AI coaching sounds clever. But most tools fall flat when it comes to soft skills. Why? As Petr Kanaev explains, it’s not just about algorithms, it’s about data. Without context, outcomes and expert input, the feedback fails. The good news? With the right foundation, AI can help your employees grow
AI-powered coaching tools are gaining traction in workplace learning – especially as organisations seek scalable ways to develop soft skills in hybrid and remote settings. From conversational agents to automated feedback platforms, the promise is clear: consistent, personalised support that helps people build behavioural capability.
But here’s the reality: without the right data – and without meaningful connections between data points – no AI solution will deliver on that promise.
To provide meaningful feedback, AI needs more than conversation transcripts or keyword tracking
This article explores the kind of data that actually matters, the common pitfalls that undermine AI-driven learning tools, and what L&D teams can do to create a foundation for sustainable, measurable impact.
What your AI needs to be effective
To provide meaningful feedback, AI needs more than conversation transcripts or keyword tracking. It requires structured access to:
- A behavioural competency model with clearly defined indicators of effective behaviour.
- Contextual information – including roles, goals, constraints and emotional tone.
- Complete interaction data – capturing contributions from both sides.
- A clearly defined outcome – ideally with a measurable result.
- A set of expert-reviewed examples – showing how professionals evaluate similar behaviours in comparable situations.
Without this foundation, AI is left guessing – and the feedback becomes superficial or misleading.
Three reasons AI coaching fails in practice
These are the most common mistakes that cause AI coaching tools to miss the mark – and how to avoid them:
1. Vague behavioural models
If your competency model relies on broad, unstructured language, AI will struggle to assess performance reliably.
Example: The phrase “asks effective questions” sounds actionable – but what does it mean? Does it involve open-ended phrasing, strategic sequencing or clarifying intent?
Without specific behavioural indicators, feedback becomes vague or irrelevant.
2. Missing context and outcomes
Soft skills are deeply contextual. Whether a behaviour is effective depends not only on what the employee says or does, but also on how the other party responds and what result the interaction produces.
Example: An employee offers a 4% discount. On the surface, that seems reasonable, especially if company policy allows up to 10%. But if the client was ready to accept full price, the discount reflects a missed opportunity.
Without insight into the client’s goals, reactions and constraints – and without knowing the outcome – the AI cannot judge whether the decision was appropriate or wasteful.
That’s why the most effective AI learning systems use structured role-play simulations, where both sides’ motivations are predefined and outcomes are measurable. This gives AI the context it needs to evaluate behaviour with real-world relevance.
3. No access to expert judgment
AI coaching in soft skills does not improve simply by observing interactions. It needs structured human input – examples that are labelled, explained and contextualised by experts.
Example: In one scenario, assertiveness might be seen as confident leadership. In another, the same behaviour could harm trust and derail collaboration.
Without expert framing, the AI cannot recognise these nuances – or reflect them in its feedback.
Unfortunately, datasets like this are rarely available in the public domain. That’s why it’s essential to build your own feedback loop internally: a system where skilled facilitators or coaches annotate soft skill scenarios with real-world insight. This ongoing expert input is what allows AI to stay relevant, improve over time, and reflect how people assess soft skill effectiveness.
Final thought: It’s all in the data
Whether you’re building an AI-powered coaching tool in-house or buying one off the shelf, ask this first: What kind of data is it built on – and how does it connect that data to context, outcomes and expert understanding?
Petr Kanaev is CEO and Founder of The Reflection – Next-Gen LXP for Skill Mastery
