After coaching my fiftieth manager one-on-one, I noticed something that changed how I thought about the entire coaching industry. The insight that actually helped them wasn’t rare. It wasn’t some breakthrough only a seasoned executive coach could deliver. It was usually a reframe they could have reached with the right questions asked at the right time.
The bottleneck was never insight. It was repetition. It was the five days between sessions where the manager went back to their team, faced the same delegation challenge or the same conflict with a direct report, and had no one to think it through with. By the time we met again, they’d either muscled through it (learning nothing) or avoided it entirely.
That observation led me to build Risely. And this guide is written from that perspective: someone who has coached managers personally, studied what works, built an AI that tries to do it, and has a clear view of where it falls short. I’ll be honest about all four.
Why AI Coaching Exists
The case for AI coaching isn’t a technology story. It’s a math problem.
82% of managers receive zero formal training before or after their promotion. The Conference Board has tracked this for years, and the number barely moves. Not because companies don’t care about management quality, but because the economics of coaching make broad deployment nearly impossible.
Executive coaching costs $200 to $500 per hour. A mid-size company with 200 managers that wants each one to get two sessions per month is looking at $1.4M to $2.4M per year. No L&D budget survives that math. So what happens in practice? The company coaches the top 10 to 15 leaders. Everyone else gets a workshop, an e-learning module, or nothing.
The question that matters isn’t “Is AI coaching as good as human coaching?” It’s “What do we do about the 90% who can’t get a coach at all?”
That’s the gap AI coaching exists to fill. Not replacing the executive coach working with your VP of Engineering. Reaching the 200 managers below that VP who are making daily decisions about how to lead their teams and currently doing it with zero support.
The managers I’ve seen struggle most in my career weren’t the ones who got bad coaching. They were the ones who got no coaching. They were promoted for technical excellence, handed a team, and left to figure out people management through trial and error. Most of them wanted to be good at it. They just didn’t have anyone to practice with.
What AI Coaching Actually Is
The term “AI coaching” gets applied to everything from a chatbot with motivational quotes to a structured skill development platform. Those are not the same thing, and the difference matters if you’re evaluating this for your team.
Three components that make it coaching (not just conversation)
Structured skill progression. Real coaching tracks where someone is in their development and builds on it. If a manager is working on delegation this month, the AI should know what they tried last week, what worked, what didn’t, and what to explore next. This requires a defined skill framework, not just free-form conversation.
Behavioral reinforcement between sessions. A coaching conversation once a month changes very little behavior. The research on skill development is clear: practice needs to happen close to the moment of application. AI coaching that only shows up when the user opens the app misses the point. Daily nudges, micro-practices, and contextual reminders are what bridge the gap between understanding a concept and actually doing it.
The third requirement is contextual conversation. The AI needs to coach on the user’s real situations, not generic scenarios. When a manager says “My direct report keeps missing deadlines and I don’t know how to address it without damaging the relationship,” the coaching response should explore that specific situation, not deliver a lecture on accountability frameworks.
If a platform has all three, it’s doing something that resembles coaching. If it’s missing one or more, it’s something else. Maybe useful, but not coaching.
AI coaching vs chatbot vs LMS
This comparison comes up in almost every conversation I have with HR leaders.
| Capability | AI Coaching Platform | General AI Chatbot (ChatGPT, Claude) | LMS / E-Learning |
|---|---|---|---|
| Conversation memory | Remembers past sessions, tracks ongoing challenges | Starts fresh each conversation (or limited context) | No conversation at all |
| Skill tracking | Measures specific competencies over time | No skill measurement | Tracks course completion, not skill change |
| Progression | Adapts difficulty and focus based on growth | Same level regardless of user growth | Linear course sequence |
| Proactive outreach | Daily nudges, reminders, contextual prompts | Only responds when prompted | Email reminders to complete modules |
| Organizational analytics | HR dashboard showing team-level skill data | None | Completion rates and quiz scores |
| Cost per user | $10 to $80/month | $0 to $20/month | $5 to $30/month |
A chatbot can answer a coaching question. It can’t coach you over 12 weeks toward measurable improvement in a specific skill. An LMS can teach you the theory of effective feedback. It can’t help you practice it on Tuesday morning before your one-on-one. These distinctions matter when you’re building a business case. For a deeper look at how AI coaching compares to live training programs, we’ve written a separate analysis.
