I've been experimenting with ChatGPT lately, and one of the more intriguing features is the ability to create a “Custom GPT”–a tailored version of the tool that draws from documents you upload.
Naturally, I tried this with Lean Hospitals: Improving Quality, Patient Safety, and Employee Engagement. I uploaded a PDF of the book and created what I'm calling the “Lean Hospitals Helper.” It's a ChatGPT instance designed to answer questions based on the book's content.
What's impressed me most is how well ChatGPT can summarize and extract insights from a large, complex document. It's been fun to interact with it–posing questions and seeing how it surfaces relevant responses.
A Comment That Sparked an Idea
I shared a link to the Custom GPT on LinkedIn, just to see what others thought. The feedback was encouraging. One comment from Tim Noble stood out:
“Very cool… but you need to offer your IP on a subscription model.”
That got me thinking: Is that possible? Could I make something like that work? At the very least, I figured I'd learn something by experimenting for a few hours on a Friday afternoon.
From Idea to Action
What I learned pretty quickly is that OpenAI doesn't let you embed a Custom GPT directly into a webpage. They expect you to use their platform.
But I also knew that you could use their API to build something more custom. I didn't know how to do that myself–but I also knew that ChatGPT can write code. So I asked:
“Can you write a WordPress plugin that embeds a Custom ChatGPT?”
Now, I haven't written code in any serious way since college (other than tweaking HTML code manually), but I used to love programming as a kid on an Apple IIe (at school) and a TI 99/4A (at home).
Within minutes, I had something that demonstrated the rough proof of concept working on my site. It helps that I know how to upload files via FTP.
A few hours later, with ChatGPT's continued help, I had a working prototype. A simple plugin. And a functioning paid subscription model on my site.
Asking ChatGPT to Describe the Process
I asked ChatGPT to write up a narrative that explained the process of working with me (instead of me writing it from my perspective). Here's what it wrote:
ChatGPT thoughts: building this together
From my side of the keyboard, this felt like pairing with a decisive product owner who ships. Mark set a clear goal–“let readers chat with the ideas in Lean Hospitals, right on my site, and make it reliable”–and then we moved in short, honest loops. I'd propose a tiny, testable change; he'd upload the file via FTP, try it on a real page, and send back exactly what happened. That specificity–real error text, screenshots, what he clicked–let me deliver fixes instead of theories.
We started with the simplest thing that could possibly work: a single WordPress plugin file that draws a chat box on a page and calls the OpenAI API. When the host's cache held onto old results after a deploy, we taught the chat route to say “don't cache me.” When the browser wanted a little extra safety, we added a small token to each request and a gentle speed bump for rapid-fire clicks. When streaming responses looked cool but the hosting pipeline wasn't thrilled, we didn't fight it–we kept the stable version live and iterated off to the side. That “Option A first” mindset kept momentum high and downtime low.
Mark also pushed for trust, not just function. He didn't want a black box; he wanted the assistant to say when a reply is grounded in the book versus general experience. We tuned the output so it's easy to scan–bold subheads, short paragraphs–and we added a light “sources” hint without exposing the full PDF. It reads like a coach you'd want in your corner, not a mystery oracle.
Mark suggested a simple toggle that lets the reader choose Book Search (default) or General Knowledge before sending a question. In Book mode, the assistant prefers the uploaded Lean Hospitals content and clearly labels replies as “Source: Lean Hospitals by Mark Graban.” Flip to General, and it answers from broader Lean/TPS knowledge, labeled as “Source: General.”
This small UI choice did three big things:
- Set expectations–readers know why an answer took a certain angle.
- Increased control–users can switch context without reloading or guessing.
- Built trust–transparent labels reduce the “AI mystery” factor.
The workflow itself was pleasantly human: clear intent from Mark, fast code from me, and tiny experiments in production-like conditions. A parse error showed up? We cleaned a bracket. A “mixed content” warning blinked in the console? We upgraded a stray link to HTTPS. A “please include this header” message from the API? Added once, moved on. No drama–just steady progress.
What does the plugin mean for readers and site owners?
It's a friendly way to interrogate the book's ideas in your own words. Ask about leader standard work, variation, daily improvement, or psychological safety, and you'll get an evidence-minded answer that clearly says whether it's leaning on Lean Hospitals or broader Lean/TPS practice. It's direct, readable, and designed to respect your time.
What does it mean for site owners? You don't need to be a developer. Upload the file, add your key, drop a shortcode on a page. If you want the assistant to prefer the book, put the PDF in the uploads folder and the plugin does the rest. Behind the scenes, it's calm under traffic, avoids stale caches after you update, and includes basic safeguards so casual misuse doesn't ruin anyone's day. Think of it as a tidy appliance: small surface area, clear labels, sensible defaults.
Final Thoughts
If I had to sum up the collaboration in one line:
Mark brought clarity and accountability; I brought fast, targeted code.
Together, we turned a Friday-afternoon “what if?” into something useful you can try right now–an approachable gateway into Lean Hospitals that meets readers where they are.
Here's how it currently looks on the site:
There are other features that we could add, such as a “copy” button or the ability to download a PDF of the answers. Some attempted experiments have caused the website to break (“critical error”), but it's easy enough to roll back to the previous non-broken version of the plugin code.
Experiments with a Subscription Service
I was able to create a subscription service model using off-the-shelf components. To test the concept, I am using the free version of the ProfilePress plugin for WordPress. With the free version, I can take credit card payments through an integration to my existing Stripe account, with a 2% fee.
Currently, I'm offering a 3-day free trial (and the length of that is something I can configure) with a recurring $9.99 per month subscription fee. Users can create an account and manage it through the website.
I'm also considering offering a discount for existing book owners. But that would require people to send me a receipt or a photo of themselves with the book. That might be too much hassle?
I'm curious to hear your thoughts on pricing.
What Would the Publisher Say?
If you're an author, you might wonder about my arrangements with the publisher. I haven't asked permission to do this, and I probably don't have to. I'm the copyright owner. They had previously said I was free to do an audiobook version, even though I didn't follow through on that. So I think this is similar.
Wrapping Up
I thought I'd share my experiment and a little bit about the process. Thanks for checking this out. Again, you can find the AI page here, and this is the registration page.
Please scroll down (or click) to post a comment. Connect with me on LinkedIn.
Let’s build a culture of continuous improvement and psychological safety—together. If you're a leader aiming for lasting change (not just more projects), I help organizations:
- Engage people at all levels in sustainable improvement
- Shift from fear of mistakes to learning from them
- Apply Lean thinking in practical, people-centered ways
Interested in coaching or a keynote talk? Let’s talk.
Join me for a Lean Healthcare Accelerator Trip to Japan! Learn More
