As enterprises accelerate their AI adoption, the business process management (BPM) industry is under pressure to reinvent itself. While fears of disruption and job displacement persist, Genpact believes the real opportunity lies in shifting from a labour-led outsourcing model to one centred on business outcomes.Vijay Vijayasankar, Global Agentic AI Officer, Genpact, discusses why most AI projects fail to move beyond pilots, the enterprise debt holding organisations back, and how AI will reshape jobs rather than eliminate them.Has the BPM industry been disproportionately affected by AI?Today, there is intra-company competition among BPOs. Whoever uses AI better will have an edge. This criticism is largely from investors and analysts who say that software companies or model providers could disrupt you. The BPM industry, or at least Genpact, has a different view. We don’t sell labour, but an outcome. Neither software companies nor LLMs commit to an outcome.LLM providers charge based on token usage, while software companies typically charge per user. Neither guarantees a business outcome, such as reducing invoice processing costs from $5 to $2. However, we can underwrite an outcome. My thesis is that we will not be disintermediated because we are changing our business model to stand behind outcomes and not continue with a labor-based model. Meanwhile, the BPO companies sticking to a labor-based model will be disintermediated by technology.Which BPM functions are being transformed the fastest by AI, and which will continue to require human judgment?Everything will need a human touch. The AI we use is based on deep learning; it’s a probability-driven system and not deterministic. For instance, an LLM doesn’t calculate 2 + 2 like a computer or a calculator. It looks through data to see past responses.Hallucination is not a bug, but a feature. Many guardrails can be programmed, but it also needs human guardrails. Autonomous should not mean ungoverned. Many assume that AI will take over the world and do everything autonomously. This is a utopian vision.However, AI can take over 80% of repetitive, low-judgment activities, while 20% will need some human judgment. AI will take over the vast majority of coding, but not software engineering.AI is inherently dumb about architecture, whereas good software is all about great architecture. Human judgment is incredibly important, and the mass narrative misses these nuances.