AI in Dentistry: Radiograph Analysis and Case Presentation with Overjet Founder Dr. Wardah Inam

Welcome everybody to another episode of the Dental Boardroom podcast. I'm continuing on my series on AI in dentistry and I have just a wonderful guest on the show today who has been involved in AI for a number of years, I think was very much ahead of the curve and has just a very, very prestigious resume and I think is going to share a lot of really good insight onto what's happening in AI and dentistry and, and will also sort of spotlight her company as well. So Warda, thank you for joining. It's Warda Enum, is that right?

Yeah, that's right. Thank you for having me.

Happy to have you on the show. You are the founder of a company called Overjet AI, is that correct?

Yes.

Excellent. Well, I'm going to do my best here, Warda, just to give a little rundown of what is Overjet. Go ahead and clarify for me and then I want to give a little bit more on your background leading into the startup of Overjet. So here's Overjet AI. It is a Boston based dental artificial intelligence platform founded in 2018 by you. It is the first FDA cleared AI solution in dentistry for both caries, cavities detection and bone level measurement, making it a trusted clinical partner for dentists, D CSOs, payers and educators. And here are the four main areas that I found on what Overtjet is doing. Number one, smart radiograph analysis. So it instantly scanned, scans X rays during appointments, highlights areas of decay or bone loss with color coded overlays and helps prevent misdiagnosis. Number two, patient communication booster. Annotated images aid in educating patients which has led to a 10 to 20% increase in treatment acceptance and case in case studies. Number three, streamline operation. Supports insurance claim accuracy and efficiency by enhancing documentation, reducing denials and accelerating reviewer decisions. And then lastly, number four, workflow. Ready and scalable. It integrates with existing imaging and practice management systems with minimal disruption adopted across solo practices, large DSOs, insurers and dental schools. And Overjet's mission is to standardize diagnostic precision and drive patient centered oral health care, blending advanced AI with real world dental workflow. How did I do, Warda?

You did pretty good. That seems a very comprehensive overview about Overjet.

Great. If there's nothing else you want to add to that general description of Overjift. And I just want to mention your background, which is as impressive as they come. So you have a PhD in electrical engineering and computer science at MIT Massachusetts Institute of Technology. You also have a degree in entrepreneurial management from Harvard and also a Master's degree from mit, also in electrical engineering and computer science. And that is quite an impressive resume and I think tees you up pretty well to do what you do. So let's start off with the questions. I want to start off with why and how you got into AI and dental, specifically with Overjet.

So I, I was. And for your listeners as well, I do have some hiccups. So if I, you know, if I sound a little weird, it is because of that. So I got into or started with AI back in 2013.

When we were focused on how to make energy systems more intelligent. And with that then I focused on AI in healthcare in 2016, 2017.

And really made this conscious decision to move towards healthcare because I felt that we could have real impact very quickly utilizing AI and and then basically came to dentistry as a patient. I went, I changed my dentist after changing my neighborhood and got a treatment plan which was very different than I had received before. And the more I learned, the more obsessed I got about dentistry, like all the way from diagnosis, but how cool it was with all the different treatments that were happening. I think the more I understood about the industry, the more I realized that technology could play a big role here. And so I quit my job and started shadowing dentists, trying to understand how it could help. And, and that's how Overjet started.

Let's go into a general question now. With this backdrop of who you are and how you got into dentistry and AI, specifically in dentistry and of course the startup of Overjet, how do you see where is AI in its evolution inside of dental offices? And what do you forecast over the next two to three years? Are you looking to buy or sell a dental practice? If you're a seller, how do you find a strong list of potential buyers? There's no MLS or Zillow for dental practice sales. In such a fragmented market, many dentists selling their practice feel discouraged. That's why I built practiceOrbit.com practice orbit is modernizing how dental practice are bought and sold. It's a platform technology. Y' all know them Zillow, Airbnb, Uber, and OpenTable, just to name a few. These technologies do two things. First, they help match buyers and sellers by centralizing the marketplace. Second, once the match between buyer and seller is made, these technologies help them navigate the sale to a successful close. Here are a few of the features inside of Practice Orbit. Users can determine a reasonable sale price for practice in just minutes. All they need is the most recent business tax return. Practice owners can use the directory to Find important service providers like practice brokers, attorneys, accountants and bankers. Practice owners and their brokers can invite those service providers into the platform. So all parties are coordinating the sale together. Practice owners will have a beautiful digital listing of their practice like a home on Redfin. And importantly, the listing will be anonymous and personal information like your address, name and pictures are only released after an NDA is signed and approved by you or your broker. Practice buyers can create saved searches to be autonotified when new listings come on the platform that fit their desired specialty and location. And hey dental practice brokers, you can list all your practices on the site for free. By doing so, you'll get access to to Practice Orbit's large pool of buyers, all of whom have created profiles and saved searches to be auto notified when new listings come for sale. You can use the platform to introduce your listings to a targeted set of buyers. Additionally and optionally, you can use the platform tools such as a built in NDA, a customizable LOI and a workflow tool to navigate the sale from LOI to closing. In fact, you can manage all your listings by on the system from listing the practice to NDA to LOI and finally to closing the deal. Create your free account today at www.practancyorbit.com.

