0:00

0:41:00

How AI Tools for Therapists Are Built: Trust, Data Privacy, and Choosing What to Adopt – An Interview with Ian Knox and Megan Toomey of SimplePractice

Apple Podcasts Stitcher Spotify Subscribe on Android Download Leave a Review Keep me updated!

How AI Tools for Therapists Are Built: Trust, Data Privacy, and Choosing What to Adopt – An Interview with Ian Knox and Megan Toomey

Spend five minutes in a therapist Facebook group and you will find two camps: clinicians convinced AI is coming for their jobs, and clinicians quietly relieved to have their evenings back now that a note taker drafts their progress notes. Somewhere between the panic and the hype is the harder question most of us never get to ask: how is this stuff actually built, and how much of it can we trust?

Curt and Katie pull back the curtain with two of the people who build it. As part of the show’s AI month, and the show’s partnership with SimplePractice, they sit down with Ian Knox, Chief Product Officer at SimplePractice, and Megan Toomey, Senior Director of Product Management, to move past the buzzwords and talk about what happens behind the scenes: how AI tools get engineered, what “training the model” really means, how client data is handled, and how therapists can decide what to adopt without handing over their clinical judgment.

Transcript

Click here to scroll to the podcast transcript.

(Show notes provided in collaboration with Otter.ai and Claude AI.)

About Our Guests: Ian Knox and Megan Toomey from SimplePractice

Ian Knox brings deep experience in product leadership, having previously led global teams at Expedia and Microsoft. He leads SimplePractice’s product strategy with humility and a growth mindset, creating tools that help clinicians focus on improving client care and spend less time on administrative burden. Outside of work, Ian enjoys time with family outdoors in the Pacific Northwest and is an avid triathlete. He also traveled around the world after college, making him a great source of inspiration for future bucket-list trips.

Megan Toomey is a product leader with over fifteen years of experience in tech. She is currently the Sr. Director of Clinical Support AI Product Management at SimplePractice, where she manages a team focused on delivering AI-powered solutions for behavioral health clinicians, helping them drastically reduce the administrative burden of providing care while giving them the tools to improve overall client outcomes. Prior to her work in health tech, Megan built and scaled major product initiatives at organizations including Microsoft and Amazon. Outside of work, Megan has been a Lady Gaga fan since day one (from The Fame to Mayhem) and has seen her in concert eight times.

In this Podcast Episode: How AI Tools for Therapists Get Built, Trained, and Adopted

Ian and Megan walk through how an EHR actually develops AI features, starting from a map of the everyday jobs a clinician does and asking where AI is mature enough to help. They get specific about what therapists should evaluate before adopting a tool, including vendor trust, HIPAA compliance, and data practices, and explain why note taking is the most developed use case while insurance, scheduling, intake, and referral matching are still emerging. The conversation also covers what “training the model” does and does not mean, how transcripts are retained and the new opt-in for de-identified data, the fear that AI-first companies want to replace therapists, and the best practices Curt and Katie have been pressing all month: transparency, consent, and reviewing anything you put your name on.

Key Takeaways for Therapists: Adopting AI Tools, Protecting Client Data, and Keeping Clinicians at the Center

“AI used the right way can be hugely beneficial for clinicians, but it requires trust, it requires a lot of control.”

— Ian Knox, Chief Product Officer, SimplePractice

Start with trust, then HIPAA and data practices. Before adopting any tool, get comfortable with the vendor and exactly how it handles your data. A “HIPAA compliant” label is not proof of a strong security posture.
Match AI to your weak spots, not the hype. Not every problem needs AI. Note taking is the most mature use case today; benefits verification, scheduling and utilization, intake, and referral matching are still emerging. Megan’s test: where could you use “a Robin to your Batman”?
Whatever you sign, you own. Anything you put your name on, you are responsible for, so review every AI-drafted note and do not assume it caught everything. SimplePractice prompts you to confirm you reviewed a note before you can sign it.
Transparency and consent come first. Be transparent with clients about how AI is used in their care, and get their consent before using it.
“Training the model” is not what most people think. SimplePractice says it does not train an underlying large language model on client data. It uses a private model and improves results through better prompting and a clinician-graded “golden data set” of strong notes.
Know your data’s lifecycle. With Note Taker, the transcript persists only briefly and is deleted once you sign the note. Retaining de-identified, decoupled transcripts to improve the tool is a separate, opt-in choice.
Expect the tools to keep changing. Capabilities move month to month, so a feature that disappointed you six months ago is worth another look.

