The headline figure from JPMorgan Chase's Contract Intelligence (COiN) platform — 360,000 legal hours eliminated annually across roughly 12,000 commercial credit agreements — has become one of the most cited public benchmarks for AI in regulated financial services. The deeper lesson is what produced the result: not a better model, but a workflow rebuilt around the model, with governance, audit, and human-in-the-loop checkpoints engineered in from day one.
The benchmark: 360,000 hours of work, compressed to seconds
JPMorgan Chase's Contract Intelligence (COiN) platform, launched by the bank's Intelligent Solutions team in 2017, used machine learning to automate the review of commercial credit agreements. According to JPMorgan's own technology team, the system extracted approximately 150 distinct attributes from roughly 12,000 contracts each year — work that previously consumed an estimated 360,000 hours of skilled legal and loan-officer time annually1,2.
What the data tells us about agentic AI in regulated industries
The COiN engagement is one of the most cited public benchmarks for AI deployment in regulated financial services. Three points stand out for any leader weighing a similar investment.
The bottleneck is rarely the model. JPMorgan's reported gains came from rebuilding a workflow — document ingestion, classification, attribute extraction, validation, audit trail — around the model. Bloomberg, reporting on the launch, noted the bank had also reduced loan-servicing mistakes that previously stemmed from human error in interpreting wholesale contracts2.
The cost case stacks up. Reuters and others have reported that JPMorgan's annual technology budget at the time exceeded $9 billion, with COiN cited as one of the highest-leverage automation programmes in that portfolio3. The published return profile — eliminating six-figure annual labour hours on a single contract type — is consistent with the broader evidence base: McKinsey's State of AI 2025 survey found that software-engineering and back-office functions deploying AI report 10–20% cost reductions tied to automation of structured tasks4.
Adoption is no longer the differentiator. Workflow redesign is. McKinsey's 2025 data shows 88% of organisations now use AI in at least one business function — but only 6% report meaningful EBIT impact. The dividing line between leaders and laggards is end-to-end workflow rebuild, not access to better models4.
What this means for a mid-market lender or insurer
Most institutions outside the top tier of global banking will not deploy at JPMorgan's scale. But the underlying pattern — a high-volume, repetitive, high-stakes document workflow handled by skilled but expensive human reviewers — exists across mid-market banking, insurance underwriting, claims adjudication, KYC/AML review, and procurement. The economics of agentic AI in these functions are now well understood.
The question for boards is no longer whether the technology works. It is whether the operating model — data access, audit and governance controls, change management, and human-in-the-loop checkpoints — is ready to absorb the speed gain without introducing new risk.
"The 6% of organisations realising meaningful EBIT impact from AI did not get there by deploying better models. They got there by rebuilding their workflows."
Where NovasIQ comes in
NovasIQ's AI Transformation pillar is built ground-up for this kind of deployment — not retrofitted onto a legacy consulting service line. Our work spans the full agentic stack: production-ready AI agents, MCP-based tool integration, retrieval-augmented generation pipelines, GenAI integration into legacy systems, and bespoke AI modelling for regulated environments. We pair that with the operating-model discipline — governance frameworks, audit trails, human-in-the-loop design — that determines whether a deployment delivers the kind of impact JPMorgan has reported, or joins the 40%+ of agentic AI projects Gartner forecasts will be cancelled by end of 2027 due to escalating costs and unclear value5.
The capabilities that map to outcomes like these
- AI Agents & MCP Solutions — for autonomous multi-step document and decision workflows
- AI Strategy & Roadmap — for portfolio prioritisation by EBIT impact rather than novelty
- Legacy modernisation — to make the data and process estate AI-ready
- Cybersecurity — for the governance, audit, and policy guardrails regulated industries require