Playbook 01 · 90 days
Your first 90 days: one group, one rule, one use case.
You don’t need an AI strategy document. You need a few people who can say yes or no, a rule for the tools staff are already using, and one problem worth solving. This route gets you there in a quarter.
01 · Before you start
What it takes, and what you get.
Nothing here needs new staff or new software. It needs an owner, a standing meeting, and honest answers from each department.
- Who leads
- The CEO names one owner. A COO, CNO, quality director, or IT lead can do it. The title matters less than having the time and the standing to get answers from every department.
- Time
- 90 days. A 45-minute meeting every two weeks, plus the owner’s time between meetings.
- What you need
- A list of your software vendors and contracts. Your current policies on privacy, security, and acceptable use of technology. One person from the clinical side and one from the business office who will show up.
- What you’ll have at the end
- A governance group that meets on a schedule. An inventory of the AI already in the building. A one-page acceptable-use rule. One chosen use case with a baseline and a measure of success. A first report to the board.
Some of these decisions may already be made for you. Ask your system’s IT and compliance leads what exists before you write anything. Your own group still has to decide how those rules apply in your building.
02 · The route
Six mile markers, in order.
The dates overlap on purpose. The inventory and the rule can move at the same time. The use case waits until both are in hand.
-
Mile1Days 1–10
Name the owner and the group
Keep the group small enough to meet. Five or six people is plenty: the owner, a physician or advanced practice provider, a nurse leader, whoever runs IT, the compliance or privacy officer, and someone from the business office. In a small hospital, one person may hold two of those seats. That’s fine.
- The CEO has named one owner, in writing
- The group has a clinical voice, an IT voice, a compliance voice, and a revenue cycle voice
- The group knows what it decides: which AI tools may be used, for what, and with which data
- A standing meeting is on the calendar every two weeks for the quarter
- The group reports to an existing committee or to the CEO, not to a new structure
Where small hospitals get stuckWaiting to write a charter before the first meeting. Meet first. Write down what the group decides as it decides it. By day 90, those notes are the charter.
-
Mile2Days 10–30
Find the AI you already have
Most hospitals are already using AI and don’t have a list of it. It arrives inside the EHR, the dictation software, the coding and billing tools, the phone system, the email suite, and on staff members’ own phones. Ask every department head two questions. What do you use that calls itself AI? What are your staff using on their own?
- Every software vendor has been asked which features use AI, and whether they are switched on
- Each tool on the list has an owner, a purpose, and a note on what data it touches
- Tools that touch patient information are marked, with a note on whether a business associate agreement covers them
- Staff have been asked, without blame, which general chatbots they use for work
- The list lives in one place the whole group can see
The data readiness guide covers where your data lives and who can get it out. You’ll need both answers at mile 4.
-
Mile3Days 20–40
Write the acceptable-use rule
Staff are using general chatbots today. A ban won’t stop it, and silence leaves them guessing. Write one page that says what’s allowed, what isn’t, and who to ask. The line that matters most: no patient information goes into any tool the hospital doesn’t have a business associate agreement with.
- One page, in plain language, that a new hire could follow
- No patient information in any AI tool without a business associate agreement
- A short list of approved tools, and what each one is approved for
- A person checks the output before it’s used. The tool drafts. People decide
- A named person staff can ask before they try something new
- Every employee has seen it, and it’s part of onboarding
Where small hospitals get stuckCopying a large health system’s policy. It was written for an organization with a legal department and a review board. Nobody reads it, so nobody follows it. One page that people follow is worth more.
Before you finalize the rule, check the regulatory map for anything your state requires you to tell patients.
-
Mile4Days 30–55
Choose one use case
List the problems, not the products. Ask each department what takes the most time, causes the most rework, or keeps people late. Then score each problem on two things: how much it hurts, and how feasible a fix is with the people, data, and budget you have. Pick one that scores high on both. Leave the rest on the list.
- At least five candidate problems, each stated without naming a product
- Each one scored for pain: who feels it, how often, and what it costs in time or money
- Each one scored for feasibility: do you have the data, an owner, and a vendor who has done this at a hospital your size
- The first choice doesn’t make a clinical decision on its own
- One use case chosen. The others are parked, with a date to look at them again
Where small hospitals get stuckStarting from a demo. A vendor shows something impressive, and the hospital goes looking for a problem it might solve. Start from the problem. If no department would list it among its worst, it isn’t your first use case.
-
Mile5Days 50–75
Ask the vendor hard questions, and set a baseline
Before you sign or switch anything on, get answers in writing. Then measure where you stand today, so you have something to compare against. Without a baseline, every result is an opinion.
- The vendor will sign a business associate agreement
- You know where your data goes, where it is stored, and who can see it
- The contract says whether your data is used to train the vendor’s models, and you’ve decided whether that is acceptable
- The vendor has shown how the tool connects to your EHR, in your version, not in a demo
- You’ve spoken to a reference at a hospital of similar size
- You know how to leave: the notice period, what happens to your data, and what it costs
- A baseline is recorded for the one or two numbers this use case should move
-
Mile6Days 75–90
Decide how you’ll know, and tell the board
Write down, before any pilot starts, what success looks like and what would make you stop. Then tell the board what you’ve done and what comes next. A board that hears about AI from you first asks better questions later.
- One or two measures of success, each with its baseline next to it
- A stop rule: the result or the problem that ends the pilot
- A measure of use, not just purchase: who is using it, and how often
- A date for the decision to keep, change, or stop
- A short report to the board: the group, the inventory, the rule, and the use case
- A plan for what patients will be told, if the use case touches them
The board briefing is a twenty-minute agenda for that report.
03 · After day 90
Now you’re ready to test one thing.
You haven’t changed how the hospital runs yet, and that’s the point. You’ve built the small amount of structure that keeps the next step from going wrong.
The next route picks up where this one ends: a short, honest test of the use case you chose.
What done looks likeSix things you can hold up
- A governance group that meets on a schedule and keeps notes.
- An inventory of every AI tool in the building, with an owner for each.
- A one-page acceptable-use rule that every employee has seen.
- One use case, chosen on pain and feasibility, with a baseline.
- A success measure, a stop rule, and a decision date, written down before the pilot.
- A board that heard it from you first.
Stuck at a mile marker?
Tell us which step and what’s in the way. If other hospitals are stuck in the same place, the answer becomes part of this playbook.