I think where AI is really getting adopted right now is in the in the operatory which is a crazy place to start off from because you know a lot things are happening very fast. Dentists do doing so much. There's multiple people involved in an operatory with a hygienist and sometimes the assistant as so it's a pretty interesting place where it started with as you mentioned, where we helped in providing more comprehensive diagnosis but more importantly utilizing AI to communicate with patients. And as you know treatment acceptance rate is a big challenge in dentistry because patients just don't understand their oral health. They don't know how to read X rays and then a lot of information is provided and for some reason they believe if you know if it doesn't hurt it's nothing is wrong. And and then this, this technology could help and did help and is helping them provide much more visual understanding that patients can easily understand and and then accept the treatments. So that's been the first I would say major use case that happened in dentistry. And now I think the next step here is around operations, revenue cycle management, administrative work. How do we remove the administrative work that especially around claims and especially around.

Pieces which require a lot of work and that can I can do pretty.

Well as well and does overjet focus more on the clinical, the diagnostic, the treatment planning side, or is it more on the practice management, revenue cycle management, process, workflow outside of the operatory or both?

So we sit on top of a practice management system. So we have, think of it as we have three offerings right now. One is around smart imaging, so people have their imaging software. So you can actually take that imaging software and replace it by the smart imaging software, which would have AI natively from the image capture all the way to the findings, which can. And visualizations which can be used for patients, really saving people time and cost because you don't have to pay for multiple softwares in the practice. And then second is just the AI. If you want to use your own imaging software, if you want to use your own practice management software, you can just get the AI which overlays on top of that, takes the data from the practice management system, takes the data from the imaging software, provides findings for the dentist so that they can, they can make the decisions much more effectively. And then the third is around operations as well in revenue cycle management where we've introduced review pass and ability to press a button, skip manual insurance review and get a decision in real time at the treatment plan level so that you can share the cost with the patient and patient can accept and does not get surprise bills later on and then insurance verification as well. So we are actually operating on the clinical and the operation side currently. And that's where the uniqueness comes into play as well. Because even revenue cycle management in dentistry is very clinical data heavy compared to some of the other areas. And by being able to understand what that clinical data is, you are able to maximize your or optimize your operations as well.

Going to walk us through a use case, a doctor signs up for the, let's say the full suite of overjet, all of its AI features. When is the first point that the doctor is using or in the patient experience is overjet being used? And then just walk kind of step by step through the patient experience and explain how overjet is laid inside of that process from beginning to end.

That's a really good question. So from the beginning, usually in practices, three days before there is an insurance verification check that happens. So say I'm the patient, I schedule with the, with the dentist, the dentist checks or then the staff checks, whether there is, you know, whether my, I have insurance and whether I need to, they need to collect the data, etc. And that, that can be automated.

Where all the insurance checks are happening automatically and the information is being Provided for.

For the patients. Then as a, I don't see that as a patient, but that, that can happen. And then as a patient, when I walk into the dental office and I get an exam, that's where the patient sees overjet for the first time. Where as the dentist pulls up.

As the X rays get taken and photographs get taken, information.

Is provided, then the patient pulls that up. Sorry, the dentist pulls it up for the patient. And as a patient, I interact with it.

And I think the word being utilized is called co diagnosis with the patient. Right. You're showing the patient what the disease is, what the treatment is needed, and the patient understands more and then accepts the treatments as well. So think about, for example, periodontal disease. Most patients don't know what periodontal disease is. It doesn't hurt. So when the dentist says you need deep cleaning to the patient, the patient is confused of why. And they think, you know, just the dentist just trying to make money because nothing is wrong in my mouth. And that is such a serious condition in terms of, you know, we know periodontal diseases linked to heart disease conditions, diabetes, etc. It's an infection in your mouth that has to get treated. But with overjet, you can actually see bone levels, red, green, yellow, you know, depending on how large they are. It's quantified. So it's not just my opinion versus your opinion. The patients can see before, say it was 2 millimeters, now it's gone to 3 millimeters and just growing. And your, your bone is degrading. And that, that is something that helps patients understand more so that they accept the, the treatments as well. The other place where it will, you know.

Let me just walk through that journey and then I'll come back again. So then patient, as the patient goes to the front desk and says, hey, yeah, I'll get this treatment scheduled. The front desk can press a button, get the insurance company information back, saying that they will cover it, and then they can let the patient know that, hey, your insurance company will cover half of it and you will need to pay the rest. And as a patient, I walk knowing, and I set up, set the appointment knowing what I'll owe rather than getting a surprise bill later on. So that's the, that's what? Overjet. You know, if overjet is being used in the practice, in the back end, there's a few more things that, that the, the product is doing, but more for the staff. For example, people are using this also for morning huddles. So you have this, what we call the daily huddle page where you have all the patient information listed, all the findings, things that, you know, extras that should have been taken, were not taken, barrier charts that weren't taken, etc. So that you can have one view of understanding what needs to happen that day. But that is another thing that the patient might not see but really provides value for the practices. And then we have different lists for patients who've missed coming to the practice but might need a certain treatment as well.

