, training can be dynamic AI practice that checks whether they can do it.

That’s the gap this piece is about: the whole category of hard conversations, with a customer, a prospect, a direct report, things that most training was never built to prepare people for.

Why this has always been hard

Traditional role-play training runs into the same walls again and again. Acting out a scenario in front of colleagues comes with real social pressure, and that pressure tends to crowd out the actual learning. Feedback is only as good as whoever happens to be in the room that day, so two people can practice the same scenario and walk away with completely different notes. And running enough sessions for it to change how someone performs, not just how they feel about a workshop, is slow and expensive to coordinate across a team of any size. We wrote more about this in why role-play works, and how AI is transforming it.

None of this is new, and it was never really about the theory. Most training is fine at explaining what a good conversation looks like [1]. It’s why so much soft skills training ends up being a single workshop people remember for a week, rather than something that changes how they handle the moment it was meant for.

What AI changes about it

AI’s real strength here isn’t replacing traditional training. It’s what happens once you put it next to it. Courses, workshops, and decks are still a reasonably good way to teach the theory of a hard conversation. What they’ve never done well is the part after that: turning theory into something someone can do under pressure [2]. That’s where applied AI fits, giving the theory somewhere to be practiced, not replacing it.

What changed is practical: it’s now possible to practice a real conversation, one where the other side pushes back, gets it wrong, or reacts badly, without needing a colleague free at the same time or an actor on retainer. The AI characters in platforms such as Apprendly RolePlay™ are built to handle this, and a session gets scored against the goal and theory it was set up for, rather than just logged as finished. Whether it’s a manager rehearsing a hard feedback conversation, a rep running a discovery call against a prospect who pushes back, or a support rep sitting with an escalation before facing one live.

A lot of the confusion around AI in training comes down to this: people aren’t sure where it helps or where to use it, so it tends to get pointed at the part that was never the hard part. Faster course creation, auto-generated quizzes, a chatbot bolted onto a help center. All useful, none of it touching the gap this piece is about, which was never the content. It was the practice [3][4].

Topics worth building practice around

A few conversations keep coming up as the ones worth building real practice around, across leadership, sales, and customer service alike. A manager who has to give a direct report feedback they don’t want to hear, and doesn’t know how to open the conversation. A rep who gets hit with an objection they didn’t see coming and has no ready answer. A support rep who picks up the phone to a customer who’s already furious, with only a few seconds to keep it from getting worse.

These aren’t hypothetical. They’re the kind of scenario that shows up in performance reviews and coaching conversations far more often than in a course catalog.

None of it has to run on a generic framework either. Most companies already have their own playbook, training manual, or internal process that describes what good looks like for that specific team, built up over years of doing the job. Practice tends to land better when it’s built around that material directly, not a script written for someone else’s company. It’s part of why Apprendly RolePlay is built to turn a team’s own handbooks, decks, and playbooks into scenarios directly, alongside the well-known frameworks like SPIN, MEDDIC, STAR, and others for teams that use those or a combination instead.

What to look for in a platform

Once you know which of those conversations matters most to your team, the actual shortlist comes down to a handful of things.

1. Your primary scenario

Communication training covers a lot of ground: leadership conversations, customer service, sales calls. Worth being specific about which one matters most before shopping. If your team already works with AI agents in support, sales, or ops, look for a platform that can role-play that handoff moment itself, not just a generic customer conversation. If the real gap is a manager who tenses up giving hard feedback, or a rep who goes quiet the moment a deal stalls, the platform needs to be built around that exact scenario, not a broad one that happens to touch on it.

2. What “good” looks like in the platform

Worth checking: can the AI character in the scenario be wrong, rude, or incomplete, the way a real person sometimes is? Does it measure specific behaviors, like interruptions, missed steps, or tone, instead of a flat pass or fail? And can a non-technical training designer build a new scenario, for a sales pitch or a leadership conversation or a support call, without needing engineering support?

3. Vendor transparency

Worth asking: what happens to the content and conversation data your team generates? How does the AI behind the platform make its decisions, and where can a person step in and override it? Does the pricing and data handling meet your region’s requirements? That matters more for EU teams than it did two years ago. We keep this part simple on our end: Apprendly runs on European infrastructure, and nothing your team generates gets used to train the underlying model, laid out in more detail in our trust center.

4. Budget, same as always

The economics haven’t changed just because the use case did. Factor in content creation time, training administrators, and the cost of training software your team won’t end up using. A platform built around the happy path isn’t really cheaper if nobody ever practices the scenario that matters most.

5. What you get back

Even a single scored playthrough is useful. It tells you how that person handled the conversation, and gives a manager something concrete to work with. Run hundreds or thousands of them across a team, though, and clear patterns start to emerge, the kind of thing that can directly impact KPIs. Some example findings could be: people handle the opening fine but struggle the moment a customer gets defensive, or everyone’s saying the right words but landing flat on empathy. Patterns like that only show up at scale, and they’re a different kind of number than “100% completion” or a pass/fail count, closer to an ROI figure a team can use [5][6]. It’s the kind of pattern Insights, the reporting side of Apprendly RolePlay, is built to surface across a whole team, not just one session at a time. And once the gap has a name, empathy, clarity, whatever it turns out to be, training can point straight at it instead of guessing, with the same numbers showing whether the score moves afterward.

In the end, this isn’t about which platform has the flashiest AI. It’s about whether someone gets to practice the conversation before they’re in it for real, the same gap this piece opened with. That’s what Apprendly RolePlay was built to close: not more content, but somewhere to practice. Worth holding whatever else you’re looking at to the same question.