What the Research Actually Says
I want to be careful here because the research base on AI coaching is real but young, and I have an obvious bias as someone who built an AI coaching platform. So let me separate what we know from what we don’t.
The manager training gap is well-documented
The Conference Board’s data on the 82% training gap is consistent with broader research on management development. Gallup has reported that companies fail to choose the right candidate for manager 82% of the time. DDI’s Global Leadership Forecast consistently finds that leadership development pipeline quality is rated as low by over half of HR professionals surveyed. The problem isn’t controversial. Most organizations acknowledge it. They just haven’t been able to solve it at scale.
AI coaching can build working alliance
The most relevant academic work comes from Terblanche et al., published in Frontiers in Psychology, which examined the working alliance in AI coaching. Working alliance is the psychological term for the relationship quality between coach and client. It’s considered one of the strongest predictors of coaching outcomes.
Their finding: for structured skill development, the working alliance in AI coaching was comparable to what you see in human coaching. Users developed trust in the AI, felt heard, and agreed on goals. This doesn’t mean AI coaching equals human coaching across all dimensions. It means the specific mechanism through which coaching works (the relationship) can form with an AI when the coaching is structured well.
Where the evidence is still thin
I’d be misleading you if I stopped there. The honest picture includes several gaps:
Longitudinal data is sparse. Most studies look at outcomes over weeks or months. We don’t have decade-long studies comparing AI-coached leaders to human-coached leaders. We’re building that evidence base in real time.
Most studies are vendor-funded. Including our own outcome data at Risely. That doesn’t make it wrong, but it means you should weight independently conducted research more heavily.
Senior leader data is also limited. The research on AI coaching mostly covers mid-level managers and individual contributors. For senior executives, the evidence for human coaching is much stronger, and the evidence for AI coaching is thin.
Self-selection bias. People who opt into AI coaching may already be more motivated to develop. Separating the tool’s effect from the user’s motivation is hard.
I share these limitations because intellectual honesty is more persuasive than cherry-picked data, and because I think the case for AI coaching is strong enough that it doesn’t need inflated claims to stand on its own.
Where AI Coaching Works (and Where It Doesn’t)
Where it works
Daily skill practice for managers and ICs. When someone is learning to give better feedback, run more effective one-on-ones, or manage conflict constructively, they need to practice regularly with real situations. AI coaching delivers this at a cost that makes all-employee deployment possible.
Reaching everyone, not just the top 10. The single biggest advantage of AI coaching over human coaching is coverage. A company with 500 managers can give all 500 access to daily coaching for less than the cost of coaching 15 of them with humans.
There’s also a stigma problem that AI coaching quietly solves. Asking for help with a people skill is still uncomfortable in most workplaces. Talking to an AI removes the social risk. Users I’ve spoken with describe it as being able to “think out loud without being judged.” This is particularly true for managers who worry that admitting they’re struggling with their team will be seen as weakness.
Consistency. Every user gets the same quality of coaching. There’s no lottery of whether you’re matched with a great coach or a mediocre one. The methodology is applied the same way every time, and it improves for everyone simultaneously when the platform updates.
Speed to deploy. A human coaching program takes weeks to months: coach matching, intake calls, scheduling, contracts. An AI coaching program can start the same week you decide to run a pilot.
Where it doesn’t work
I’m going to be direct about this because I think overclaiming is the fastest way to lose credibility.
C-suite and senior executive development. When a Chief People Officer is dealing with a board that wants to restructure the leadership team, or a VP is managing organizational politics around a merger, the coaching challenge isn’t a skill gap. It’s navigating genuine ambiguity where the right answer isn’t clear and the consequences of getting it wrong are significant. Human coaches who understand organizational dynamics, power structures, and strategic context are better equipped for this work.
Deeply personal challenges are a second boundary. Grief, trauma, burnout that has crossed into clinical territory, mental health struggles that show up at work. These need a human professional, often a therapist rather than a coach. An AI coaching platform should recognize when someone needs a different kind of support and say so clearly. We’ve built this into Merlin, and I consider it non-negotiable for any AI coaching provider.
