SONATAnotes
What If Every Human Worker Had an AI Coach?
World bodybuilding champions Derek Lunsford and Chris Bumstead both work with legendary trainer Hany Rambod up to six times a week. Likewise, former tennis champion Conchita Martinez has coached multiple #1-ranked players on the women’s pro circuit. Outside of sports, Jeff Bezos, Steve Jobs, and Oprah Winfrey have all hired executive coaches.
As Bill Gates said in a 2013 TED talk… “Everyone needs a coach. It doesn’t matter whether you’re a basketball player, a tennis player, a gymnast or a bridge player. We all need people who give us feedback. That’s how we improve.”
Of course, while everyone can benefit from a coach, not everyone can afford one. Great coaches are rare, busy, and expensive: Christina Martinez charges up to $1,000 for a tennis lesson, and the best executive coaches charge tens or hundreds of thousands of dollars per year.
Meanwhile, most workers have to settle for less-than-stellar coaching from their bosses: which is a problem given how, in a 2018 survey, fewer than one-third of HR leaders thought their organization’s managers did a good job of coaching employees.
So, do we just have to accept a “coaching gap” between the top tier of professionals and the mass of workers? Or is there a way we could democratize coaching, and give every worker access to an expert coach?
Based on my company’s experience designing coaching programs for Fortune 500 / Global 1000 companies and major international nonprofits, I honestly believe that performance coaching is an area where AI can be a tremendous help to workers at all levels.
AI as Coach
Most of the talk around AI in the workplace has focused on machines taking over people’s jobs. But while it’s true that certain jobs will be automated away in the coming years, the bigger opportunity might be AI helping humans perform better in their jobs.
To put AI coaching in perspective, it’s worth pointing out that not all top-performing human employees make good managers or coaches, and likewise not all coaches were top performers. For every champion-turned-coach like Christina Martinez there’s a Hany Rambod, who never won a major competition in his own bodybuilding career but coached clients to a combined 24 Olympia titles, or Tim Gunn, who is famous for mentoring fashion designers like Christian Siriano but whose own professional background is in sculpture, not fashion!
The point here is that coaching someone to do a job is a distinct skill set from doing the job, and it’s a skill set that plays to AI’s strengths. While coaches may give the occasional hands-on demonstration, coaching is largely about having conversations, and that’s something AI does quite well… or at least AI can do well, if given the right direction.
That said, when we propose AI serving as a “coach” for human workers, we need to be clear exactly what we’re talking about. Before getting involved in AI development, our company helped design traditional coaching and training programs for leading organizations in fields ranging from finance to healthcare to tech, manufacturing, construction, and the nonprofit sector.
In each of those programs, coaches played slightly different roles, and the same applies when developing AI coaching experiences.
AI as Instructor
Our company has helped organizations create “blended” learning courses that combine videos or e-learning with live instructor-led training sessions. And our advice to clients is usually “Have the e-learning cover the theory and background knowledge, so instructor-led sessions can focus on application.” And the same applies to AI instructors.
Having a human or AI coach repeat the same rote information again and again is a waste of resources: videos can do that better. What humans and AIs do better is asking “OK, now how do these principles apply in situation X?” and talking learners through various scenarios.
Recognizing that, most of our work to date using AI in training courses has focused on having it generate interactive role play and simulation activities, where the user is presented with situations and asked what they would do, then given feedback by the AI – like this training simulation to help staff members at financial institutions give better advice to customers:
AI as Advisor
Perhaps the most valuable function of coaches is helping coachees think through specific real world challenges. And, when given the right source date and a framework of questions to ask, AI can be a highly knowledgeable, always available, infinitely patient advisor.
For example, see the Sales Enablement and Physiotherapy virtual coaching modules below, which are intended to guide salespeople on how to position their products and physiotherapists on how to prescribe exercises for patients, respectively.
In the physiotherapy copilot we even encouraged the AI to challenge the user’s thinking with questions if they recommended something that seemed inadvisable, in much the same manner as a human coach:
The Cost of AI Coaching at Scale
At first, it might be tempting to compare AI-generated training simulations and virtual coaches to traditional forms of “computer-based training” such as e-learning, videos, and knowledge bases. However, when you dig into the use cases for AI-powered training, they bear a much closer resemblance to human coaches than e-learning.
