AI for healthcare operations.

Episode one, your first eureka moment. Welcome to AI for Healthcare Operations. I'm Eureka, and over the next few episodes I'm going to change the way you think about AI. Not the way you use it. The way you think about it. Because that's where everything starts.

Let me ask you something. The first time you tried one of these AI tools, what did you do? If you're like most people, you typed in a question. Got an answer. Maybe it was okay, maybe it was junk, and you thought, eh. Not for me.

Here's the problem. You treated it like Google. You walked up to the most capable thinking partner you have ever had access to, and you asked it for directions. That's the mistake. And once you see it, you can't unsee it.

AI is not a search engine. It's a thought partner. A search engine gives you links. A thought partner thinks with you. But here's the catch. It can only think as well as you set it up to.

So let me give you the one framework that changes everything. Role, Context, Output.

Three things, every time. Role. Tell it who to be. You're an experienced practice manager. You're a billing specialist. You're a facilities supervisor. Context. Tell it what it needs to know. The situation, the background, the constraints. Output. Tell it what you want back. A summary? A draft email? A list of options?

Watch the difference. Write me a memo gets you generic mush. But this. You're a clinic operations manager. I need to tell my front desk team we're changing our check-in process next month, and some of them have done it the old way for fifteen years. Write me a short, warm memo that respects that.

Now you're getting something you can actually use. Same tool. Completely different result. The difference was you.

Now here's a trick for when you don't even know what to tell it. Just say, ask me five questions before you answer. Let it interview you. Most people try to write the perfect prompt. Stop. Let the AI pull it out of you. It is better at knowing what it needs than you are.

And one more, because this is the one that turns AI from an agreeable assistant into something valuable. Add this. Challenge my thinking. By default these tools agree with you. They're people pleasers. But when you say push back on this, tell me where I'm wrong, now you've got an advisor instead of a cheerleader.

Last thing. Stop typing. Use the microphone. Talk to it the way you'd talk to a sharp colleague in the passenger seat. Same rules apply. Role, context, output. You're just saying it out loud. For those of us who think better out loud anyway, this changes everything.

So here's your challenge. One real thing this week. Something on your plate right now. Open up the tool and instead of asking a question, give it a role, give it context, tell it what you want. Then say, ask me five questions first.

That's your first eureka moment. The shift from asking, to thinking together.

Everything we do from here builds on it. I'll see you in the next one.

Episode two, what you're actually allowed to do. Welcome back.

Last time we talked about AI as a thought partner. Today I want to deal with the thing that's probably sitting in the back of your mind.

Am I even supposed to be using this?

It's a fair question, and almost nobody answers it out loud. So people do one of two things. Either they avoid AI entirely because they're afraid of getting in trouble. Or they use whatever they found on their phone without ever checking. Both of those are bad, and they come from the same place. Nobody told you where the lines are.

So let's fix that. Three things to know.

First. Your organization almost certainly has approved tools. Somebody in technology went through a review process, looked at where the data goes, checked the privacy terms, and put a short list in front of you. That list is not a restriction. It's a green light. Inside that list, you're covered.

And here's the part people miss. An approved tool and the free version of the same tool are not the same thing. The approved one usually has an agreement behind it saying your information doesn't get used to train somebody else's model. The free one you signed up for with your personal email does not. Same name on the box. Completely different terms.

Second. There is a short list of things that never go into a chat box. Patient names and anything from a chart. Dates, room numbers, or details that could point to one patient. Employee records and discipline files. Anything you'd hesitate to read out loud in a public meeting. If you wouldn't email it to a stranger, don't paste it into an AI tool.

And here's the good news, because this trips people up. Most of your work does not involve any of that. Drafting a memo about a schedule change. Comparing two vendor quotes. Turning your notes into a clean procedure. Rewriting a message so it lands better. None of that touches protected information at all.

Third, and this is the one that matters most. You are responsible for what you send. Not the tool. If AI drafts something and you forward it with your name on it, that's your work now. Read it. Check it. Fix what's wrong. The tool is an assistant, not a signature.

Think of it exactly like a capable new employee. You'd let them draft something. You would not let them send it to the CEO without reading it first. Same standard.

