EXL is betting that enterprise AI’s next competitive gap isn’t model selection but model operationalization, and it’s acquiring iMerit to own that layer. The deal, now closed, combines EXL’s domain expertise across insurance, healthcare, and financial services with iMerit’s specialized capabilities in AI model training, evaluation, and reinforcement learning from human feedback (RLHF, the process of using expert judgment to improve how a model behaves in practice). iMerit founder Radha Ramaswami Basu joins EXL’s executive committee. No deal price was disclosed. Read the full EXL-iMerit acquisition announcement<\/a> for context.<\/p>\n If your organization is still treating AI model quality as something foundation model providers handle upstream, this deal is a signal that assumption is eroding fast. EXL is a 68,000-person services firm with deep roots in regulated industries, and it didn’t buy iMerit to resell AI software. It bought a human-expert annotation and evaluation operation, including physicians, scientists, and linguists, because enterprise AI deployments in healthcare and insurance fail not at the model level but at the domain-accuracy level. CIOs running AI programs in those sectors are now competing against clients of a combined EXL that has structured data curation and model validation baked into the service wrapper.<\/p>\n The more important structural claim here is that iMerit’s Ango Hub platform, a collaborative tool for building and validating multimodal training datasets, gives EXL a proprietary data flywheel that pure-play IT services firms can’t easily replicate. Most systems integrators are assembling AI practices by partnering with model providers and layering consulting on top. EXL is assembling the feedback loop itself, the part that decides whether a deployed model stays accurate six months after go-live. That’s a different business model, and it creates switching costs that consulting relationships alone don’t.<\/p>\n The vendor landscape this reshapes most directly is the middle tier of AI services firms positioning for enterprise model deployment work. A renewal or expansion conversation with a services partner that promises to “manage your AI” now needs a harder question attached: who owns your model evaluation process, and where does domain-specific correction actually happen? If the answer is “the model provider handles it” or “we review outputs periodically,” that’s a capability gap this deal just made visible. The firms that own the human-in-the-loop layer will set quality benchmarks the rest have to chase.<\/p>\nWhat this means for your business<\/h2>\n