Great. So follow up questions on that process. So client or patient, sorry, patient calls in and AI can help verify their insurance ahead of time, faster. So that's more of a process practice management step that AI is sort of condensing the time frame to do that. They come in, they get their X rays taken and overjet will create a much clearer picture that is a lot more understandable for a person, a non dentist like me, to understand and see what is the need or what is the issue clinically that I have. And visuals are just another sensory that helps us understand something. In this case it adds to what you're communicating. It adds this nice visual to help with understanding and by extension case acceptance. And then it can help with the, the, the billing. Is it helping with the revenue cycle management as well?

Not with the billing yet, but it is helping with presenting that treatment plan to the patients, knowing what will be covered and what won't be covered.

Got it. Okay. And in the operatory does overjet take notes of the conversation and summarize the clinical notes? Is that at all involved?

Oh, we're not doing that right now.

Okay. And talk to me about the radiograph analysis. So part of it is this helps patients understand the clinical need. Is it also helping the doctor to diagnose as well or is it more for the patient's understanding?

It's also helping the diagno the dentist diagnosed because if you think about it, you know, first the dentists don't have that much time to, to spend on the diagnosis. They're usually doing another treatment when the hygienist comes in, makes them go to the other office and very quickly while the patient is waiting, they have to like really quickly look at the extras, which might be 18 extras taken, all the peri charge information and then not only make the, the diagnosis, but also communicate with the patients in that time, you know, and, and the less time people have, the more things they will miss, you know, just in general. And there's been studies on that as well. So for example, periodontal disease is very underdiagnosed in, in dentistry there, there are a lot of other conditions which are underdiagnosed as well. And with this they're seeing a more comprehensive image. So it's, it's similar to like if you had a very good hygienist, where the dentist walks in, the hygienist already has pointed out the things and the dentist is like, check. Ensures that you know what they're saying is correct. Similarly, when the dentist walks in, the AI has already viewed everything and has highlighted things. They of course take a full glance at everything, but then they can really focus on the areas that have been highlighted.

Does it give a narrative, does the AI give a narrative of kind of a first take treatment plan and then the doctor will take that initial analysis and then of course apply their own judgment and further.

Yeah, so it has recommendations.

Great. You know what this reminds me of a little, a little bit is there have been a few times where I will get a financial statement, which is my X ray is a financial statements and tax returns or payroll reports and I can redact them, take the name off and I can put it into a ChatGPT or another AI model and ask for it to analyze it's thoughts financially on the health of that practice. And although it, it doesn't come near what, at least where, where, where I'm at having been done this, doing this for 15 years and really analyzing a doctor's business and their, their labor costs, their operating income, their profit percentage, their sort of debt structure, their tax planning, that kind of things. But it does give me a decent kind of first take at it to at least start with. And I think in time AI is going to come into every business. Probably at some level it's going to come into my space, especially where there's a lot of numbers and numerical analysis and it can reduce, I don't ever see it eliminating a role where you're, where you're integrating so many variables of a person's financial life, a lot of which are non quantifiable, the emotional side, the psychological side, people's relationship to money and everyone's very different there. And that's why I think the way I view AI is that I can help reduce the time and the costs of getting and preparing the analysis to then hand it into my hands, taking the baton to do a complete and full analysis and talk about applying my clinical recommendations around money for the doctor. I see it similar here in that it's condensing the amount of work time and costs needed to do a really Good treatment planning, case acceptance process with a patient. One thing I did want to ask about is you got FDA clearing. Talk to me about what that means exactly.

Yeah. So the way the FDA clearance works is pretty interesting. Where you need to prove for these devices, it gets labeled as a software, as a medical device. So our product is a medical device in a software form and then because it's, you know, it's aiding in the diagnosis and can impact patient care. So you there, there are standards that you have to meet and what the FDA guidelines here are, that not only does your AI need to be as good or better than, than humans, but more importantly, when humans utilize the product, then they need to be better. So what you need to do is what are these clinical studies where you basically have dentists? First you compare your AI compared to a dentist and you determine how good that is. And not only one dentist, you're doing multiple dentists so that you can get, get what, what they call ground truth. And then you have after that then the, the, the dentist you do the same diagnosis with utilizing overjet and, and AI and see how much better they are at, at at the diagnosis and compared to the ground truth. So these are. So there, there is many dentists that get involved in these studies and, and they go over months because then you have a washout period. So like they don't REM before. So you have to wait some time and then you have to start again. So it's a pretty interesting way that this happens. But.

The interesting thing is the bar that FDA has set.

You have to cross the bar that the human has to be better with utilizing AI. So with that.

The good thing is when we go into clinical and into a practice, the dentists, it's not about me versus the AI. It's oh, I'm better with the AI together.

