EXL has closed its acquisition of iMerit, betting that the next competitive moat in enterprise AI isn’t model selection but the infrastructure to make models work reliably in production. iMerit brings a credentialed human-in-the-loop workforce called Scholars, a proprietary annotation and evaluation platform called Ango Hub, and direct relationships with frontier AI labs. Founder Radha Ramaswami Basu joins EXL’s executive committee. The EXL-iMerit combination<\/a> targets regulated industries where generic model outputs aren’t enough and domain-specific tuning, evaluation, and compliance are table stakes.<\/p>\n The AI services market is bifurcating fast. On one side sit generalist integrators promising speed to deployment. On the other sit specialists who can own the harder problem: getting AI to produce trustworthy outputs in environments where a wrong answer has legal, clinical, or financial consequences. EXL is making a clear move into the second camp, and every enterprise that has stalled at the pilot-to-production gap, particularly in insurance, healthcare, or banking, needs to honestly assess whether its current AI partners have the domain depth to get it across.<\/p>\n What EXL is acquiring isn’t just a data labeling shop. iMerit’s Scholars network, which includes physicians, scientists, and engineers doing model evaluation and reinforcement learning from human feedback (RLHF, the technique that teaches a model to align with expert judgment rather than just statistical patterns), gives EXL a human quality layer that pure-software platforms can’t replicate at scale. That matters because regulated industries face a specific problem: off-the-shelf model benchmarks don’t map to their compliance requirements, and building custom evaluation pipelines in-house is expensive and slow. EXL is positioning to own that gap as an outsourced service.<\/p>\n The acquisition also signals something about where the money is moving in AI services. The firms that built their pitch around implementation and change management are getting squeezed as deployment tooling commoditizes. The next margin pool sits in model fine-tuning, continuous evaluation, and production monitoring, the work that keeps AI accurate after go-live, not just at launch. If EXL executes, the comparison set for a renewal decision shifts away from traditional BPO providers and toward specialized AI labs. Any CIO or CEO currently renewing a services contract with a vendor who hasn’t made a comparable capability bet should weigh that gap explicitly before signing.<\/p>\n RLHF is the training technique where human experts rank or correct a model’s outputs, and those judgments are fed back to reshape how the model responds, similar to an apprentice who improves not by reading more textbooks but by getting corrections from a senior practitioner in real situations. In enterprise contexts, “human” can’t mean a generic crowd. It means domain experts whose corrections actually reflect what accurate looks like in that field. That’s precisely the asset iMerit has built and EXL just bought.<\/p>\nWhat this means for your business<\/h2>\n
Concept deep-dive: Reinforcement Learning from Human Feedback<\/h2>\n