Now, if you don't know what your organization allows, that's not a dead end. That's one email. Ask your supervisor or your technology department two questions. Which AI tools are approved for staff use, and is there written guidance I should read. That's it. People ask that question every day and nobody gets in trouble for asking it.

And if the answer is that a policy is still being written, you are not behind. Plenty of organizations are still writing theirs.

So here's your challenge. Find out. Two questions, one email, this week. Or check wherever your organization keeps its approved software list.

Because here's your eureka moment. The rules are not the thing standing between you and using this well. Not knowing the rules is. And that takes about five minutes to fix.

Next time, I'll show you why chasing new AI tools is usually the wrong move, and what to do instead. See you there.

Welcome back to episode three. Go deep, not wide.

Today I want to save you some time, some money, and a lot of frustration with one idea. Go deep, not wide.

Here's what happens to a lot of people. They get excited about AI and they start collecting tools. They see one on social media that writes emails. Another that makes presentations. Another that summarizes documents. Pretty soon they've got eight different apps, they're paying for half of them, and they're not actually good at any of them.

Let me let you in on a secret the AI world does not advertise. Most of those flashy tools are not new AI. They're a wrapper. A nice looking package built on top of one of the same handful of big models underneath. Same engine. Somebody just put a new paint job on it and charged you for it.

Think about it like your tool belt. If you've got a good set of tools and you actually know how to use them, you don't need to run out and buy a specialty gadget for every single task. Most of those gadgets are doing what your good tools already do. You just hadn't learned the full range of what you had.

So before you ever pay for a new AI tool, ask one question. What does this do that my main tool can't already do? If the answer is nothing really, it just looks easier, save your money. Go learn your main tool better instead.

This matters even more at work, because your tools are already chosen and already approved. You don't need to go shopping. There are years of capability in the two or three tools on your organization's list that almost nobody ever touches. The win is not finding a new tool. The win is going deep on the ones you've got.

And there's a second reason, beyond the money. Fluency compounds. Every hour you spend in one tool makes the next task faster in that same tool. You start to know how it thinks, where it's strong, where it gets lazy. Tool hopping resets you to zero every time.

The people who get real value out of AI are not the ones with the most apps. They're the ones who picked one, learned it cold, and made it part of how they work. Depth beats novelty every single time.

So here's your challenge. Pick one tool. Just one. And this week, instead of trying something new, go deeper on the one you've got. Find one feature you didn't know was there. Push it on something messy and real, not just the easy stuff. That's where fluency actually gets built.

That's your eureka moment. You probably don't need a better tool. You need a better grip on the one already in your hand.

Next time, I'll show you how to stop starting from scratch every single time you open it. See you there.

Welcome back to episode four. Build your workspace.

Let me name something that quietly wears people down about these tools. Every time you open a new chat, it forgets everything. It doesn't know who you are. It doesn't know what you're working on. It doesn't know a single thing you told it yesterday. You start from zero, over and over.

There's a fix for that, and almost nobody uses it. It's called a workspace.

Different tools call it different things. A project. A notebook. A saved assistant. The name doesn't matter. What it does matters. A workspace gives AI a permanent home for one kind of work, so you stop re-explaining yourself every single time you sit down.

Here's how I think about it. A workspace has four parts.

One. The job description. You tell it once who it is and how you want it to work. You're my proposal review assistant. You're helping a patient access supervisor. Write plainly, no fluff. Write that once and it applies to everything you do in there.

Two. The brain. This is what you load in. Your policies. Your templates. Your standard forms. The reference material you're always digging around for. Put it in once and it stays.

Three. The tools. What it's allowed to reach. Your files, the web, whatever your organization has turned on. Some of that depends on what you've been given, and that's fine. Just know it's part of the setup.

Four. The tasks. Your conversations. Every chat you start inside that workspace already has the other three standing behind it.

Job description. Brain. Tools. Tasks. That's a workspace.

Now here's the part almost everybody misses, and it's the whole ballgame. Loading documents is not the same as using them. You can fill a workspace with excellent material and it will still answer out of its own training unless you tell it to look.

