Mark Birkhead is firmwide chief data officer at JPMorgan Chase. In a conversation published by McKinsey & Company on 5 June 2025, he spoke with Kevin Buehler, a senior partner in McKinsey’s New York office, about how the bank manages its data now that AI sits at the centre of its plans. McKinsey’s introduction puts the bank at about $4 trillion in assets and more than 300,000 employees, and notes that it began a new data transformation in 2023. The discussion ranged across the size of the bank’s data estate, how responsibility for data is split, what AI-ready data means, the tests AI must pass before production, and his advice to new data chiefs.
An estate measured in exabytes
Birkhead described a footprint spanning almost 100 countries and businesses from Chase consumer banking to investment banking, asset and wealth management, and payments. He put the firm’s data at more than an exabyte, held in structured and unstructured forms that include voice and video, and called it a major strategic asset.
He tied data strategy directly to serving customers and clients, and said the push to use data, analytics and AI comes from chief executive Jamie Dimon, so the mandate starts at the top. The aim, as he set it out, is data that people can find, that is highly accurate and that is tightly governed. Privacy runs through all of it, because customers rely on the bank to protect their information.
Who owns the data
In mid-2023 the bank created a firmwide chief data and analytics office to bring every data initiative under one roof. Teresa Heitsenrether, the chief data and analytics officer, reports to Dimon and sits on the operating committee. Birkhead’s central team sets the data strategy for the whole firm. Business lines and corporate functions put it into practice, since they sit closest to the data and should own it. The centre supplies the privacy and governance strategies, along with shared tooling and capabilities.
On risk, he said the firm starts by identifying the data risks specific to it, then monitors how data is kept accurate, stored, protected and eventually destroyed. It also checks that data is used only for legitimate purposes that match the privacy commitments made to customers and clients.
Four priorities, one of them first
Birkhead was clear about the order.
“Our top priority is to keep our customers’ and clients’ data safe”
Beyond protection, he listed three aims. The first is AI-ready data. Teams are studying what AI and generative AI mean for data strategy, with a multiyear goal of taking people out of routine work fixing data. Part of that effort is publishing data in a consistent form that large language models can interpret, done jointly with technology partners and data teams.
Data products come second. Several have already worked well, and the next step is to make them interchangeable, interoperable and reusable. Speed is the third aim: he wants data available within milliseconds.
The bar for production AI
Any use case heading into production, he said, must draw the right data, in the right format, from the right places. Before deployment it must also meet standards for accuracy, explainability and proper permissions.
He described where the investment is going. Governance platforms built up over the years are being consolidated into one central platform. LLM Suite gives employees access to large language models within guardrails, and he cited 200,000 users. Other spending goes on engineering-heavy data platforms, including ways to deploy AI agents.
Birkhead sounded excited about the moment. Compute capacity, investment and capability have finally lined up, he said, and he singled out AI agents and systems that can reason as the developments he finds most promising. His caveat was that no new technology can be deployed without continued investment in data, so the pairing of data and AI will matter for the firm and for its customers.
Advice for first-time data chiefs
He noted that the average tenure of a chief data officer is little more than two years, a figure that has proved stubborn, and said he had exceeded it in several roles. He put that down to having a solid strategy and being able to explain what needs doing and what investment it requires. His method is to study the organisation’s challenges, its business goals, its clients’ needs and the data in your care, then build the data and analytics strategy from that analysis.
What it means for leaders
- For data leaders, the lesson is that AI readiness is largely a publishing job: data has to be consistent enough for a model to read without a person cleaning it first.
- JPMorgan’s split, with the centre setting standards while business lines own and run their data, offers a workable answer to the long argument over whether data should be managed centrally or by each business.
- Writing down a production gate covering accuracy, explainability and permissions costs little and settles disputes before they start.
- LLM Suite’s 200,000 users suggest that staff will take up language models in large numbers when the firm offers them with guardrails already built in.
- New chief data officers who want to outlast the two-year average should arrive with a written strategy and a costed investment case, and lead with both.