“Short answer: we’re not training an AI model. That is not happening.”

— Megan Toomey, Sr. Director of Clinical Support AI Product Management, SimplePractice

How SimplePractice Thinks About AI: Four Areas

Ian described the product roadmap as four buckets. At a high level:

Clinical care and intelligence. Note taking is just the first step; treatment plans, measurement-based care, and client and intake summaries come next.
Insurance and billing. Benefits verification, cleaner first-time claims, and handling denials and rejections in the background.
Practice operations. A “business advisor” view: analytics, scheduling, and catching issues early for clinicians who never trained to run a business.
Referrals and caseload. Not just any referral, but matches that fit the clinician, including links to payers and health systems through Therapy Finder.

Listen to the full episode for how they decide what to build next, and what they mean by working with a “white label” LLM under the hood.

Resources on AI in Therapy, EHR Tools, and Data Privacy

We’ve pulled together resources mentioned in this episode and put together some handy-dandy links. Please note that some of the links below may be affiliate links, so if you purchase after clicking below, we may get a little bit of cash in our pockets. We thank you in advance!

SimplePractice: simplepractice.com
SimplePractice Trust Center (privacy, security, and data policies): simplepractice.com/trust-center
Reach the guests directly, as offered on the episode: Ian Knox, ian.knox@simplepractice.com; Megan Toomey, megan.toomey@simplepractice.com
Concepts discussed: HIPAA and HITRUST, the “golden data set,” de-identified transcript retention (opt-in), prompting versus model training, and augment versus replace

Relevant Episodes of MTSG Podcast

Meet the Hosts: Curt Widhalm & Katie Vernoy

Curt Widhalm, LMFT

Curt Widhalm is in private practice in the Los Angeles area. He is the cofounder of the Therapy Reimagined conference, an Adjunct Professor at Pepperdine University and CSUN, a former Subject Matter Expert for the California Board of Behavioral Sciences, former CFO of the California Association of Marriage and Family Therapists, and a loving husband and father. He is 1/2 great person, 1/2 provocateur, and 1/2 geek, in that order. He dabbles in the dark art of making “dad jokes” and usually has a half-empty cup of coffee somewhere nearby. Learn more at: http://www.curtwidhalm.com

Katie Vernoy, LMFT

Katie Vernoy is a Licensed Marriage and Family Therapist, coach, and consultant supporting leaders, visionaries, executives, and helping professionals to create sustainable careers. Katie, with Curt, has developed workshops and a conference, Therapy Reimagined, to support therapists navigating through the modern challenges of this profession. Katie is also a former President of the California Association of Marriage and Family Therapists. In her spare time, Katie is secretly siphoning off Curt’s youthful energy, so that she can take over the world. Learn more at: http://www.katievernoy.com

A Quick Note:

Our opinions are our own. We are only speaking for ourselves – except when we speak for each other, or over each other. We’re working on it.

Our guests are also only speaking for themselves and have their own opinions. We aren’t trying to take their voice, and no one speaks for us either. Mostly because they don’t want to, but hey.

Join the Modern Therapist Community:

Linktree

Patreon | Buy Me A Coffee

Podcast Homepage | Therapy Reimagined Homepage

Consultation services with Curt Widhalm or Katie Vernoy:

The Fifty-Minute Hour

Connect with the Modern Therapist Community:

Our Facebook Group – The Modern Therapists Group

Modern Therapist’s Survival Guide Creative Credits:

Voice Over by DW McCann https://www.facebook.com/McCannDW

Music by Crystal Grooms Mangano https://groomsymusic.com

Transcript for this episode of the Modern Therapist’s Survival Guide podcast (Autogenerated):

Transcripts do not include advertisements just a reference to the advertising break (as such timing does not account for advertisements)

… 0:00
(Opening Advertisement)

Announcer 0:00
You’re listening to the Modern Therapist’s Survival Guide, where therapists live, breathe, and practice as human beings. To support you as a whole person and a therapist, h ere are your hosts, Curt Widhalm and Katie Vernoy.

