The gap between companies that successfully transform with AI and those that flop is not complex. It's a gap in having the money, a great rollout strategy, and cultural change management.
There is a recognisable pattern to do this well: they start internally, move deliberately, and treat AI adoption as a people problem. This article breaks down this pattern through the lens of JP Morgan Chase — one of the most studied examples of enterprise AI transformation at scale.
Everyone knows JP Morgan Chase
It's the largest bank in the United States by assets, and employs over 300,000 people globally. Most of my college friends salivate at the thought of working there.
The investment commitment
An $18 billion dollar investment in 2025, with an upfront and honest acknowledgement by CEO Jamie Dimon that AI would "eliminate some careers" while augmenting others. Respectable, honestly. To JP Morgan, AI was their new operating assumption.
Internal-first LLM rollout
Their cornerstone was the in-house development of a proprietary generative AI platform built entirely in-house and released in summer 2024. The firm made it optional for employees to use it over mandating adoption. According to insiders, it created "healthy competition" as teams leaned in to find valuable use cases.
It went from 60,000 users to 140,000, and 200,000 by eight months. That's two-thirds of their workforce. We can see here that:
- Opt-in works better than mandatory adoption
- They used multiple LLM providers — no vendor lock-in
- It was deployed within their own infrastructure
Governance with scale
We need to acknowledge the back-end work taken to figure out risk and compliance considerations too. Data security was the primary constraint, and only once the security posture was established did the firm launch its AI initiative.
Dual-track strategy
AI is about engaging both your executives and your business units. JP Morgan ran a dual-track strategy:
- Top-down. Focused executive attention and input on four domains: credit, fraud, marketing, and operations.
- Bottom-up. Gave individuals innovation space in their individual workflows.
Transformation was not too centralised, or too fragmented. Not easy.
What were the outcomes?
- Client advisory response times improved by 95% during market volatility
- Gross sales in asset and wealth management went up 20%
- Net income of $14.6 billion in Q1 2025, up 9% year over year
- 450+ AI use cases in active deployment across front, middle, and back offices
The playbook principles
Principle 1: Start internally
Deploy AI internally with your employees before incorporating anything customer-facing. Internal adoption creates internal champions — your most credible advocates.
Principle 2: Make adoption voluntary, then make it viral
If you force people to adopt AI, people only comply because "my boss told me so." Opt-in models produce faster and deeper adoption. Companies can use structured onsite workshops for employees to build actual workflows.
Principle 3: Invest in change management, don't forget about it
JP Morgan ran more than 30,000 AI education sessions, and their investment in people was proportional to the investment in technology. That's what made the AI investments sticky.
Principle 4: Governance precedes scale
Organisations that build governance frameworks first beat organisations that merely respond to security incidents. Don't play catch-up.
Principle 5: Measure what matters
JP Morgan set up test and control groups to measure benefits from AI tools. Without baselines and defined success metrics, there's no signal about what to iterate on.
Principle 6 (most important): Rearchitect workflows
Workflow re-architecture with AI and agents produces order-of-magnitude productivity changes. Surface-level, chat-based AI adoption produces only marginal gains.
The diagnosis-first imperative: it's actually not about the budget
Across every case study we've examined, and within our own customers, the diagnosis of your workflows becomes the differentiator.
Map before you deploy
Find where AI creates genuine leverage — not from "what tools are available?" The right question is: "Which workflows, redesigned around human-AI collaboration, would produce the greatest change in output per person?"
Find your asymmetric use cases
Every organisation we've worked with has had this problem. A very small number of workflows end up accounting for a disproportionate share of time and workload cost. These are the workflows where AI deployment produces asymmetric returns beyond productivity gains.
Cost optimisation is an architecture problem
Cost discipline has been a recent issue, even in our own projects. Organisations are spending too much on tokens, and AI usage drives integration costs up. Organisations need solutions to route simple, repetitive tasks to cheaper models, while reserving expensive compute for complex reasoning. The right tech architecture matters.
Trust and speed during deployment have to be balanced. Organisations that scale AI fastest are not the ones that move quickest — they are the ones where their people actually believe in AI transformation.
Where we come in
We understand AI transformation is daunting. But this is where we know we can help: human-focused and diagnosis-first. We run the diagnosis, plan your tech architecture, and design your AI adoption journey for cost efficiency from day one. Governance and change management are what we're used to working with across our healthcare, finance, retail, and government customers. The investment in AI stays proportional to the returns, not to the headline number.
Sources
- JP Morgan Chase AI Strategy (AI News, 2025): artificialintelligence-news.com
- LLM Suite Drives AI Transformation (The Digital Banker): thedigitalbanker.com
- 450 Use Cases and Lessons Learned (Tearsheet): tearsheet.co
- Building an AI-First Bank Culture (McKinsey): mckinsey.com