So put a line right in the job description. Something like this. Always check the documents in this workspace before you answer. Don't fall back on general knowledge when the answer is in my files.

Then say it again when you open a new chat. Before we start, review the materials in here, then help me with this.

Think about a sharp new employee. The binder is sitting on their desk. Sometimes you still have to say, check the binder first.

Let me make this real. Say you review vendor proposals against the same standards a few times a month. In the job description: you're my proposal review assistant, always check the documents here first, flag anything that misses our standards and tell me what to ask the vendor. In the brain: your standard terms, your evaluation checklist, two past proposals you thought were strong.

Then every time a new one lands, you open a chat, drop it in, and say, review this against our standards, where does it fall short.

You built it once. Now an hour of squinting is a two-minute job, every time. The setup is the work. After that it just runs.

So here's your challenge. Pick one job you do over and over from the same reference material. Reviewing something against a standard. Answering the same kind of question. Checking work against a policy. Open your assistant and build that workspace. Write the line that tells it to check your documents first. Then test it, and watch whether the answer actually comes back out of what you loaded.

That's your eureka moment. You stop starting from zero, and the thing you built keeps paying you every week.

Next time, I'll show you how to stop it sounding like a textbook and start it sounding like you. See you there.

Welcome back to episode five. Make it sound like you.

Let me ask you something. Have you ever handed AI a long document, asked for a summary, and what came back was completely accurate and completely useless? Correct, flat, and written like an encyclopedia. You'd have to rewrite the whole thing before you showed it to anybody.

Most people see that, shrug, and think, well, it's fine. And they leave the best part sitting on the table.

Because here's what almost nobody does. You can tell it how to sound. Once. And it remembers.

Here's the problem. Out of the box, it's writing for nobody in particular. So it writes like a textbook. It doesn't know whether you're a supervisor with four minutes between meetings or an analyst who wants every number. It doesn't know you can't stand jargon. Nobody ever told it.

So tell it. Every one of these tools has a place to give it standing instructions. Sometimes that's a setting for the whole tool. Sometimes it lives inside the workspace. Poke around, you'll find it. And it's not code. You just talk to it in plain language.

Something like this. Speak to me like a sharp colleague, not a textbook. Lead with the bottom line. Skip the jargon. I'm an operations person, not a tech person. If something matters, say it in the first sentence.

Three or four sentences. That's the whole thing. Who you are, how you want it to talk to you, what to skip, what to lead with.

And from that moment on, everything it makes for you follows those rules. Every summary. Every draft. Every breakdown. That's the difference between something you'd have to rewrite before anybody saw it, and something you'd forward as is.

Watch the difference in real work. Say a new policy lands in your inbox Friday afternoon and you're briefing your team Monday morning. You drop it into your source-grounded tool, which works from the sources you give it, anchored to your material instead of the open web. But this time you've already told it who you are and how you think. What comes back isn't a generic summary. It's a briefing, in your language, ready to use.

Two habits make this even better.

First, don't pile everything into one giant workspace. Give each one a job. One for a project. One for a topic you're learning. When the sources are focused, the answers are sharp. When it's a junk drawer, you get junk drawer answers.

Second, most of these tools give you a chance to steer each piece before they make it. Take it. Say what you want this particular summary to cover, who's reading it, how long it should be. Skipping that step is the number one reason people end up with something generic.

Now, the honest part. This doesn't make it perfect. It makes it yours. You still read what it gives you before it goes anywhere with your name on it. That never changes.

So here's your challenge. Find the place where your tool takes standing instructions, and in three or four sentences, tell it how you want to be spoken to. Then take a document you already ran through it and ask for the same thing again.

Read the two side by side. Same tool, same document. The only thing that changed was that you introduced yourself.

That's your eureka moment. Out of the box it sounds like a textbook. Set up right, it sounds like you.

Next time, we get into the one that costs people the most sleep. The message you've been putting off. See you there.

Welcome back to episode six. The message you've been putting off.

Let me ask you something. Is there a message sitting on your list right now that you have not sent?

