If you run an independent specialty group, you already know the squeeze. Staff are stretched, costs keep climbing, every payer wants more paperwork, and the larger groups and health systems around you have bigger budgets for technology. Then every week someone pitches you a new AI tool to fix it. An ambient scribe, a prior auth bot, a phone agent, an “AI-powered” add-on to something you already pay for.
Most of those pitches skip the questions that actually decide whether AI helps your practice. Which work is ready for AI? Who checks it when it’s wrong? Will your team use it, or work around it? And how will you know, in hours and dollars, that it paid off?
We help practices answer those questions, and we run our own business the same way. So we took one part of HealthSpaces, our marketing and business development work, and rebuilt how it runs with AI in the middle of it. This post walks through it with screenshots, and after each part, what the same approach could look like inside a practice like yours. The biggest lesson is one every practice leader should hear: with AI, you can now build as well as buy, and the practices that plan both together will get far more out of AI than the ones that keep bolting on one more tool.
Our Problem Looked a Lot Like Yours
Like a lot of firms our size, our marketing and business development runs lean, on one person. Before this project, that work was spread across a stack of tools that didn’t always talk to each other:
- A CRM for contacts, companies and deals
- An email platform for newsletters
- A website builder for the site and blog
- A scheduling link for booking calls
- A cookie banner service, an SEO plugin and a handful of other add-ons
To be fair, a big all-in-one marketing platform covered a lot of this. But it came at a real price, and it was built for every kind of business, not for how we actually work. We paid for features we never touched and still filled the gaps by hand: copying leads between tools, rebuilding lists and keeping a running to-do list in our heads, because no tool knew what our week actually looked like.
If you run a practice, swap in your EHR, your PM system, your phone system, the fax queue, the patient reminder tool, the prior auth portal and the spreadsheets your front desk keeps “just in case.” Same story, bigger stakes. Your staff become the integration layer between systems that were never designed to work together, and that’s where the hours go.
That’s the real opportunity for AI in a practice. Not a shiny new tool on top of the pile. Taking work off the people who are stuck in the middle of it.
Step One Wasn’t AI. It Was Writing Down the Work.
We say it to every practice: optimize before you automate. Hand AI a messy process and you get a messy process that runs faster. So before we touched any AI, we wrote down everything our team actually does, on what day, and how long it takes.
That list covered every daily, weekly, monthly and quarterly task, plus the ones triggered by dates, like a conference three weeks out. It became the blueprint for everything we built. If a task wasn’t on the list, nothing got built for it. If something got built that didn’t serve the list, it got cut.
The second thing we did was look at how our existing tools were actually performing, not how we assumed they were. That’s where it got humbling. We found that our old website’s contact form had likely never reported a single lead back to Google Ads. The form showed a thank-you message and never redirected, so the ad platform couldn’t see the conversion. We’d been running ads and judging them with no real signal.
No AI found that. Asking “does this actually work?” found it. That’s the point of mapping first: half the value shows up before any automation does.
If you do one thing after reading this, do this part. Pick one role in your practice, front desk, billing, referral coordinator, and write down every recurring task they do and how long it takes. You’ll learn more about where AI fits from that list than from any vendor demo.
The Big Shift: Build, Don’t Just Buy
For years, the math on custom tools was simple. Building meant a development team, months of work and a big budget. So you bought. You picked the closest off-the-shelf product, paid the license every year, and bent your workflow to fit it.
AI changed that math. Coding with AI meant we could build exactly what our map called for, and build it fast. What would have been a months-long project took about two weeks to get to a working version. When something didn’t fit, we changed it that day instead of filing a request and waiting on a vendor’s roadmap.
Speed is only half of it. The bigger win is that everything we built works together. The website, the console, contacts, account planning, booking, leads and reporting all share one set of data. A form submission on the site shows up in the leads inbox, on the right account, in the daily task list and in the reports, without anyone copying it anywhere. And it connects to the tools we chose to keep, like our old CRM for the sales side, Google Analytics and Google Ads, instead of replacing everything at once.
That’s the difference between a pile of tools and an ecosystem. A pile of tools makes your people the glue. An ecosystem does the connecting for them. Practices will thrive off of software that fit their workflows instead of having to change them to an out of the box product.