Curt Widhalm 0:16
Welcome back, Modern Therapists. This is the Modern Therapist’s Survival Guide. I’m Curt Widhalm with Katie Vernoy, and this is a podcast for therapist about the things that go on in our practices and the things that go on in our profession, and we are smack dab in the middle of AI month here on the podcast, and we are looking at all of the different aspects of how we are coming into AI work. W e want to also kind of dig in and look at since this isn’t a passing tech trend, we want to talk with the people who actually know how artificial intelligence is being built and moving past all of the buzzwords and pulling back on the curtain of how this stuff is being engineered. How it looks behind the scenes. So we are joined today by Megan Toomey and Ian Knox from our friends over at Simple Practice, and they’re going to help us understand how all of this kind of stuff works. So thank you very much for joining us and being a part of the show.

Katie Vernoy 1:19
We are so glad to have you here, and to have this conversation with you as part of our partnership with Simple Practice. But before we jump into the conversation, I want to ask you the question we ask all of our guests, which is, Who are you, and what are you putting out into the world?

Megan Toomey 1:33
I’m Megan, on the product team at Simple Practice, which was only part of my life. Mostly, I’m a mom to a five and three quarters very specific year old girl, wife, daughter, sister, and god, pal, to many. B ut probably I’m one of the most passionate fans of Seattle sports. Someday, someday the Mariners will get the World Series. B ut I’ve been working in technology for my whole career, and I found that often in tech you look for positions with interesting problems that need to be solved or challenges, but you might not be that passionate about the subject matter itself, so you’re chasing the interesting opportunity and maybe not the subject. And I think it can be a rare opportunity to work at a tech company where the problems we’re solving are hard and interesting, but the work is mission driven, and to me it’s a privilege to be in this position with Simple Practice.

Ian Knox 2:25
Yeah, hi everyone. My name is Ian Knox. I’m the Chief Product Officer for Simple Practice. As Megan said, also have a lot of family stuff. I have three kids, a bit older than Megan’s now, all in high school. I’ve been also in technology for a really long time, worked at places like Microsoft and Expedia, but been at SimpleP ractice for the last four years. I n terms of AI, i t’s really interesting, even in my dissertation at Computer Science years ago was about AI. So this has been around for a long time, but obviously accelerating quite a bit. I would say, in terms of putting out into the world, you know what I love is helping customers, and I think once I started investigating SimpleP ractice to join, I talked with a bunch of therapists and just such amazing people looking to transform other people’s lives, and so anything we can do to make their lives easier is like hugely beneficial to us, so I’m excited to be at Simple Practice and helping your community.

Curt Widhalm 3:21
Like a lot of EHRs SimpleP ractice has been starting to develop artificial intelligence tools. There’s a lot of different feelings that a lot of consumers have about this, so I would like to maybe start with a question around, can you give us a quick overview of what Simple Practice’s history of AI tools is, how it got started, anything that’s kind of surprised you along the way.

Ian Knox 3:49
Yeah, well, we’ve been, you know, watching this like everyone else and getting involved with the technology since, you know, that first quarter when Chat GPT came out, and people are starting to recognize, well, this has pretty big implications, and I think there’s been a huge amount of caution about AI and mental health. You know, over time, we’ve seen cases of AI encourage delusions, psychosis, even suicide, which makes us very cautious about the use of AI, and you know I think we all recognize AI chat bots are not mental health care professionals, right. B ut I would say AI using the right way can be hugely beneficial for clinicians, but you know the thing I’ve been most surprised about, Curt was just that the adoption, the rapid adoption of it, even when we started with Note taker, first, you know, I saw therapist people we talked to every day were just worried about AI, like, is it going to take that job? So, like, what’s going to happen? The second is mental health is way more sensitive than, like, Zoom recording a call and summarizing s tuff, so there was a lot of trepidation around that, and then lastly you have to speak with a client, right, to get permission, and there’s another sort of hurdle there, but what we’ve seen is the benefits in terms of clinicians being more present in sessions, in terms of all the time saved writing notes, and other things have been so big that people are willing to try it, right. And so we’ve seen a classic early adopter phase, which is now kind of broadening out. So I’m very hopeful about the use of AI, but it requires trust, it requires a lot of control, and hopefully we can get into some of those topics as part of this.

Katie Vernoy 5:38
What do therapists need to know when they’re adding AI to their practice? I know that’s a broad question. To be clear, I’ve added a lot of the SimpleP ractice AI tools into my own practice, so I’m asking more for the audience members who are cautious or just starting out and wanting to have a sense of what they should be thinking about, with the caveat, we’ve, we’ve done episodes on the ethics, we’ve done episodes on some of the basics, but if we’re looking at the nuts and bolts, what are the things that they should be paying attention to when they’re deciding which tools to incorporate into their practice?