Do you ever feel lost in your practice finances? And how about your personal finances? Ever wonder where your hard earned money goes and why you're not seeing the financial progress you've worked so hard for? If so, listen up. As a dental CPA and financial planner, I've created our company Practice CFO just for you. With almost two decades of experience, Practice CFO integrates great CPA services like accounting and tax with customized financial planning, both business and personal. Think of it this way. Our dental focused CPA services give us the x rays we need to clinically plan your global finances and be effective financial planners. It's a highly customized approach we call the CFO model. Historically having a CFO or chief financial officer was an advantage of large corporations only. But today Practice CFO gives that same advantage to private dental practice owners. As a client of Practice cfo, you'll be assigned one of our dedicated CFO advisors. You'll receive insightful financial reports each month, have regular financial review meetings, reduce taxes, grow your wealth, and fill more in charge of your numbers. Because we believe that in the competitive landscape of dental practice ownership today, you have to be an effective business owner, not just a capable clinician. Reach out to learn more at www.practicecfo.com.

So FDA will, at least for the foreseeable future, probably indefinitely require a licensed dentist to interpret the results of the AI before recommending and performing any treatment on a patient. And the FDA has approved Overjet as a medical device that is that can be reliably used for dentists to do to do their thing as Dennis Am I understanding that right?

Absolutely. Yeah.

And when it comes to the data, I think most listeners at this point have used ChatGPT or Grok or any of the other ones and they know how to use it as a prompt to get or retrieve information. I, I've said this many times, I rarely use Google anymore. I may for certain things, but I am almost exclusively using ChatGPT in fact, on my, on my iPhone. You know, in the bottom of the iPhone you have your three or four kind of key apps that you use the most. That's sacred land right there. And I ChatGPT has made its way as one of the four, so I use it all the time there. And it's always so impressive to me how detailed and accurate it feels accurate. At least so many of these answers are. But in a very summarized view, what AI is doing is it receives a prompt, it goes into a large database, a large pool of data called these large language models. And it doesn't just retrieve specific text and paste them together. It actually organizes a completely brand new response based on interpreting from a lot of different sources and then it provides a synthesized output back to the prompter, back to me in a nutshell, that's what it's doing. Now what we're trying to do with AI is use it to create automation. So not just a super or glorified Google, but actually to do tasks that historically have required human thinking. And that's where I think the largest effects are going to be in the business world and in, you know, your business, my business, dental offices, is how do you use it to replace, I don't want to say replace humans, but to do some of the thinking that humans have done in the past. So when it comes to this clinical analysis of the radiograph, the input is, the radiograph itself is how I'm seeing this at least the input is like that's your prompt, you put it in there and it's asking the large language model to go into its large massive database and to pull out its analysis of that radiograph, saying what this patient needs based on its view of, of that. So my question is where is overjet pulling from what, what is its large language model that it's pulling from to do its analysis and recommendations?

Yeah, so I think the easiest way to think about this would be in the end, however it learns. Think of it as a large matrix, right? You have the X. You know, take a multidimensional matrix and you learning what the variables in the matrix are which represent the model of the world. When you're talking about like very large language models or the model of you know, dental X rays, if you're talking about smaller models, but basically we're taking real world data and condensing into the vector space and putting into a multidimensional matrix. Once you have this matrix, all you're saying is when you get an input.

And the input becomes another vector, so it's like a matrix as well. So an image becomes a matrix. It's actually math. So from a finance perspective it gets back to math anyways. You have take the image, you convert it into another matrix and you're inputting this matrix with the matrix of the world that you have and you're getting an output. So in the end, so there are the kind of models we use, we're not hallucinating stuff and making stuff up. We're literally determining what the output would be based on the matrix. So that means if you send the same image three times, you're going to get the same output. If you send three times, you prompt charge three times, you will get different answers because for them they're generating new outputs. So they're doing, they're predicting the next word. You know, that's and, and then they're generating every time there, there's a little bit of by purpose, on purpose they've created where it can create, you know, create stuff for us, for our needs. We don't need to be creating new kind of extras. We just, we, we could if we wanted to. But we've, we've think of it as like we fine tuned it to be able to put an X ray every time you put it give you the same output. And, and then you'll have, you know, another, you know, think of it as another out, which is the predictions on, on the X ray. And then you can visualize that and see how you know and see what, what's really present. So I think that's like the underlying way to think about these kind of where you're, you're finding things in images.

I, I'm just curious, this may be too granular of a question, but how is AI taking this image? I get how it takes text. It takes text and behind the scene, it, it converts them into these things called tokens. Is, is my understanding. And then it goes out and based on those tokens, it uses math and probabilities to understand what the prompt or the question is. And then it goes out and it, it assembles an answer and comes back and puts it back into text. How is it exactly converting an image into a language or into a series of tokens to then do that analysis?

Yeah, so an image, if you look at, you know, if you look at an image, how it's represented in a computer system, it's represented in, in numbers. Right. So good point.

Yep.

You know, think about the RGB that you have, you know, you have it in a matrix like X, Y of the image and then the RGB will tell you, you know, the, what the color, et cetera is. And, and if you have, so those numbers get tokenized, if you want to put it as tokens and put it into the system. So it's basically, you know, think of it as a matrix and you could have, you could have just fed it 1, 0, 0, 5, you know, like all the numbers anyways, but that, that's how an image in a computer system is represented as numbers. And those numbers then get fed into, into the system.