Maybe it is a family member who is upset and is not wrong to be upset. Maybe it is a vendor who missed a date and you need to say so in writing. Maybe it is telling your team something they are not going to like. It has been there two days. You have opened it twice and closed it.

Here's the problem. It is not that you do not know what to say. You know exactly what happened and exactly what needs to happen next. What is stopping you is the first sentence. Because the tone has to be right, and once you hit send it is a record. It can be forwarded. It can end up in front of somebody you did not write it for. So you leave it.

Here is the shift. Do not ask AI to write the message. Ask it to help you land the message. Those are two different jobs, and the second one is where it is actually good.

So here is how you set it up. Same three things you always give it. Who it is, what the situation is, what you want back. But for a hard message you add one more, and this is the one people skip. Tell it the reaction you want.

Not, write a response to an upset family member. Instead. I want this family member to feel heard, I want them to understand what we are doing about it, and I do not want this going to my director.

That one line changes everything, because now it is not writing a message. It is writing toward an outcome.

Give it the facts. What happened, what you can actually promise, and just as important, what you cannot. If you are not authorized to say we will replace it, say so up front and it will stop offering that.

Now here is the move that makes this worth your time. Ask for three versions. One warm, one plain and factual, one firm. Read all three. You will know within about ten seconds which one is right, and you probably could not have told me in advance which one you wanted.

Then, before you send it, do this. Paste your chosen version back in and say, read this the way somebody who is already angry would read it. Where does it land wrong?

Watch the difference. It will find the sentence you thought was neutral that reads as blame. It will catch the word that sounds like you are making an excuse. That is a rehearsal you cannot get anywhere else at eleven at night.

One rule while you do this. Describe the situation without the names. A patient, a staff member, a family. It does not need who, it needs what happened. You will get the same quality of draft and nothing personal ever leaves your hands.

And you still read every word before it goes. Your name is on it. That never changes.

So here's your challenge. Take the one you have been avoiding. Give it four sentences of context and the reaction you want. Ask for three versions. Then ask it how the person on the other end is going to read the one you picked.

That's your eureka moment. The hard part was never the words. It was going first. Now something else goes first, and you decide.

Next time, what all of this looks like if your day does not happen at a desk. See you there.

Welcome back to episode seven. If you're on your feet all day.

This one is for the people whose day does not happen at a desk. You are on a unit, in a supply room, in a kitchen, on a loading dock, or driving between patient homes. You might sit down at a terminal twice a day if it is a good day.

Here's the problem with almost everything you have heard about AI. It assumes you have a keyboard and twenty quiet minutes. You have neither. You have a phone in your pocket and about ninety seconds between one thing and the next.

So here is the shift, and it is the whole episode. Stop typing. Talk to it.

Every one of these tools takes your voice. You open it, you tap the microphone, and you talk the way you would talk to a sharp coworker riding along with you. Same three things as always. Who it should be, what is going on, what you want back. You are just saying them out loud instead of thumbing them in.

Let me make it real.

You just walked a building. Instead of trying to remember it until you are back at a terminal, you step into a quiet hallway and talk for two minutes. Ceiling tile stained outside the second floor elevator, a door on the east wing not latching, two lights out in the waiting room, the air handler on the roof is louder than it was last month. Then you say, turn that into a punch list, group it by trade, and tell me what should not wait.

You get back a clean list. You did not write it. You said it.

Supply deliveries run late three days in a row and you need to tell the units something. You talk out what happened and what you are doing, and you ask for a short notice that does not sound like an excuse.

The truck comes in and half of what is on the packing slip is not what is on the dock. You talk through the discrepancy and ask for a clear note to the supplier with the questions you need answered.

A vendor is coming Thursday to service equipment you have not dealt with before. You tell it what the job is and say, give me the eight questions I should ask this technician before they leave. Now you are not the one nodding along hoping it is fine.

And here is one more that people are slow to try. Most of these tools let you show it a picture. A part you do not recognize. A label on a piece of equipment. A page of instructions you are squinting at. You show it and ask what you are looking at. It will not always be right, and you will know when it is not, because you know your building. But it gets you a starting point right where you are standing instead of a phone call you have to wait on.