For starters, the cost of AI training is variable. Where an organization can produce an hour-long e-learning module for a one-time cost of $10,000 to $20,000 then deliver it to tens or hundreds of thousands of workers at near-zero additional cost, AI simulations cost more the more people use them.
This is because the makers of AI models like ChatGPT, Gemini, and Claude charge for usage based on “tokens” – units of data representing words, parts of words (such as prefixes or suffixes), and even punctuation marks.
While the math for calculating the number of tokens consumed during an average AI conversation gets convoluted, the cost to run a 20-minute session where the user submits 15 messages to the AI using the latest model of ChatGPT would cost around $1.25 just for the computing power involved. Meanwhile, the best AI interaction designers (“prompt engineers”) typically charge a license fee for their work on top of what the platform provider charges for processing, bringing the cost of our hypothetical session up to $2, $3 or more.
AI Coaching vs. E-Learning
If we compare the cost of AI sessions to “traditional” computer-based training, it costs $5,000 to produce a 20-minute e-learning module and $5,000 to produce an AI interaction that people interact with for 20 minutes, the organization will need to spend another $30,000 to let 10,000 people run the AI simulation three times, whereas it will cost nothing extra for them to view the e-learning.
But, again, e-learning is a poor point of comparison. E-learning can’t tailor the substance of a session to every individual learner, it can’t listen to what a user says, and it can’t offer truly personalized feedback. Those are things only AI and human coaches can do.
AI Coaching vs. Human Coaching
So… how does the cost of AI coaching compare to coaching delivered by a human?
If we examined the same 10,000-person training audience, it would take human coaches around ten thousand hours – or 1,250 working days – to walk everyone through the same 20-minute coaching conversation or role play activity three times. So we’re already talking about hiring a team of coaches, not an individual coach.
If the coaches only cost $65 per hour (a typical rate for a freelance nurse educator, without any agency markup) that would be $195,000, and that’s assuming zero time for preparation and no travel expenses. And if the coaches cost $350 per hour (a typical rate for a reputable sales or management skills trainer) the total would be $1,050,000.
Suddenly, paying $35,000 or $65,000 or, in the case of the million-dollar sales training program, $245,000 for the AI option seems like a tremendous value.
The Role of AI On a Coaching Team
Are we saying that organizations should replace their human coaches with AI? Not at all.
Going back to the point made earlier, the real promise of AI lies not in replacing human workers, but in enhancing and extending human capabilities, and allowing individuals and organizations to accomplish things that would have been impossible (or unimaginable) before. And this applies to coaches as well.
In our hypothetical 10,000-participant coaching program, for most organizations the idea of hiring twelve human coaches to work 100 days each to provide each participant with an hour of one-to-one consultation, role play exercises, or assistance with specific tasks is simply out of the question, financially and logistically. However, with AI that same personalized attention can be delivered to everyone in the program within a single day (or on demand, at each learner’s convenience) and human coaches could even review the transcripts and feedback to help them tailor the content of live group sessions.
Recognizing that, even if we said an AI’s time was only worth half or a third of a human coach’s time in terms of impact (setting aside the possibility that a well-designed AI coach might actually be better in some cases), it would still make good sense to leverage AI coaches, for instructional and financial reasons.
Given all this, one can easily imagine AI serving a similar function to teaching assistants in a large university class or a “nurse practitioner” on a medical team, capable of performing many of the more routine tasks of an experienced professor or physician, but at a (dramatically) lower hourly rate, thus making the training program more accessible to all. Plus, the AI assistant coach would be available to participants 24/7 for months or years after a training workshop, allowing it to take over some of the on-the-job training and mentoring traditionally left to busy, overworked managers.
Conclusion
As our company has started rolling out the first wave of customized AI coaches for clients, no one can say what their impact will be or how the technology will evolve over time.
Will AI coaches reduce the volume of e-learning or instructor-led training time involved in training programs? Probably not, however they will make it possible to give every worker their own coach, not just the CEO or the superstar salesperson. And as such it represents an opportunity to have an exponential impact on learning outcomes, with only a marginal increase in training spend.
Hopefully this article provided some new information and clarity around the potential for using AI coaches in learning programs!