Ian Knox 6:14
Yeah, I mean, I can obviously start with that. I think, like, the most important thing is trust, right. W hen you’re using AI tools, you need to be sure about the vendor you’re using and HIPAA compliance and policies. I would like dig into that. There’s a lot of discussion on online communities and people exchanging notes, and certainly our customers ask a lot of questions, and so I think with whatever AI you’re using, just go make sure you feel comfortable with that company and what they’re doing. Then, secondly, I think it’s understanding the different use cases or different things you can do with AI and where you feel comfortable to start with. So, you know, there’s a few different buckets of things we see, like, the most prevalent one today is note taking, and that’s really where most people have experience today, but there are others that are emerging. T hings, like, you know, how you’re dealing with insurance. There’s a lot AI can do to help with that, even how you’re doing referrals and matching and building your caseload, as we can talk about in a minute, going from note taking more into this whole intelligent clinical care space is really incredible w hat AI can help with. B ut then one of the other things that I’m really excited about is a lot of therapists come in to running a practice and they’ve never run a business before, they don’t know how to, what should they be doing, what should they be charging, like what should they be watching out for, and AI is also really good at that, like summarizing where you’re at, what you should be looking at, if something changes, especially if you’re running a practice with a few clinicians, there’s a lot of things you can figure out. S o I would say, you know, it’s like looking at where you feel most comfortable trying, and then exploring the tools, and just getting a feel for how they work, right.

… 8:05
(Advertisement Break)

Megan Toomey 8:06
I think one thing I would add, too, is that AI can be a really great way to help you with things maybe you’re not as strong at. To Ian’s point, I say all the time, I’m like, therapist probably didn’t go get their MBA to run a small business, but we can help support you in that, in that way, and I think of AI as being a way to just like augment what you’re already doing and pick up the pieces when things are you’re not super great at. B ut like, how does AI help you but not replace you? And I think that that’s one way we’re potentially looking at different AI solutions, it’s like, where do you feel like you could use a Robin to your Batman, and the AI is really good at that.

Curt Widhalm 8:44
So, maybe diving into that a little bit more. How are some of the places where clinicians could implement or streamline some of their practices with artificial intelligence?

Ian Knox 8:55
Well, I think, like the way we look at anyway, is we look at everything that a collection does on a day-to-day basis, and then that can be cool. Jobs to be done, it can be cool where you just break down the things that people do. So we sort of have a map of that, and it’s everything from some of the things I mentioned earlier, like how are you getting referrals, how are you doing intake, how you’re doing scheduling and intakes, how you’re dealing with insurance, so once you break down all those things, really start to decide, okay, where’s AI at right in terms of maturity, the tools, and what can I go adopt? So, as we mentioned, Notetaker and some of the care stuff is more advanced right now, but we’re starting to see the ability for AI to do things like look at your calendar and figure out like what’s your utilization like or if someone cancels an appointment last minute how can that go be filled so you don’t sort of miss out on appointments during the week, t hings like where you’re doing verification of benefits or trying to u nderstand why a rejection or denial comes back. AI can be really helpful there. So, Curt, the way we look at it, sort of, all the things that questions do, and then usually where the most challenges is, I think, for a question personally, where a therapist is today, you should figure out, like, where am I struggling most, and then what tool might I try? Obviously, there’s things in EHRs like us that are within the platform, and they’re sort of add-ons you can look at as well, but it’s moving very quickly. I would say, you know, new tools are coming into play every few months, right now. So, I think that’s one of the challenges, just keeping up and figuring out, like, what to go try.

Katie Vernoy 10:36
When I think about incorporating AI into a practice, there are so many tools available, and there’s also a range of capacity that clinicians have to understand some of the technology, or to even be interested to engage in the technology, and my experience working with some of the AI tools has been mixed. I think there’s a sense of efficacy that I’ve found in, in using like a note taker or scribe that can help make sure that I’m not missing anything in the session, it’s it’s making sure I’m hitting all the points in my notes, I have presence, and sometimes there is time that I spend updating the template or giving feedback to the scribe, and making sure that I’m really getting the note that I want. By and large, I really enjoy it, and it’s a process that I think someone needs to know that they’re signing up for. I think some of the front, front of office kind of ideas, whether it’s a front door, a an ability to get referrals, or to help reschedule, or doing the things with the client; that sounds fascinating, and I also get a little bit worried about how clients will perceive it. A nd so those are just two examples, but when you’re looking at the types of tools that go into a practice, and maybe Megan, this is something that you can talk through, as, as you know, an expert in the products. Not everyone’s going to want every piece, and so when they’re looking at their own processes, what are the what are the types of, what are the pieces of advice you’d give to our audience members about how to choose what parts they’re incorporating into their practice.

