.png)
A salesperson at a Fortune 50 company told Sumanta Chakraborty he had used his employer's AI training tool exactly once, long enough to clear the required score, and would never open it again. That is what you buy when you measure completion instead of learning, and most training budgets are buying exactly that.
Who is on this episode
Sumanta Chakraborty is the founder and CEO of nSpire AI, a talent development platform he describes as a flight simulator for critical conversations. He spent close to 30 years building technology at NVIDIA, Nuance, Siemens, Sony and LG, across AI, robotics and autonomous systems, and holds an MBA from UCLA. He has operated in Asia, the United States and Europe, and the one thing every one of those very different cultures had in common, he says, was how little real thought went into how people get hired and how they get grown.
What they covered
The conversation starts with the two moments where companies lose the most value and invest the least. The first is the interview: nobody trains interviewers to interview, so a hiring decision rests on whoever happened to be on the panel that week. The second is the promotion from individual contributor to manager, a completely different job that most companies hand over with a document library and no practice. In between sits onboarding, which Sumanta calls static content handed to someone who is facing new products, new customers, new buyers and a new culture all at once.
From there Imran and Sumanta get into the annual performance review, which Sumanta says is not a process so much as a scramble to remember what you did eleven months ago because Workday has assigned you a task. His argument is that coaching has to happen in the moment, and that AI finally makes in the moment possible at scale. The sales example is the sharpest version: you can let a new rep practise objections and discovery calls against a simulated buyer, or you can let them practise on your actual pipeline and find out six months later that they never understood why you beat the competition.
"We spend all that money and what we measure is completion, not learning. How many people completed the courses? What were their scores?"
Sumanta Chakraborty
They also take on the line between coaching and surveillance. Sumanta's position is that the practice data belongs to the individual and the organisation should only see trends, because a coach only works on someone who is willing to be vulnerable, and nobody is vulnerable in a system they think is watching them. He closes with what an engineer notices about HR from the outside: engineering ships, observes, takes feedback and improves in a loop, and HR runs the most important asset in the company without one.
Key insights
- The practice gap sits at the transitions. Nobody is trained to run an interview, and nobody is trained to run a one on one, a raise conversation or a performance improvement plan before they are expected to do all three.
- Completion rates are a vanity metric in the age of chatbots. A learner can paste the assessment into a model and score full marks, so a score tells you nothing about capability.
- Mastery comes from reps, not reading. Sumanta's framing of the 10,000 hour idea is that it was never 10,000 hours of reading, it was 10,000 hours of practising.
- Coaching only works if the employee can be honest, so split the data: the individual keeps the session, the organisation gets aggregate trends like "my managers are weak on active listening" and can act on that.
- The numbers that changed Sumanta's own mind about whether people would practise with AI at all: roughly 15,000 users, over 100,000 interactions with the system, and 30,000 hours of recorded practice conversations.
Watch the full conversation on YouTube, or listen to the audio version on Apple Podcasts.