But build doesn’t replace buy. It adds to it. A practice still buys necessities like its phone system and the network switch in the clinic, and still needs the lights kept on. The real trap is the opposite: AI showing up as one more thing to buy and bolt on, one problem at a time, until the patchwork is worse than before. The answer is deciding each piece on purpose, with a plan for how it all fits together.
Then We Put It All in One Place
With that list in hand, we built our own system: the upcoming new healthspaces.com plus a private console behind it that we call hTide. One sign-in, one place for the website, the blog, contacts, account planning, booking, reporting and our daily work.
Here’s the part that surprises people. We have developers who build custom software for our clients, but they didn’t build this. Our marketing team (of one person) did, without much coding experience, working side by side with Claude Code. It took about two weeks to go from first build to a working staging site. That’s what AI changes: the people who know the work best can now build the tools for it, instead of writing a wish list and waiting their turn. We own the code and the data, and it runs on Google Cloud under our control. Here’s a walk through the parts that matter most. Keep in mind these screenshots are an early version. It already does the job, and it’ll look and work better every round from here.
A note on the screenshots: every practice, contact, email and phone number shown is made-up demo data. None of it belongs to a real client, prospect or contact.
My Work: The Day Starts With a List, Not an Inbox

The day starts with a list built from the actual work, not an inbox.
The console opens on My Work, not a dashboard. It takes every recurring task we wrote down and turns it into today’s list: what’s due, when, and roughly how long it should take. Each task has a short “how” checklist and an Open button that goes straight to the screen where the work happens.
Every task also has a small question mark that explains why it exists. “Why check Google Ads every Monday?” has an answer right there. That sounds minor, but it’s what keeps a routine from turning into busywork. If we can’t explain why a task matters, it comes off the list.
Some tasks come from the calendar (newsletter week, Tuesday’s LinkedIn post). Others come from dates in the system, like a client review coming up in three weeks or a conference follow-up due within 48 hours of the event.
In your practice: picture your front desk or billing team starting the day with a short list of what’s due, built from how the work actually flows, instead of digging through the fax queue, the portal inbox and a sticky note. Same idea. The list comes from mapping the work, and AI helps prepare what’s on it.
Accounts and Contacts: Find Anyone in a Second

Every account in one filterable list, with fit, stage and next step side by side.
We brought in thousands of companies and contacts from our old CRM, linked contacts to their practices, and made every field filterable. Need every practice administrator at an orthopedic group in Colorado who visited our booth at a conference? That’s a couple of filters, saved as a view the whole team can use.
Our sales side still lives in the old CRM for now, and that’s fine. hTide pulls updates from it every night and pushes every new website lead into it right away, so nobody copies anything by hand and nobody has to switch tools before they’re ready.
In your practice: this is the lesson for any practice nervous about ripping out a system people depend on. You don’t have to replace everything on day one. Connect what you have, move work over in pieces, and let people switch when the new way is clearly easier.
Accounts: Focus on the Practices That Fit

The fit score shows its math, and every seat on the buying committee is accounted for.
This is the part that replaced our mental to-do list of “who should we be paying attention to.” Every practice gets a fit score out of 100 based on the kind of group we serve best: specialty, number of providers, number of users, locations, surgery centers and ownership. The score shows exactly where its points came from and which fields are still missing, so it’s never a black box.
Each account also shows its buying committee. For a specialty group that’s usually four seats: the administrator or COO, the managing partner or board, the CFO or revenue cycle lead, and the IT lead. Empty seats show up dashed, so we can see at a glance that we know the COO at a practice but have never met anyone on the finance side.
Accounts get a tier by hand, by us. Tier 1 is up to about 15 practices we work one to one. Tier 2 is a few dozen we work by specialty. Everything else is Tier 3. The tool tells us what to do for each tier. People decide who’s in which.

Every target account by stage, and what moved this week.
The overview shows every target account by stage, from Unaware through Customer, and what moved since last week. That’s the slide we used to build by hand for our Friday owners meeting. Now it’s a button that copies a plain-text summary.
In your practice: the same pattern works for any queue where your team has to decide what to work first. Denials, prior auths, referral follow-ups, recall lists. A clear, explainable score puts the highest-value work at the top. Your people still make the call. They just stop starting every day from a blank page.