Megan Toomey 12:23
Yeah, I have been saying the phrase, we are trying to solve common pain points with common and uncommon solutions. So, not everyone should use AI for everything. AI is cool, sure, it’s a fun, hot thing to build, but we don’t want to solve every problem with AI, because when AI’s not good at everything, we don’t need to, and it’s just like you don’t need AI to do everything for you. So, we want to make sure we have options. So, for example, something new that we are doing some final polish on right now addresses the session lifecycle. So, the back to Ian’s point about the infamous product jobs to be done, it’s we want to help clinicians prepare for a session, conduct a session, and follow up from a session, and within this we’re providing a collection of tools to help you do all those things. Do we think everyone’s going to use all those? No, no, definitely not. But some of them might be appealing to you in the way that you choose to give care, or the things that you find useful and helpful to you, and so we’re trying to again offer a variety that stands like just what we know these different pain points are, and we encourage folks to try them. We want you to get in there, try it, and see what you like, what doesn’t work for you yet. And there’s no expectation that all of these will be slam dunks for every clinician, but we want to offer variety, because everyone does offers care differently, so we want to be flexible in that respect, but really encourage folks to just try it out and see if it seems to work well for you.

Ian Knox 13:49
You brought up a really good point, like we’re still early in all this, right? Like, I think we’ve come a long way in the last year, but these tools are evolving, you know, month to month, and I think if you experiment you probably have a pretty good feel now for as you do a note like what will come out, what you edit, how you prompt, but these are improving all the time, and hopefully we’ll get into a little bit on like how they’re getting better, because I think it’s really interesting to understand that and how actually therapists are helping these things get better, but, but just even experience, I think, is useful, just to understand, like, how do they work, and how, how do I interact, and what can I trust and not trust, and so that’s that’s been really, I mean, I’ve been using AI quite a bit in my job, and you start to get a feel for it, you get started to get a feel for how it works and what’s beneficial, and then what you don’t trust, right? Because you can’t trust everything with AI, like it’s generally pretty accurate, and the more focused the job and the better context you can give it, the better it will do. So a lot of this is sort of working through all these different things, and to your point about client trust how they feel about that. Well, that’s something we’ve got to unpack and figure that out with with our customers.

Katie Vernoy 15:07
Well and I think because there are so many different ways to do it, I think those options are very helpful. I have some clients that the note taker runs in the background, and there’s some clients that I use dictation, and some I type something in, and it’s old fashioned, right? So, there’s there’s a lot of ways that I can respond to some of the hesitancy that I think folks have about AI, and I’m developing processes around those things, and I think it’s something that I like the feedback to play around, because if you don’t play around with the tool, and to your point, do you, if you don’t play around with the tool, and then again later when it’s been updated, and then again later when it’s been more refined, you won’t have a sense of what the utility might actually be, and so I think there’s this, this element of potentially overwhelm related to how should I make those decisions, because there are so many different options, and and I like the idea of play around, but maybe, maybe we’ve got some other pieces of advice to kind of refine a little bit on on what people might be looking at when they’re trying to set up their practice.

… 16:09
(Advertisement Break)

Curt Widhalm 16:10
I really like that the AI experts on this episode are saying you can’t fully just trust AI, and I think that there’s a lot of good discussion that’s out there, just as far as the learning curve that Ian, that you’re talking about, where it does take kind of some experience with some of these things to understand the utility of them, and I think that a lot of what I’ve seen in our community is the hesitancy to adopt some of these tools, is it’s not working perfectly for me right out of the gate, and therefore this is terrible, and nobody should ever use these kinds of things. We have already talked earlier this month in our podcast about some of the best practices of artificial intelligence use with clients. We’ve talked about the AI ethics updates that Katie and I worked on, the informed consent, the due diligence of practicing with some of these tools that we’re pushing, that anybody, anything that you create with AI that you put your name on, you’re ultimately going to be seen as responsible for. But bringing this back to the two of you, what does SimpleP ractice see as some of the best practices, and how do you support clinicians and using the AI tools available for them.