Does it ever hallucinate?

So it depends on what, what we're trying to do. Right. So if we are trying to create a narrative, say if I was trying to do a transcription, it might. Right. But in the particular, when it comes to computer vision, when we're taking an image and we're getting an output, it doesn't hallucinate in the sense that we're, how we see like large language GPT hallucinate because you're again predicting the next word. We're not predicting the next word. We're not creating stuff. When you're starting to create stuff, does it get things wrong? Yes, it does. Right. So like I would take them differently or understand them differently and here it can get things off and like it's Humans aren't perfect at diagnosis. So it's, and the data that gone. And even though we've trained on so much data as well as, you know, we've had multiple people look at the same image and give, you know, come to a conclusion and, and have that input in it as well. But you know, X rays are like especially incipient carious lesions. You know, they're, they're, they're hard to understand or say if there's a secondary caries underneath a large restoration that's, you know, some dentists might say there is something that might say there, there isn't. So even in our training data set there might be some errors and because of that, you know, you of course over large data they should reduce significantly and that's why the models are so good. But there's always like some error that that's going to be there.

And when it's pulling from the data and it's doing its analysis, how much of it is, is you with all of your users? This is the amazing thing about AI is that it, it, it self learns in some ways. And as you gather more data, more X rays from more dental offices and it goes into this, this, this LLM, this, this model is that how overjet is constantly improving itself is by aggregating the data from multiple dental offices in order to continually and incrementally improve its output clinically, or is it going out to sort of the general LLMs to do that, explain how you're building that language.

So we're utilizing data. We also have, we employ hundreds of dentists who are cleaning up our data and improving it and making it better than what you would just find in a dental practice. You know, really identifying, outlining things on these X rays and that helps you get to the kind of accuracies we need. You know, as you mentioned, if you put your financial sheet into, you know, a GPD model, you know, there, there are like sometimes it might say 82 when it should say 84. You know, like, and while we have to get really, we have to get really good at like better than humans, you know, better than not just any humans. We need to get better than like dentists who are specialized in this stuff. So you, you need to get very good and, and, and you know, so all kinds of techniques are used to make that happen, happen.

Does it learn from the doctor, her or himself? When the doctor will take the overjet analysis of the radiograph and give its kind of recommendations or its analysis, is there a way for the doctor to add some comments or notes for the AI to say, okay, thank you for adding that. I see how you're now interpreting this and then factoring that into future.

Cases.

Yeah. Similarly to charge, you might see of like a thumbs up, thumbs down. How did you like my answer? That's probably the best feedback that you can get. If you got something wrong and somebody tells you that you got something wrong, that's the best feedback that you can get and feed that into the model. So we have what we've provided and we've done it by design. So on every finding, even at every surface level, like we allow you to delete, you know, so for example bone, bone levels, we do it on mesial and distal and you can actually remove like the mesial one and say I disagree with it and that we can take back and learn from and then come back and the molds will be much better. Also it won't show in your system next time. Right. Like you open it. So it's valuable for the dentist to do so so that the next time they open it they don't see something that disagree with. But it's very important for us as well because then we can say, okay, there, there's something wrong that happened. It might be not wrong, maybe they did it by mistake. So there's actual review that happens internally with people and, and, and then, so we don't just like create a self learning system which can just go haywire. You know, there, there's a check that happens, make sure that everything, you know, if we use the new data, we get better than where we were. If we're not getting better, it's not going to go in.

And so you, you have a team. So when it comes to training the model, I did a podcast the other day and the guest said that when you adopt AI, it's a little bit like hiring an employee and you have to train it, there's some training involved in it. So it, it kind of learns your, some of the nuances about your style, your approach, your process, whatever that is. When it comes to the training the AI model, it, it sounds like you've got some members of your team that will take feedback from users and help train the model to make it more and more intelligent. Is that how the shaping and the incremental improvement as a thinking software, as a thinking tool is improving? Is it through human training it, or is it self kind of learning automatically from feedback from the users?

So I would say the feedback of the users is going, going to humans and the humans are helping it. Even with, if you look at ChatGPT, the way it got there was reinforced learning human feedback that went into it. So when outputs came out somebody had to say hey, this doesn't make any sense and it went back into it. There are newer techniques where, but when you talk about again the kind of accuracies that we're looking for, the kind of precision we're looking for, we're not saying hey, make me any kind of cat, right? Like we're saying, you know, for this patient, identify and outline where diseases and you know, do it as accurately as possible. So that, that's why, you know, it's like the best way you can train these models as well. There is, there is some other techniques with more self supervised learning stuff that's happening as well. But there is also human competent to, to it.

Yeah. Even with ChatGPT, periodically it will give me two answers and ask me to choose which one I like better and that it's learning from that as well. So I, I imagine it's something similar to that in the way that you're continually honing this thinking engine known as known as AI in your software to become better and better and better.

Humans are very smart. So you can, if you can get your feedback that that's really good.