Blog and SEO: Off the Old Website Without Losing Anything
We moved all 40 of our existing blog posts over with their images, and every old website address either kept its exact URL or points to the right new page. Hundreds of old links, including PDFs linked from past emails, keep working. The blog editor has the SEO fields built in, with a Google preview and a short checklist, so there’s no separate plugin to manage.
In your practice: moving off an old system is where most technology projects go sideways, whether it’s a website, a PM system or an EHR. The work that keeps it from breaking is unglamorous: knowing every link, every report and every workflow that touches the old system before you turn it off.
Leads and Booking: Every Form Goes Somewhere

Every form lands in one inbox, with the form, score and status at a glance.
Remember the contact form that never reported a lead? Now every form on the site lands in one Leads inbox, sends the team an email, gets checked for spam, and is tracked properly in Google Analytics and Google Ads. The Book a Call page books straight onto our calendar with real availability and a video link, instead of a back-and-forth email chain.
In your practice: how many patient calls, referral faxes and portal messages land in different places today, and how many fall through? Getting every request into one place, tracked, is often worth more than any AI layered on top. It’s also what makes AI useful later, because it finally has one clean place to work from.
Dashboard: The Numbers in One Place
Website traffic, search performance and ad spend sit on one dashboard with one date picker, instead of three logins and three different definitions of “last month.”
In your practice: most groups we talk to pull their numbers from the PM system, the EHR, the clearinghouse and a spreadsheet, and each one tells a slightly different story. One view of the numbers that matter, with one definition of “last month,” is where good decisions start.
Where the AI Actually Fits
It would be easy to call all of this “AI-powered” and leave it there. That’s exactly the kind of vague claim we tell practices to push back on. So here’s specifically where AI does the work and where it doesn’t.
AI built most of it with us. Claude wrote the large majority of the code. Our job was the part AI can’t do: knowing what the work is, deciding what to build, saying “no, that’s not how this should work,” and checking every screen before it shipped. Think of it less like hiring a developer and more like having a very fast one who needs a clear brief every single time.
AI drafts. People approve. Inside the console, AI writes first drafts: SEO titles and descriptions for blog posts, research on accounts, and soon outreach emails that land in a queue for our team to review. Nothing it writes gets sent or published on its own. That’s the same rule we hold for every practice we work with: a person reviews before anything acts.
AI has guardrails, not just a prompt. We didn’t just tell the AI to follow our rules and hope. When it drafts copy, the system checks the draft before anyone on our team sees it. It fixes punctuation that doesn’t match our style, rejects any compliance claim we don’t make, and rejects any number that isn’t already in the post. If a draft invents a statistic, it never makes it to anyone’s screen.
AI checked AI. Before we went to launch, we had two separate AI reviews go through the whole codebase looking for security and reliability problems. They found real issues, things like how stored links were displayed and how forms behaved if a visitor closed the tab mid-submit. We held the launch until every one was fixed and checked. AI didn’t make the call to hold. We did. But it found problems a one-person team never would have caught alone.
AI doesn’t decide who matters. The fit score is a formula we wrote and can explain line by line. Tiers are set by people. The AI helps us act on those decisions faster. It doesn’t make them.
In your practice: these same rules matter far more when patients are involved. AI can draft a prior auth letter, a referral summary or a reply to a portal message. A person on your team still reviews it before it goes anywhere. The guardrails should be built into the system, not left to a policy nobody reads. And anywhere patient data is involved, the vendor signs a BAA, no exceptions.
For what it’s worth, the console holds business contacts only, no patient information. But we built it the same way we’d build anything for a practice: on Google Cloud, with sign-in limited to our team and every role checked on the server, not just hidden in the interface.
What We Got Wrong
Building the tool turned out to be the easy part. Using it was harder.
Once the console was working, we got pulled into polishing website copy. Day after day. Meanwhile the task list the system put up every morning sat there, mostly unchecked. We told ourselves most of it was blocked on account work we weren’t ready for yet. When we actually counted, more than half of it wasn’t. It was routine work that could have been done that day.