Megan Toomey 17:24
I think one of the most important things is maintaining the trust with your clients and being transparent about how AI is being used and getting their consent. That’s been a very important focus for us, is making sure that we not only enable that, but encourage that. I think that that’s really critical.

Ian Knox 17:46
Yeah, and I’d also add to your point, like, you know, for me AI can be absolutely amazing, right? In terms of the output you get, you’re kind of blown away, and then occasionally it’ll miss something. So, you know, best practices always review. T o your point, Curt. You mentioned, like, it’s it’s your name, so you need to review that note, and just don’t assume AI captured everything, although it’s probably got a pretty good, and then over time you get a good feel for what that is. So, I think those are probably two of the key things I’ve seen.

Katie Vernoy 18:18
And one of the things that’s really interesting is there are definitely times when I need to do a lot of editing on a note, but as you said, Ian, sometimes it’s perfectly right, and if I go to sign it and I’ve edited nothing, there’s a prompt that comes up and says, “Did you review this?” before I can actually? I have to click that before it lets me sign the note, and so I appreciate that, that there is that element of at least a cross check at that point to make sure that we’re truly owning our own documentation. B efore we get too far into this, because I think that that both of you have alluded to some of the trust and the data issues and that kind of stuff, I’d really like to talk about that element of the best practice of how do we make sure beyond somebody saying they’re HIPAA compliant, which, as we’ve seen in a lot of other episodes that we’ve talked about AI, sometimes they say that and it’s not actually HIPAA compliant. We know Simple Practice is, so we don’t need to belabor that point. But what are the ways in which data is we’re going to have a whole data episode, but high level, what should clinicians be aware of around data when, when using an AI platform?

Ian Knox 19:29
Well, first of all, you know it’s it’s your data, right? So you know, as you’re evaluating platforms, you should know the data practices, and like you know different aspects of that, right? When you’re looking at different companies and different ways they handle data, like over time they’ve built trust and experience on that. I’d say, you know, there are so many aspects to data security, which I think we can probably get into in another episode, because you could spend a lot of time that, but it’s part of its cultural, right, in terms of within the company, how much do they lock down data, and, like, you know, make sure people don’t have access to what they don’t, don’t need, or shouldn’t be looking at. I think, secondly, is all sort of safeguards and systems they have in place to prevent anyone getting in and accessing data, is like super important, you can be HIPAA compliant, but then maybe your security posture isn’t super strong, and so investing in that is really important, and then you know the ability for you to access and download data as you should need it is really important, so I think you know there’s there’s a number of aspects of this often come down, I think, whereas sometimes you know, if I’m using different technology, it’s who is who are these people, right? You know, is that a startup where everyone can just like go poke around and see what’s happening, or have they published how they’re operating, right? And how they feel secure, because that’s like, you know, that’s the biggest nightmare, right? It’s like your data gets exposed, right, and that’s something obviously we take super seriously at Simple Practice.

Megan Toomey 21:06
Did that, n o, we’ve recently even built out our own trust center to be even more transparent about a lot of these policies.

Curt Widhalm 21:12
The Trust Center, I know that some of the reaction around this from the community that we’ve heard is just, you know, the fear that S impleP ractice is going to be like a lot of the other venture capitalist companies that are looking to replace therapists and taking on all of everything that’s going on, even the data of us appearing. I’m going to put you on the spot and say, is that what you’re doing?

Ian Knox 21:38
Yeah, absolutely not. Very fair, I think that’s a fair question, though, right? Because you know, I think that is a fear of people, and you know, when we saw AI coming a couple of years ago, we sat down and we just really thought about our principles and what the company is about, and fundamental to our belief is that clinicians should be at the center of care, you know. I think if you, if you think about this long enough, or you’re in this industry, is is AI – does it have empathy? Can it, like, pick up visual cues? Does it have the training? It can simulate some of those things, but will never do as good of a job in our view as a professional. Now, can it help? Absolutely, right , with clinical support and be able to answer questions, etc. But you know, we came out very early on, and with our principles, right, and what we believe, and I think that’s really important to state. The second thing we did, our Chief Marketing Officer, Doug Scott, he came from Twitch, which is an interesting company. It’s all about live streams and other things and community, and so right at the beginning he got us all on live streams with customers. Like, can we do that now every month? Jonathan, our CEO, is on the live stream, and I’m on one, you know, every quarter. Megan’s been off to quite a few. And so we have those direct conversations with customers, because in some ways engaging face to face is a way to build trust and go ask those questions, plus doing things like coming on and talking with you, right, to go deep on But But I think ultimately, you know, if you get to like, what data are you sharing, or what’s being used, that should be a decision for a clinician, right? And you should know exactly if a company has your data, what could they be doing with it, or what are they doing with it, right? And so that transparency is important, and then being educated on what those use cases are, and what’s allowable and what isn’t, is super important, but our position has, and always will be, that questions that are sent to the care under the tools we build will be a way to help clinicians, not replace them.