Two rules. No names. Not patients, not staff, and never a photo with a patient or a chart in it. Describe the situation, not the person. And it does not know your building. It does not know that air handler is thirty years old and the part is on back order. That part is you. It is a fast set of hands, not a replacement for twenty years of knowing what that noise means.

So here's your challenge. Next time you finish a walkthrough, do not write anything down. Step somewhere quiet when you are done, open your phone, tap the microphone, and talk for two minutes. Then ask for the clean version.

That's your eureka moment. You do not have to go back to a desk to use this. It rides along with you.

Next time, the same idea for the people who do sit at one. See you there.

Welcome back to episode eight. If you work at a desk.

This one is for the desk. Patient access, scheduling, billing, coding, medical records, technology support. Anywhere the work arrives in a queue and the queue refills while you are looking at it.

Let me describe your Tuesday. Eleven tabs open. A spreadsheet somebody sent that is almost the format you asked for. A form filled out wrong that you have to send back politely. Two people waiting at the front desk and the phone ringing. And the thing you actually needed to finish today is still sitting in the corner of your screen.

Here's the problem. When people talk about AI saving time at a desk, they talk about typing faster. That is not your bottleneck. Your bottleneck is that you handle forty things that all look a little different, and every one of them takes a decision.

So here is the shift. The win at a desk is not speed. It is that you stop starting over.

Because here is what is true about desk work that is not true about much else. You do the same kinds of things over and over. The same kind of request. The same kind of comparison. The same letter with different details. That repetition is the opening. Set it up once, use it every week.

Here is what that looks like. Say quotes come across your desk. Rather than reading three of them side by side every time, you write down once what you always need to know. Total cost, what is included, what is not, delivery time, what is missing that you would have to ask about. Save that. Now every time quotes come in, you hand them over with that same instruction and you get the same comparison in the same shape. You read it, you make the call, you move on.

Or the question you answer forty times a week at the front desk. Ask it to write the plain language answer, short, no jargon, at a level anybody can read, and ask for a Spanish version if your patients need one. Now you have a card instead of a conversation.

Or the policy that landed in your inbox that you have to explain to eight people who are not going to read it. You drop it into your source-grounded tool, the tool that works from the material you give it, and ask for the four things that change for our department and what people are going to ask.

And here is my favorite one for a desk, because it costs nothing. The second set of eyes. Before the message goes to every clinic manager or the whole department, paste it in and ask, what is going to be misunderstood here, and what question am I going to get back? It finds the ambiguous sentence every time. That is one round trip you just saved yourself.

Two rules, same as always. Records stay out. No names, no patient information, nothing from a chart, nothing that belongs in a personnel file. Describe the situation, not the person. And check anything that looks like a number or a date, because it will be confident either way. More on that in the last episode.

So here's your challenge. Pick the one thing you do the same way every single week. Do it once with AI, and this time write down the instruction you used. Save it somewhere you will find it. Next week, use it again.

That's your eureka moment. You are not trying to do Tuesday faster. You are trying to never build Tuesday from scratch again.

Next time, the thing that lives in one person's head. See you there.

Welcome back to episode nine. Before it walks out the door.

Every operation I have ever seen has at least one of these. The person who knows how to get the old scanner talking to the system, and it is not what the manual says. The one who knows which payer kicks back the same claim every single month. The one who knows what to do at six in the morning when the system is down and nobody is answering the phone.

None of it is written anywhere. It is in their head. And everybody knows it, including them.

Now let me ask you something. How many times has somebody said we really should document that? And how many times did it get documented?

Here's the problem, and it has nothing to do with anybody being lazy. The knowledge is not the hard part. The writing is the hard part. Sitting down and turning twenty years of knowing something into numbered steps is miserable work that nobody has an afternoon for. So it does not happen. Then somebody retires and the team spends a year rediscovering what that person could have told them in fifteen minutes.

Here is the shift. This was never a documentation problem. It was a transcription problem. And that is the one thing AI is genuinely excellent at.

So here is the move, start to finish.

