Q1. Could you start by giving us a brief overview of your professional background, particularly focusing on your expertise in the industry?

I have over 10 years of industry experience, starting in content moderation and progressing to data labeling and AI/ML operations. I’ve worked with both small and large teams—often managing 1,000+ people—focusing on quality, training, and stabilizing products at scale.

Over time, I’ve also worked with AI tools, LLMs, and data-driven systems to improve overall performance. Currently, I’m a Service Delivery Manager, handling operations across North America and Europe for an AI-focused company.

Q2. Across the industry, how quickly is automated AI moderation 'cannibalizing' the headcount of traditional BPO contracts, and is the new 'AI-Training' work filling that revenue gap 1:1?

AI is reducing headcount in traditional BPO roles, especially in areas like customer support, data entry, and basic problem-solving. Much of the first-level work is now handled by automation.

However, it’s not a complete replacement—roles are evolving. You need fewer people, but with higher skills. And while AI-training work is growing, it’s not yet replacing the lost volume 1:1. It’s more specialized and harder to scale.

Q3. In which specific IT service workflows do you believe human delivery will remain the 'Gold Standard' for the next decade, despite the rapid advancement of LLMs?

Even with rapid progress in AI, some areas will remain human-led for the next decade. This includes things like complex decision-making, cybersecurity, client interactions, and high-stakes QA.

You also can’t replace humans in compliance, creative problem-solving, and AI oversight. Anywhere there’s risk, ambiguity, or accountability, humans will continue to be the gold standard, with AI mainly acting as a support layer.

Q4. What percentage of the AI safety workflow is currently successfully 'automated,' and where do you see the 'hard ceiling' that requires human-in-the-loop (HITL) indefinitely?

I’d say roughly 60–70% of AI safety work is automated today—things like filtering, classification, and obvious violations. But the remaining 30–40% still need humans, especially for context, edge cases, and judgment calls. That’s where AI still struggles, and honestly, that gap isn’t going away anytime soon.

Q5. Are you observing a shift towards synthetic data for safety training to save costs, and what measurable 'safety decay' do you observe when the human-in-the-loop is removed?

Yes, there’s definitely a shift toward synthetic data to scale faster and cut costs. But when you reduce human input, you start to see a 10–25% drop in safety quality—mostly in edge cases and real-world unpredictability. The best setups right now are hybrid, with humans still validating and refining what AI generates.

Q6. What are the emerging best practices for 'Data Provenance'—ensuring that the training data hasn't been tampered with—that you believe will become a 'must-have' for enterprise AI in the next 12 months?

Data provenance is becoming a must-have. Companies are focusing on traceability, audit trails, and data integrity checks to make sure nothing is tampered with. Strong access control and compliance are also key. Without this, it will be very hard to build trust, especially with enterprise clients.

Q7. If you were an investor looking at companies within the space, what critical question would you pose to their senior management?

As an investor, I’d ask: “What’s your real moat?”

With AI becoming more accessible, the differentiator is how well a company can balance automation with human quality, scale responsibly, and maintain trust. That’s what will actually last.