Passive or hostile users. When coaching is assigned as a corrective measure and the person has already decided they don’t have a problem, AI coaching produces nothing. Human coaching can sometimes break through that resistance because a skilled coach can build rapport and earn trust over multiple sessions. An AI has a harder time with someone who’s determined not to engage.
The coaching conversations I’ve seen fail most often weren’t where the person was struggling. They were where the person had already decided they didn’t have a problem. No amount of good coaching methodology fixes that, but a human coach has more tools to work with than an AI does.
Organizational dysfunction. If the problem is a toxic culture, a structurally broken reporting relationship, or an executive team that’s misaligned on strategy, coaching the individual manager to cope better is treating the symptom. The organization needs to change. AI coaching can help someone survive a bad system for a while, but it can’t fix the system.
The honest answer
AI coaching and human coaching are complementary. The right question for most organizations isn’t “which one?” but “who gets which?” For a detailed breakdown of how they compare across every dimension, we’ve published a full comparison.
How to Evaluate an AI Coaching Platform
If you’re considering AI coaching for your team, these are the five questions I’d ask any vendor, including us. For a broader view of coaching platform options for scale, the leadership coaching platform guide covers the full landscape.
1. What’s the coaching methodology?
Can the vendor show you their skill framework? How many skills do they cover? How did they build it? A coaching platform grounded in behavioral science should be able to explain its approach in detail. If the answer is “we use GPT-4 with a coaching prompt,” that’s a chatbot with branding, not a coaching platform. Ask to see the skill taxonomy. If one doesn’t exist, move on.
2. How does it measure skill change, not just completion?
Course completion and session counts tell you nothing about whether people are actually improving. Ask: can you show me how a user’s delegation skill moved from week 1 to week 12? Can I see team-level trends? If the platform only tracks “sessions completed” or “satisfaction scores,” it can’t prove its own value. That makes your business case impossible to defend when renewal comes up.
3. Does it reach users between sessions?
The difference between a coaching conversation and actual behavior change is what happens in the days between conversations. Ask about daily nudges, micro-practices, and contextual reinforcement. A platform that only works when the user opens it is fighting against how busy people actually operate. This is where most platforms are weakest, and it’s where the real impact happens.
4. Can you pilot with 10 people this week?
If the answer involves a 6-month enterprise sales cycle, POC planning committee, and three rounds of procurement, you’ll lose organizational momentum before anyone gets coached. The best way to evaluate AI coaching is to try it with a small group and look at the data. Ask whether you can start small and expand based on results. If the vendor needs a large commitment upfront, ask why.
5. What does HR see vs what does the user see?
Privacy architecture matters more than vendors acknowledge. If users suspect their coaching conversations are visible to their manager or HR, they won’t be honest in sessions, and the coaching won’t work. Ask specifically: who can see conversation content? What data does HR receive? Is user data used to train the model?
I’ll add one more question that I think vendors should be asked even though it’s uncomfortable for us: Where does your platform fall short? If the vendor can’t name a single limitation, they’re not being straight with you.
What We Built at Risely and Why
I’m going to shift from category-level advice to our specific product, so you know where the line is. Everything above applies regardless of which platform you choose. This section is about our design decisions and the reasoning behind them.
Merlin: what a real coaching conversation looks like
Merlin is the AI coach inside Risely. Rather than describe features, let me walk through what a session actually looks like.
A manager named Sarah has a direct report who’s been consistently late on deliverables. She’s avoided the conversation for two weeks because the last time she gave this person feedback, it went badly and the relationship got strained.
Sarah opens Merlin in Slack. She types: “I need to have a difficult conversation with my direct report about missed deadlines but I’m worried about making things worse.”
Merlin doesn’t give her a script. Instead, it asks: “What happened the last time you gave them feedback? What part of it went differently than you expected?”
Sarah explains the previous situation. Merlin helps her identify that she combined feedback on the missed deadline with frustration about a separate issue, which made the direct report feel ambushed. Together, they work through how to separate the two topics, what to say first, and how to create space for the direct report to respond.