We caught it in our own weekly review, and it stung a little. The tool can’t give you time back if you work around it. Now the routines come first, and anything we do that isn’t on the list gets added to the list, so the work is visible instead of disappearing into “we were busy.”
We also had a wrong assumption about the account tiers. Somewhere along the way, the idea crept in that the system would assign Tier 1, 2 and 3 on its own. It doesn’t, on purpose. People pick the accounts that matter most, and the tool runs the plays once they do. It was a good reminder to learn our own tools as well as we expect clients to learn theirs.
That’s the honest lesson for any practice rolling out AI: adoption is the project. The technology working isn’t the finish line. Your team changing how they work is. That’s why we stay involved long after go-live, and why we’d be skeptical of anyone who promises a tool will fix a workflow by itself.
What’s Next
This is just the start. What you see above is the first version, and that’s on purpose. We built what we needed most, started using it, and now every week of real use shows what to fix, simplify or add next. Each round makes it a little better, the same way we’d roll out AI in a practice: start small, use it, learn, improve. Up next in hTide:
- Email campaigns and the newsletter moving off our old email platform
- Engagement scoring, so the system notices when a practice starts paying attention
- AI-drafted outreach for each tier, landing in an approval queue
- Gifting and follow-ups tied to real moments like conferences and client reviews
What This Looks Like From Your Seat
AI in a practice touches everyone, but not the same way. Here’s how we’d think about it depending on where you sit.
If you’re the CEO or managing partner, this is a strategy question before it’s a technology one. Staying independent means doing more with the team you have, and AI is one of the few levers that can actually do that. The risk isn’t moving too slowly. It’s buying tools one at a time with no plan, and ending up with the same pile of disconnected systems we started with, just with AI stickers on them. AI makes it realistic to build where it sets you apart, on top of the systems you buy, instead of renting everything off the shelf.
If you’re the COO or practice administrator, this is about your team’s day. Start with the work they do over and over, map it, and put AI where it takes something off their plate. Not where it adds one more screen to check. Adoption is your project, and it’s the part that decides whether any of this pays off.
If you’re a physician, the question is simple: does this give you time back with patients, or add clicks? Good AI in a practice does the drafting and preparation, and leaves clinical judgment and the final say with you and your team. If a tool can’t explain what it did, it shouldn’t be acting on its own.
If you’re the CFO, ask what every tool costs to license, connect and support, and what it gives back. Measure in staff hours and in revenue that stops leaking, like denials worked faster or referrals that don’t fall through. If a vendor can’t tell you how to measure it, that’s your answer. And it’s worth running the build-versus-buy math again. With AI, the build side of that math looks very different than it did even two years ago.
What This Means for Your Practice
What you should get from AI isn’t a model, a license or a strategy deck. It’s a workflow that takes less of your team’s time than it did before, and you should be able to say how much. Getting there takes three things working together:
- A clear strategy for which work is ready and which isn’t
- A secure IT foundation for the AI to run on, with signed BAAs anywhere patient data is involved
- Someone keeping it running after go-live, and helping your team actually use it
Miss any one and the project stalls. A great tool on shaky IT breaks. Solid IT with no plan automates the wrong thing. A great plan nobody follows changes nothing, as we found out firsthand.
If you’re a practice administrator, COO or managing partner thinking about AI, here’s where we’d start:
- Write down the work. Pick one role and list every recurring task, how often it happens and how long it takes. Be honest about the workarounds.
- Check what you already have. Before adding anything, make sure your current systems do what you think they do. We found a form that hadn’t reported a lead in who knows how long.
- Plan build and buy together. Keep buying what should be bought, and before the next AI add-on, ask whether something built around your workflow and connected to your other systems would fit better.
- Pick what’s ready. Start with repeatable, rules-based work where a mistake is easy to spot and easy to fix. Leave judgment calls with people.
- Keep a person approving. AI drafts and prepares. Your team decides and sends. Build that into the workflow, not just the policy.
- Ask for guardrails, not promises. Any AI tool touching your practice should have checks built in, not just good intentions. Ask vendors what stops the AI from making something up.
- Measure in hours. If you can’t say how much time it gives back, you can’t say it’s working.
- Plan for adoption. Budget real time for your team to change how they work. The tool is the easy part.