Sit down with the person. Fifteen or twenty minutes, that is all. Turn on the recorder on your phone and ask them to walk you through it the way they would tell a new hire. Not formally. Just talk. What do you do first, then what, and what do you do when it does not work.

Let them ramble. Rambling is good. The tangents are where the real knowledge lives.

Then take that recording, get the transcript, and hand it over with an instruction like this. You are a technical writer. This is a recording of an experienced person explaining how they do a job. Turn it into a numbered procedure in plain language that somebody new could follow.

And then add the line that makes this worth doing. Say this. Flag anything the speaker assumed I would already know.

That is the gold right there. Because an expert skips steps. Not on purpose. They skip them because after twenty years those steps are not steps anymore, they are just breathing. AI does not know any of it, so it notices every gap. You get a list of the holes, you go back to that person for five more minutes, and you fill them in.

Then the draft goes back to them. They read it, they fix what is wrong, and they sign off. That matters, and not just for accuracy. That procedure is theirs. Their name goes on it. They did not lose anything by writing it down. They became the person who defined how this gets done here.

And you can do this for yourself. You do not need permission and you do not need a project. There is something you do that nobody else does the same way. Twenty minutes of talking and it exists.

So here's your challenge. Pick one thing. The one where, if you were out for two weeks, somebody would be calling you. Record yourself explaining it for fifteen minutes. Get the procedure. Read what comes back.

That's your eureka moment. It is not paperwork. It is your team keeping what you know.

One more to go, and it is the one you have been waiting for. What happens when AI gets it wrong. See you there.

Welcome back to episode ten. When it gets it wrong.

If you have been waiting for the episode where I admit this stuff is not perfect, here it is.

AI gets things wrong. Not rarely. And here is the part that gets people. It is wrong in exactly the same confident voice it uses when it is right. A number that does not exist. A policy that was never written. A date that is off by a year. All delivered like it is reading from the manual.

That is the number one reason people try this twice and quit. They catch it in a lie, and they decide the whole thing cannot be trusted.

Here's the problem with quitting there. Most of the time, it was not lying. It was guessing, because you did not give it anything to work from.

Think about it this way. Walk up to somebody on their first morning and say, write me a summary of how we handle purchase orders here. What are you going to get? Something that sounds about right and is wrong in half the details. Not because they are dishonest. Because they have no idea, and they are trying to be useful.

That is exactly what is happening. You asked a question with no context, so it filled the gaps with the most likely sounding answer.

So here is the fix, and it is two parts.

Part one. Give it your material. Your policy, your contract, your actual numbers. When it is working from what you handed it instead of from thin air, the guessing mostly stops.

And when you do, say this out loud in the instruction. Use the documents I gave you. Do not fall back on general knowledge when the answer is in my files. Because loading a document and telling it to read the document are two different things, and almost everybody does the first one and skips the second.

Part two. Four checks, every time, and they take about thirty seconds.

Check the facts. Anything that looks like a number, a date, a name, or a citation, verify it. Especially if it is going in front of somebody.

Notice that it never hedges. A person would say, I think, or I am not sure. It will not. So the confidence in the answer tells you nothing about whether it is right. You have to be the one who is unsure.

Make it show its work. Say, walk me through how you got that, or, what are you assuming here. The weak spot usually falls out on its own.

And correct it. When something does not match what you know, say so. That is not our process, ours works like this, try again. It will adjust immediately, and the rest of that conversation gets sharper.

Here is the part I want you to hold on to. You are not checking it because the technology is bad. You are checking it because you are the one who knows. It does not know your building, your patients, your payers, or what happened last spring. That is not a gap in the tool. That is your value, and it is the reason nothing goes out without you reading it first.

So here's your challenge. Go back to something it gave you this week. Read it again, slowly. Find the one thing that does not quite sit right. Then tell it exactly what is wrong and ask it to fix it. Watch how fast it corrects when you give it something real.

That's your eureka moment. Anybody can get an answer. Knowing how to catch the bad one is the whole skill.

And that is the series. You have got the recipe, you know what you are allowed to do, you know to go deep instead of wide, you have built a workspace, you have made it sound like you, and you have pointed it at the work you actually do.

You are the pilot. That never changes. Go use it.