The session takes 12 minutes. Sarah walks into the conversation that afternoon with a plan. The next morning, Merlin follows up: “How did the conversation go? What worked and what would you do differently next time?”
That follow-up is where the learning actually happens. Not in the planning session, but in the reflection after the real interaction. This is the piece that traditional coaching misses when sessions happen monthly.
83 skills, daily practice, measurable outcomes
When I designed Risely’s skill framework, I started from a specific question: what are the actual behaviors that make someone effective at leading people?
Not competency labels. Behaviors. The difference between a manager who says “I’m good at delegation” and one who actually assigns the right tasks to the right people with the right context and follows up without micromanaging.
That research produced 83 distinct skills covering over 1,000 O*NET occupations. Each skill has defined behavioral indicators, so when Merlin coaches someone on delegation, it knows what “good delegation” looks like and can track whether the user is progressing toward it.
The outcomes we’ve measured across 5,000+ users in 40+ organizations: 26% average skill improvement in 12 weeks. 87% of invited users engage in week one. 82% are still active at day 30. 73% engage with daily nudges.
Those numbers are strong, and they’re our own data, so weight them accordingly. What I find more convincing than aggregate statistics is watching individual users develop. A manager who couldn’t delegate without re-doing the work themselves, learning over six weeks to actually let go. A new team lead who was terrified of giving constructive feedback, getting comfortable with it through repeated practice in Merlin before trying it live.
What Risely doesn’t do
Senior executive development. If you need coaching for your C-suite, hire a human executive coach. Merlin is built for managers and individual contributors working on workplace skills.
Crisis support. Merlin will recognize when a conversation moves into mental health territory and recommend professional support. It won’t try to play therapist.
Organizations where the structural problem is at the top. If leadership behavior is the issue, coaching the middle managers to cope isn’t the answer. We’ll tell you that in the sales conversation too.
For a look at how Risely compares to specific alternatives, we maintain a comparison of the top AI coaching platforms.
Frequently Asked Questions
What is AI coaching?
AI coaching uses artificial intelligence to deliver structured coaching conversations that help people develop workplace skills. Unlike generic AI assistants, a real AI coaching platform has a defined skill methodology, tracks progress over time, remembers past sessions, and provides daily reinforcement between conversations. It’s designed to make coaching accessible to people who would never get access to a human coach.
Is AI coaching as effective as human coaching?
For structured skill development (delegation, feedback, conflict resolution, communication), research shows that AI coaching produces comparable outcomes. For senior leadership development, complex organizational challenges, and deeply personal growth work, human coaching is more effective. Most organizations benefit from using both: AI coaching for all employees, human coaching for senior leaders.
How much does AI coaching cost?
AI coaching platforms range from $10 to $80 per user per month. Risely is $59/user/month or $399 for a team of 5. Enterprise pricing runs $700 to $1,000 per user per year. Human coaching through platforms like BetterUp or CoachHub costs $3,000 to $5,000 per user per year. Independent executive coaches charge $200 to $500 per hour.
What’s the difference between AI coaching and just using ChatGPT?
ChatGPT can answer a coaching question. It can’t coach you. It has no skill framework, no memory of your development journey, no way to track whether you’re improving, no daily nudges to reinforce behavior, and no organizational analytics. It starts fresh every conversation. Using ChatGPT for coaching is like asking a knowledgeable stranger for advice each time you have a problem. It might be helpful in the moment, but it doesn’t compound into development.
Can AI coaching be used alongside human coaching?
Yes, and this is the approach I recommend for organizations that can afford both. Human coaching handles the strategic, relationship-intensive development work for senior leaders. AI coaching provides the daily practice layer for everyone else. Some organizations also use AI coaching as the between-session reinforcement for leaders who have human coaches, so the coaching doesn’t stop when the hour is over.
Try It Yourself
I’ve given you 3,000 words of context, research, and opinion. But the best way to evaluate AI coaching is to experience it.
Open Merlin, bring a real situation you’re dealing with at work right now, and have a coaching conversation. It takes about 5 minutes. No signup required, no credit card, no one will call you.
Then decide for yourself whether this is coaching.