What I would do in my first 90 days as a health system chief data officer
If I started as chief data officer at a health system tomorrow, I wouldn’t spend the first 90 days writing a strategy deck. I’d spend them learning where the money is and getting one real win on the board.
The timing would matter. Kaufman Hall’s October report found about 43% of hospitals operating at a loss. Medicaid work requirements start Jan. 1, 2027 in 44 states. And AI is showing up in more of the systems a health system already runs. A new data chief walks into all of that at once.
Here’s roughly how I’d split the time.
Days 1 to 30: listen to the people who own the money
I’d meet the CFO, the COO, and the leaders of revenue cycle, supply chain, and clinical operations. Not to present. To ask.
I’d ask each of them the same questions:
- Where do you think we’re losing money or time right now?
- Which number do you not trust?
- Which of our patients could lose coverage next year, and what would that cost us?
- What have you asked the data team for that you never got?
- If I could fix one thing for you this year, what would it be?
I’d write the answers down and look for overlap. The same few problems tend to come up across leaders, sometimes in different words.
I’d also spend real time with the data team. Less on their roadmap, more on their actual week. What they work on, what’s waiting in the queue, what breaks, and what they’d stop doing if they could. Most data teams know exactly where the problems are. Few have been asked.
And I’d gather the basic facts. What the data platform costs. Which reports exist and how often they’re opened. How many requests are open and how old they are. That gives me a starting point I can measure against later.
I’d add one more item: a list of every AI tool already in use. Many will be there before you are. An AJMC study found that 62.6% of US hospitals on Epic had adopted ambient AI documentation by June 2025. Vendors have added AI features to other systems too, often without anyone keeping track.
By day 30, I’d want a short list: the few problems that matter most to the business, each with an owner and a rough dollar size.
Days 31 to 60: pick one win and settle a few numbers
Two things would happen at once.
First, I’d pick one problem from that list to solve fast. It should be visible to executives, owned by a business leader, measurable in dollars, and doable in a couple of months. A single repeat denial cause, pricing errors on invoices, or overtime in one department are typical candidates. I’d put strong people on it and clear their other work.
Second, I’d pick the handful of numbers leadership argues about most. Often it’s measures like length of stay, cost per case, or denial rate, where finance and operations each keep their own version. I’d get the owners in a room, agree on one definition for each, and make that version the official one. This is slow, unglamorous work. It’s also where a new data leader earns the most trust, because it ends arguments that may have been running for years.
I’d hold off on reorganizing the team. It’s tempting to restructure early and show you’re in charge. I’d rather understand how the work actually flows first, and let the early projects show me what’s missing.
Days 61 to 90: show the result and set a few rules
By now the first project should have a result. I’d report it in dollars, with the business owner presenting it alongside me and finance agreeing to the number. Even a modest win, shown honestly, changes how leadership sees the data function.
Then I’d set a few simple rules for how the team works from here:
- Every major project has a business owner and a dollar goal before it starts.
- New dashboards need a stated decision and a named user, and unused ones get retired each quarter.
- Pilots have a control group and a decision agreed on up front.
- Results get reported the same way every month.
- Every AI tool has an owner, and someone checks how well it’s working.
That last rule carries more weight now. In October, HHS’s final HTI-5 rule dropped the AI “model card” and risk-management requirements for predictive tools in certified health IT. Knowing how a vendor’s model works is now your job. The Joint Commission’s voluntary AI certification is a fair guide to what good oversight looks like.
Last, I’d write a one-page plan for the next year. I’d keep it short: a ranked list of the next problems to solve, each with an owner and a dollar size, built from what I heard in the first 30 days. The technology choices would follow from that list, rather than lead it.
What I wouldn’t do in these 90 days matters as much. I wouldn’t announce a platform replacement or promise an AI program. Both may make sense later. Done first, they spend a new leader’s credibility before there’s any to spend. Taking stock of the AI already running is different. That’s housekeeping, and it belongs in the first 90 days.
The first 90 days are about earning the right to do bigger things. Leaders will judge a new data chief on one simple question: did anything get better? By day 90, you want to be able to answer yes, with a number finance agrees with.
If you’re hiring for this role, ask your finalists to walk you through their first 90 days. Listen for whether they start with your business problems or with technology. Hire the one who starts with the problems.
Discussion