… 23:53
(Advertisement Break)

Katie Vernoy 23:54
We are planning to do an episode, and actually kind of a CE episode, on data and privacy in the fall with somebody from Simple Practice, so we will be getting into this more, so I promise that I want to still stay at the high level. Building AI requires some mechanism to train the AI, and so if we can have a little bit of a conversation about how AI systems are typically trained. How SimpleP ractice has been training. I know that recently there’s been some updates around transcripts and that kind of stuff, and so I don’t know what the most effective question is here, but if we can kind of talk through a little bit about training AI systems and doing what you’re talking about, Ian, which is keeping data really secure and making it a clinician-driven decision around what data is used, because I can imagine it would be very hard to have a good AI platform if you can’t train on actual real data, so.

Megan Toomey 24:58
I would love t o take that question, I’m so glad you asked it. Short answer, we’re not, we’re not training an AI model, that is not happening. And I wanted to talk a little bit about, like, what training an AI model is technically is very complex. I just want to try to, like, demystify what that means, and so I have an analogy about a puppy, mine is sleeping next to me over here. So, think of like training an AI model, like teaching a puppy how to behave. When you bring home a new puppy, it’s kind of a nightmare. This is my life a year ago, like chewing things, not potty training, it’s just kind of a disaster. And it’s no one knows what to do. You don’t know how to train the puppy. The puppy doesn’t know what to do. You ask it to sit, it goes and finds your socks. Like, no, it’s not working. So, you kind of have to start from square one, and you spend a lot of time and a lot of repetition to train your puppy. Or you could do what we did, and we took our puppy to a boarding training facility, and had the experts train our puppy for us. And when she came home, she knew how to sit, and all we had to do was figure out, oh, this is what the command she knows to sit, and so we would learn that command, tell her to sit, and she would sit. So think about that as it relates to like an LLM or AI. We’re taking the same route as I did, where we let an expert train our puppy, and we have experts who are training the model. It is not us. We are not doing anything like that. What we are doing is trying to figure out the best way to talk to the model. We’re trying to figure out what the puppy sit command is. What’s the most effective way to work with the puppy to get the output that we want, but in no way are we actually doing any of the actual technical training or rewiring of the models, none of that, and no data from Simple Practice is involved in that at all. What we’re just trying to do is be like, what do we need to give the model to give you the best output, and so just the difference of like we’re not actually doing the training, so it’s not part of Simple Practice at all.

Katie Vernoy 27:00
Does that mean, is it a white label LLM?

Ian Knox 27:03
Yeah, so under the covers, like just as Megan said, we have access to LLM models, right? And when you, what we do is, as Megan said, to get like really good quality, like you do when you prompt, right, you’re putting in it’s how you’re talking to the model, and part of this is the model understanding like what a good data is. So we have a clinical team at Simple Practice that works on as we generate different notes, grades those notes, right? So it starts to show as an input as you talk to the model, like here’s what grade looks like, and so it’s very, very important that we have clinicians on staff that can show what a good note looks like based on a session, right. And so, in the terms of AI, this is called like a golden data set. It’s very common in terms of how you grade and then how you show the model in terms of how you talk to it, like this is what great looks like, but you know we don’t train like an underlying LLM, as Megan said, so we don’t feed questions data into like a an actual LLM model, and that’s pretty common, actually.

Curt Widhalm 28:16
So for our audience, for people who are out there listening, when they are using AI, what is what happens once it goes into the black box of the systems of SimpleP ractice? How does all of this stuff end up getting incorporated into everything, and I guess continue to demystify this from the user end.