Let's now move into the treatment is done and delivered and now we're submitting a claim and I think if I heard you right, overjet is helping that.

The notes or the claim to be more accurate and quicker. Did I understand you right?

So I would say, you know, by the end of the year we'll have more of those features. But currently what we're doing is, you know, if you went to the market like went to a, you know, and said like buy, what do you get? We are helping get an answer of basically doing an insurance check for you on the, the data data and you can do it at the treatment plan level. You don't even have to wait till a claim is formed because that's, that's probably sometimes too late. You've already done the treatment. Right. So you want to understand and give that estimate to the patient. So it's not the pre treatment estimate.

Got it.

Provide a pre treatment estimate to the patients so that there's more clarity there. They accept again they know exactly what they're going to pay so that they're not confused. Like you probably, you know, will not deny a treatment just because you know it costs you $1,000 but many patients can't afford it. So like knowing that, hey, 500 will be covered by insurance.

They will go ahead with the treatment and that and not getting a surprise bill later on and trying to figure out, but if they can, they like how do, how do they get a payment plan or whatever that might be so that they're not getting a surprise bill and then going into collections.

Got it. What, looking ahead, what's, what's coming down the next 6, 12, 18 months for overjet.

So we're really focused on revenue cycle management. So all the things that you were saying are coming. Our goal is how do we help dentists do what they do best, which is providing care. And, and then how do we remove all this administrative overhead such that it's all happening good, it's happening fast, it's happening correctly, and, and then they can to, you know, spend more time with patients and their staff can spend more time with patients. You know, when the patient walks in, they can understand the patient more rather than trying to fill all this paperwork.

Yeah, the way that I've visualized AI in dental offices is there's, there's AI that's happening inside the operatory, which is around the clinical side, treatment planning, case acceptance, presentation to patients, clinical analysis and that. And then there's everything outside of the operatory, which is from a, from a chatbot on your website, an AI chatbot, for example, to insurance verification, to submitting the EOBs, dealing with denials, managing your revenue cycle management. In marketing, there's AI making inroads there. I mean, AI is really coming into a lot of different places inside of a dental office. How much do you think in the future that dental AI companies are going to sort of focus on their, their specialty? Ours is in the operatory and we focus on clinical only, or we over here focus on the front office and automating front office tasks there versus how much and if, and if that's the case, and let's say there's two, three, four AI companies that a dentist hires or sort of signs up for and pays whatever the fee is or the subscription for those, is that going to create conflict or integration issues? Could it add so much to the tech stack and the dental office that it almost becomes confusing for everyone to learn how to use these different AI based technologies? Or do you think that there will be one, two or three that end up ruling them all, so to speak? An Amazon style of AI and dentistry that kind of handles front office and the clinical office. Where do you see it evolving from that standpoint?

I think there's going to be two things. There's going to be the practice management system and think of it as a system of record, right? And then you're going to have an intelligent system, you know, where you can think of it as overjet. And, and that intelligence system can do many things because the way you understand an image is the same way that you understand text. Now these used to be completely two different fields. You know, when, when I started in the field, when I was in school, there, there was a computer. You either did computer vision or you did natural language processing. These were two separate areas. But very quickly with transformer models came out. That's the underlying technology for GPT. The T in the GPT is transformer. And that's the, when those, when that architecture came out, these two things merged and interestingly enough, even for now, like, you know, so it's, it's, you know, if you had asked me like say six years ago or something, I would have said, oh yeah, that would be hard for a company to like, you know, kind of built capabilities on both sides. But now I would say they've merged so much that, you know, you can go either way very easily. So it's almost. And you and the tools. Tooling has gotten so much easier right now. Building AI gotten so much easier. So I think you are going to see that convergence happen. And then there is the consolidation of software happening in general. Right. Like we, as a company, we don't want to have like, no, I think we have like 60 software. So we utilize in the company. Right. Like how many softwares are we going to have? If any of our vendors can do multiple things, even if there's like a 10%, you know, improvement that we might have, we don't want to deal with the billing, separate billing, we don't want to deal with the data loss, you know, not having all the data in one place as well. So for example, you know, your claims data can help your treatment plan data can help your claims data. So like these two things need to go together. Knowing more about your patient data, knowing more when you're doing a, you know, somebody calls in will be important as well. So like, if you can understand more data across the system, you can actually build better systems. So I do think it's going to consolidate. Also. Dentistry is not that big. You know, it's a $160 billion industry, but you're talking in a 3 trillion plus, you know, medical space. So it's a very small space, you know, compared to, and having companies really invest in R and D at that scale, it gets harder. So it is, you know, you're not, you know, I think there's going to be more consolidation that happens.

Yeah. I'll tell you what technology, I love technology. And it is, I feel sometimes as much of a curse as it is a blessing in, in a business. And I find where the curse is is not inside of a given software itself. It's the how to get different softwares to communicate to each other in a workflow that spans across different technologies.

