Microsoft’s Q3 FY2026 earnings call reported over 20 million paid Microsoft 365 Copilot seats against a base of more than 450 million paid M365 commercial seats. That is roughly a 4.4% paid attach rate. Not usage. Not trial. Just the percentage of the installed base that is paying for the product at all.
The tool works. The organizational readiness does not. Employees get access, try it once or twice, cannot figure out where it fits in their actual workflow, and go back to doing things the old way. A Recon Analytics survey of over 150,000 U.S. respondents found a 35.8% workplace conversion rate among paid AI subscribers with workplace access. For organizations buying Copilot licenses at $30 per user per month, that gap between licenses purchased and licenses used adds up fast.
The banks you are about to read about made this same mistake early. They rolled out AI tools and watched usage plateau. But what separates them from the rest of the market is that they diagnosed the problem correctly: this was never a technology deployment. It was an organizational build. And they treated it like one.
What follows is one of the most detailed public case studies available of an enterprise moving from zero to production-grade AI operations. It is a blueprint for how structured progression through distinct stages of capability creates outcomes that random tool deployment never will.
From Blocking ChatGPT to Production AI in Two Years
After ChatGPT launched in late 2022, JPMorgan Chase blocked it across the entire firm to protect data security. Senior leadership was briefed immediately. By March 2023, the bank had signed a Microsoft Azure agreement to bring large language model capabilities inside its own walls. A central intake portal captured over 1,000 use case ideas within weeks. They locked the door, assessed what they were dealing with, and opened it on their own terms.
Through 2023, the bank ran AI Days in Asset & Wealth Management, tied directly to performance reviews, signaling that AI literacy was not optional. Over 100 use cases were built across business units. An internal Shark Tank-style competition generated 500 proposals from early-career analysts, one of whom later took on AI leadership responsibilities across the bank. This was not a pilot program. It was a deliberate investment in organizational capability before scaling anything.
By mid-2024, the cracks appeared. Teams across the firm had built siloed GPT solutions for investments, philanthropy, taxes, and market views. Each worked in isolation. None talked to each other. Advisors were confused by the proliferation. Dozens of successful islands, no connective tissue between them. This is the exact point where most enterprises stall. They celebrate pilot wins, report strong numbers to the board, and never build the orchestration layer that turns isolated tools into enterprise capability.
JPMorgan built it. Coach, a unified platform where a single orchestrator routes queries to more than two dozen specialized agents, launched in October 2024 to 3,000 Private Bank advisors through the Connect digital platform. LLM Suite scaled to 200,000 users, roughly two-thirds of the workforce, through an opt-in strategy. Asset & Wealth Management alone hit 1 million prompts per month. Twelve levels of guardrails governed input safety, output safety, and content filtering. Tens of thousands of queries were reviewed by subject matter experts before the system earned enough trust to operate at scale.
Today, JPMorgan is using generative AI in cyberthreat modeling and says it has a production model intended to run against every code change, more than four million per year, to identify failures and risks earlier. The bank is exploring agentic architecture patterns, including builder/verifier workflows in the software supply chain. In Asset & Wealth Management, it is working to hyper-personalize Coach for advisors by drawing on notes, transaction patterns, and client-specific data. These capabilities only exist because the organization spent two years building every layer underneath them.
Where the Money Sits Tells You Everything
JPMorgan expects to spend about $19.8 billion on technology in 2026, up 10% year over year. The company says it is spending about $1.2 billion more this year on major projects, with AI initiatives among several ongoing investment areas alongside customer experience work and platform build-outs. AI is not a side experiment at JPMorgan. It sits inside the same investment envelope as core infrastructure.
That positioning matters. When AI spending is buried inside a discretionary innovation fund, the organization is treating it as something it is trying. When it moves into the same category as data centers and trading platforms, the organization is treating it as something it runs on. Most enterprises have not made that shift. The leaders made it two years ago.
Apply this test to your own organization. If your AI budget is still managed by one executive champion out of a discretionary fund, you are funding experiments. If it has moved to business unit P&Ls tied to operational outcomes, you are funding capability. Where the money sits is one of the most honest signals of organizational maturity available.
The Gap Is Widening Every Quarter You Wait
JPMorgan did not get here by buying better tools. They got here by building organizational capability in sequence: literacy first, then experimentation, then orchestration, then production-grade operations. Each stage required different leadership, different governance, and different infrastructure. Skipping a stage did not save time. It created debt that slowed everything down later.
GAI Insights maps organizations against a maturity model built from exactly this progression: from research and education through islands of experimentation through enterprise-wide orchestration to emergent intelligence. The organizations that know where they sit on that model make better buying decisions, deploy faster, and waste less money on tools they are not ready to use.
As of Jamie Dimon’s 2023 shareholder letter, JPMorgan had more than 2,000 AI/ML specialists and data scientists. It had the full backing of the Operating Committee and a technology budget approaching $20 billion. It still took two years of structured, disciplined work to get from blocking ChatGPT to running production AI at enterprise scale. If that is what it takes for one of the most technologically advanced banks in the world, where does your organization actually stand? How many quarters has the gap been open? And what is it costing you to leave it that way?
The uncomfortable truth is that most organizations cannot answer those questions. They know what tools they have bought. They do not know what stage they are operating at, what the next stage requires, or which vendors actually serve that transition. That is the gap the Buyer’s Guide is built to close.
The 2026 Corporate Buyer’s Guide to Enterprise Intelligence Applications, publishing in June, evaluates 30-plus vendors across the three buying decisions that define your enterprise AI stack. It is built to help you see where you are, what to buy next, and what to skip.
By GAI Insights Team
-1.png?width=427&height=240&name=Frame%209%20(6)-1.png)
