Most firms overestimate AI progress, EXL study finds
A new study from data and AI company EXL (NASDAQ: EXLS) has exposed a striking disconnect at the heart of enterprise AI adoption: three quarters of US companies believe they are ahead of their competitors on artificial intelligence, yet only one in ten meets the criteria to be classified an AI Leader. The gap, EXL argues, is not a technology deficit. It is an operating model failure.
The third annual EXL U.S. Enterprise AI Study surveyed 322 C-suite and senior decision-makers across banking, insurance, retail, utilities, life sciences, and healthcare payer industries. Its findings arrive at a moment when boardrooms across every major sector are allocating significant capital to AI transformation programmes, making the reported gulf between perception and performance a strategically significant data point for investors and executives alike.
Leaders rebuild; laggards adapt
The study draws a clear line between organisations that have embedded AI into core workflows and those that have bolted it onto existing processes. AI Leaders, the top 10% of respondents, have moved beyond pilots to redesign enterprise-wide operating models. Forty-four percent of Leaders have completely restructured how their organisations operate to accommodate AI at the core, compared with just 23% of laggards.
The returns from that structural commitment are substantial. Within the specific areas where AI has been deployed, Leaders report an estimated 27% increase in revenue, a 26% reduction in costs, and a 22% improvement in margin. Leaders also report greater resilience in volatile market conditions and more effective customer engagement. Laggards trail on all three financial metrics.
"What separates the leaders is that they've stopped trying to fit AI into the way they already work, and started asking a more fundamental question: if AI were built in from the start, how would this workflow, this team, this decision look different?" said Anand Logani, EXL's chief AI officer. "Moving from AI experimentation to AI execution requires more than technology investment; it requires operating model transformation."
Data readiness remains the most commonly cited barrier to scaling AI, with seven in ten respondents describing it as a challenge. Data privacy and security (34%), siloed data across multiple sources (31%), and limited model transparency (31%) were the three most frequently named obstacles. Among laggards, 83% still contend with data locked inside individual business functions, compared to 44% of Leaders who have achieved enterprise-wide data accessibility.
The convergence read-across
The implications of this study extend well beyond the enterprise software market. For investors tracking the AI infrastructure build-out, the findings suggest that GPU clusters and foundation model licences are necessary but not sufficient conditions for AI-driven returns. The bottleneck is organisational architecture, not compute capacity. That has meaningful consequences for the capital currently flowing into AI services and consulting firms positioned to deliver operating model transformation at scale.
Across sectors, the pattern maps onto a broader structural shift. In financial services, insurers and banks that have centralised their data estates and rebuilt underwriting or credit decisioning workflows around AI are already pulling ahead on efficiency metrics. In life sciences, the same logic applies to clinical operations and regulatory submissions. The EXL data provides rare quantified benchmarks for the magnitude of that divergence.
For cross-sector strategists, the study also sharpens the risk calculus around AI-related equity valuations. Companies currently priced on the assumption of AI-led productivity gains may need to demonstrate operating model transformation, not just tool adoption, to justify those multiples. The 10% figure may prove a useful stress-test for portfolio positioning across any sector where AI efficiency narratives are driving premium valuations.