And I'll be curious to see how this develops in AI. But what AI agents essentially are supposed to do in some ways is to be able to take data from one place, do thinking, do an analysis on it as a human brain would, and then produce a result that moves on the assembly down to the next point of that workflow, whatever that workflow is, which may go to a human, it may go into another software, whatever that, whatever that is. And this is less of a question, more just a statement that I'll be interested to see how these AI agents build into workflows to not solve one problem while creating a new problem.

Yeah. So the thing of even like for example, our insurance verification, Right. I can call it in the agentic world, the way you would state it is like there's an agent that's going onto the website, getting the information and bringing it back and putting it back into the practice management system.

Right.

Like that's. So it's not only doing the thinking behind it, but it's also doing the action that a human would have done and then bringing, bringing in the information back to and putting it in there. So I think, you know that that's at least. And then you of course can string them, you know, if X happens, then do Y and string them along because you don't want just a very long chain of. Just try doing a lot of prompt, you know, very, very large prompts. You'll see what the results are. You're better off seeing the result than doing another prompt and then seeing the results and doing another prompt.

You know, in my space, and I've thought a lot about AI in my space as a, as a dental CP and financial planner for, you know, for a very specific audience, which are dentists. One area that I always say, how can I use AI to give me more tools and data, to be a more effective advisor, thinking partner, coach and every. Everything here at practice CFO is about driving financial success for doctors. Everything is to drive that on a very personal level. And so I don't think that my clients really care what's happening behind the curtain on how we're preparing a tax return. And so when I think of it that way, I say, okay, well, how can AI come in and do work that isn't diminishing anything that the client is experiencing? In fact, how can I use AI to enhance what the client is experiencing? And so a clear example, just to put a visualization on this is one of the biggest challenges on the tax side is we get all the 1040 documentation, we get 1099s, we get W2s, we get K1s, we get all kinds of things that have to go on the 1040. And you got to get the documents, you have to organize them in a way, put them together that the tax preparer can. Can sort of make comments and that a reviewer can go in. And we call them tick marks and tick off everything on these different documents that are then put accurately in our tax software called Ultra Tax cs. And then from Ultra Tax, of course, we file the return, we e file the return and we send it to the client's portal where they can sign and see, see the tax work. Well, one area that I'm now seeing AI come in is if you ward off and if I were doing your tax work, you could upload all of your tax documents in whatever order you want. You could combine them in one PDF, you can send them piecemeal, you can mail. Hopefully you're not mailing them, but if you were to mail them and we scan them as one document, and then I put it into this software, one that I'm looking at is called Stanford Tax, and it integrates with our Ultra tax software. But it will take all those documents and it will see from the prior year what the client had by way of 1099s and K1s and W2s, and it will read through it and it will automatically organize the documents and then populate the tax software with that data. And for my clients, I don't think they care how it gets in there. What they care about is, is it then analyze to figure out how to optimize the tax savings. And that's where AI I think, will absolutely play a role in giving, just like it does on a radiograph. But I think it'll be the same way. It will dish me out sort of an initial analysis of what the AI thinks are some strategies, such as doing Roth IRA conversions, setting up 529 savings accounts, maybe timing certain things around itemized deductions. But at the End of the day, I still think, and I don't know if this will ever be fully replaced, a human mind who knows the client, knows what matters to them, knows what they're sensitive about, knows their kid situation, knows what their goals are, can then overlay that context onto the data itself to produce the ultimate clinical or the kind of final clinical treatment plan around their money. And that's where I see the relationship of AI and the human involvement coming together to produce a better outcome for clients while creating some efficiency for the clinician, in this case the financial clinician, me, allowing me to have more time doing the, doing the critical thinking, doing the coaching, and less time having to organize documents together and do that behind the curtains. Things that really don't provide any perceived value to the client. So that's a little case study as the way that I'm seeing it. All right, with that in mind, let's go into here. My final question. This is going to be a two part question. Okay. What are some of the risks that you're seeing in AI adoption in dentistry? That's number one. What are risks? Because I would still say that we're in this early adopter phase of the AI lifecycle inside of a dental office. I know among my clients it would be relatively few of them who are using an Overjet or a Pearl or some of these other AI companies. But there are early adopters and you know that there are some people who love tech and they are the ones who are willing to take that first iteration in MVP1, that first iteration of a software or a technology to play with it, to try it, trial and error. And then they give a lot of feedback to the developers and the tech people to make it better and better. And then at some point it gets so good that you get more of a mass adoption to it. Maybe. I'm going to parse this out in more questions, but where do you think we are in the adoption phase of AI in dental offices? Do you think it's 5%? Do you think we're at 30%? Are we still in the adopter period?

No, we're still in the adopter period. And I think you nailed it. You know, and here I think the risk I see is, which is like with some other technologies as well, like full self driving, for example, right? Like you're a dentist, you're still responsible for making the right decision. AI is an input that you want to take, right? You just don't want to be like, okay, everything they say I'm going to do and that's something that hopefully with the new generation, that's why we are going in dental schools as well, like teaching them about AI etc, rather than just saying, hey, even in Tesla, do not just, you know, go to sleep, keep your hands on the, on the steering wheels. Similarly with AI, it's an input that'll help you and for you as well, you know, when you're, you're going to use your AI software, just do not just submit it, just make sure everything is checked. So I think it applies across the board, not only for dental, that you guys, you know, we're still in the, in the driving seat right now and we need to make sure that the patient's responsibility is on the make sure that we're doing the right thing for the patient.

Got it. What do you think are the risks then? If we're still in this early adoption.

Period, that's what I mentioned is like the risk is people being like too comfortable and saying, okay, I'm going to, after it gets it right, like five, you know, eight times, nine times, you're like, okay, I'll just accept it. And just like, you know, rather than saying, oh no, let me still check it, you know, because that's how I would want my gear to be and, and that's how the gear should be.

That makes sense. And so the analogy back to the self driving car, for example, you start to get really comfortable because it's been driving through plenty of lights and making turns the right way. And then suddenly, suddenly it, I'll say hallucinates, but it misinterprets something in its, in its field of vision. And next thing you know, somebody's hit or there's a car accident happening. So that's on a, metaphorically. That's kind of what could happen in a misinterpretation.

No crashes happen because we're in a very low, like lower risk. You know, the dentist still has to drill with, with their own hands. Right. So hopefully they're looking at it and just checking and saying, making sure this is, this is the right thing. Right. And that's what the clearances are for as well. So you know, the dentist is still making the decision and just like having that, not getting too reliant on it, where you just like, hey, you know, I, I don't, I don't even need to look at what, what, what was really there.

And what would you tell a dentist who's still hesitant? Maybe they're, you know, 50s or 60s, they're maybe not using ChatGPT, they've heard about AI, but still pretty hesitant to adopt it or experiment with it. What would you tell that, doctor?

Yeah, I think it's like free will. So people can do, if they don't want to do it, don't do it. But for people who want to, like, try it, it's easy to try and like get started on and provides a lot of value. But again, everybody's different, right? Like, it might not be for everyone. Hopefully, you know, the good thing is the patients will benefit. You will see the first time you show it to a patient, the person like, oh, now I finally get what you have been saying. You know, it completely does change, but I think, you know, people have to experience that as well. And over time I think it's going to go, you know, whether it takes a little more time, but it's going to be everywhere.

And if I adopt it, if I say, let's do it, overjet is, I'm all in. What is the onboarding or setup time and onboarding process look like?

So we have on our website, we have a demo. You can even call, you know, and schedule a demo, see how it all works. So you can't, you know, just don't take my word for it, See, see how it really works. And then you can get it installed in your practice. And it should be, I think it's within like our average time to value is like in eight days or something. So depending on if you're ready to schedule very fast because you need to, you know, the I sometimes because the software is, is installed in the server room or on the server. So sometimes dentists don't know about that. They, they need the IT to do it. So if you can schedule the IT and you can, you know, get the, the training, you can go in, you know, in a few days.

Great. And you have support, as they would need support. Possibly you have just a call in that they can call in and get technical support, right?

Yes.

Great. You know, my final comments here as we wrap up this episode is that I think there are advantages and disadvantages to being an early adopter. The disadvantages, the software may not be as perfect and polished as it will be two years from now, but the advantage, which I think significantly outweighs the disadvantage, is that by starting early and using these softwares, even if it's a little bit experimental at first, you are going to be ahead of that curve. So as new improvements come out, you're going to be the one who's speaking this language of how AI is working in A dental office. And if you stay ahead of that curve, it's going to provide a competitive advantage for you as a dental office. And it may take time for you to figure out exactly how to layer it into your processes to get your teams trained on using it and, and therefore using it to do things faster, better, less expensive, and ultimately a better clinical outcome.

And doing it, for example, like, you know, maybe I'm aging myself, but like I, I look at Snapchat and I'm like, I don't even know what this means. Right. So you don't want to be in that position where you kind of, you know, did not get on the technology curve. And then after a few years you're like, I don't even know what's happening now. So with that, you know, like as you've used ChatGPT, it is pretty, you know, the value, even if it's experimental, the value of it is much greater than the improvement that, you know, it's good enough. And I would say where our AI is even further out where it is even FDA cleared.

Right?

ChatGPT is not FDA cleared for medical diagnosis. We are FDA cleared to help in the medical diagnosis and that that's even a higher bar to cross.

I think it's great that it's in the dental schools because then they learn from the ground up. They learn the process with AI embedded in the, in that process. And so it's not even about adopting it, it's just their way of, of doing a clinical treatment planning. So that's pretty exciting to hear. I know, I know the big institutional money, the DSOs, the PE backed dental corporations out there are very much investing into this as one would expect from a well run business, kind of a larger business. And so I love the fact that companies like Overjet are coming in and saying, hey, private dental office. And I know you can support the DSOs as well, but the private dental office can get the same advantage. Implementation, I don't think is very difficult with these technologies.

No, absolutely, absolutely not. But thank you, thank you for Wes for having me here.

Yep. Thank you for being on the show Ward Eye. It's been a total pleasure and excited to see where Overjet goes and help it, how it helps dental offices as we go.

Take care. Thank you so much.