Megan Toomey 28:40
So today, when you know Katie, if you’re using S impleP ractice and you’re using Note taker, for example, you use Note taker, it will record, create a transcript for you, and then the transcript will persist for seven days, and then it’s completely deleted, or if you create a note, sign the note, the transcript’s gone immediately, it’s not, it’s gone, that’s it. So, in that respect, like, nothing is not even, not even around at all within that retention policy.

Katie Vernoy 29:11
There’s there was an updated policy around de-identified transcripts? ight? What was that? Because I’m not opted in, because I was an early adopter, but what is, what would happen, for example, if I started now and opted in to have my de-identified transcripts used to train. Yep,

Megan Toomey 29:11
Right.

Katie Vernoy 29:11
What was that? Because I’m not opted in, because I was an early adopter, but what is, what would happen, for example, if I started now and opted in to have my de-identified transcripts used to train.

Megan Toomey 29:25
Yep, so yeah, that’s the evolution that Ian was alluding to, where we will give clinicians the ability to opt in or opt out of retaining the transcript, but what we will do to it is we are going to remove all identifiable data, heavily redacted, and then we’re also going to decouple it too, so not coupled to a session, and all PHI identified data is completely removed, and the reason for this is because we want to make the features better for you, so going back to the analogy of like we want to work better with the puppy in this way, we don’t know, i t’s hard for us to know what good looks like without a really robust, as Ian mentioned earlier, golden data set. So, we do have clinical staff who help us define that, but we want to be able to determine that scale. Like, did Note taker make something up crazy? Did it miss important details? Is it just writing too much stuff? So, we need to basically create a scorecard for ourselves, and this is what that data could help us build. And we also want to improve it over time, so if the notes aren’t great, we can change how we’re talking to the LLM, we can change the prompts, and this helps us test that and make better prompts, so that the LLM outputs are better. But going back to it again, the transcripts that you know are retained would be completely de-identified and decoupled, and they can’t ever be patched up back together. It’s not possible.

Ian Knox 30:49
I just want to build on Curt, your question earlier is, are they what happens under the covers? So, as Megan mentioned, so the transcripts generated, and then what we do is pass that transcript into a private, we have a private model we’re using, so it’s not like you know, intermixed with any other company’s data or anything else. T hat’s private to us. And then you pass all that context in, and that context is what Megan’s talking about. The better we get those golden data sets and that context, the better the output comes over time. And then, as Megan mentioned, as soon as you’ve signed the note, transcript is deleted. What we have found are some pretty passionate customers are like, like we love the tool. How can we make the output better? And one of the ways is de-identified transcripts, where we’re providing that options for customers, but again, it’s totally up to a customer if they choose to do that or not.

Katie Vernoy 31:40
What are the tools that SimpleP ractice has currently, and what’s in the roadmap, and how do you make decisions on on what you’re putting into the roadmap, and what what’s next?

Ian Knox 31:52
Absolutely, I mean, I can start there, like if you go back to right the beginning of our conversation, in terms of the areas that are super interesting that I think they’d fit into kind of four buckets, and so that’s how we think about like the themes and our roadmap. The first is all about clinical care and clinical intelligence, and N ote taker is just the first step of that, but there’s so much more in terms of treatment plans and measurement-based care and client summaries and intake summaries and all that, that can help questions, you know, answer questions and get better at like working with clients, saving time, improving care. The second bucket is all around insurance and billing, and that’s everything from like verification of benefits to no compliance to making sure when you submit insurance that it’s going to get paid first time to dealing with denials and rejections and billing adjustments in the background, so that’s the second big theme. The third is around what we call practice operation, so that’s like we talked about that, what’s that business advisor, the analytics things like scheduling and intake, which is still pretty early, I would say, but that helps questions the knowledge to run their business, catch issues early, automate things in the background, and then lastly we have a theme around referrals and how you build a great caseload, and that’s not just getting – we have Therapy Finder today it’s not just getting any referral in, but like making sure that you’re getting the referrals that you want, like the ones that match you as a question in terms of what you provide, and looking at linking to payers, enterprise providers, health systems, et cetera, to get direct access to those. So those are the sort of four big buckets, Katie. And then, in terms of second question, like, how do you decide? T here’s always so many things. Like, we do this live stream with customers once every quarter, and there are so many ideas that come in for us to work on. So, there is a bit of art and science to what we pick up next, but usually what we’re looking at is like, where’s the most challenge and pain? Like, you know, we talk to a lot of questions, so you get a good feel for where to focus, and then you know we’re always looking at like direct feedback we get, so there’s always smaller things we incorporate as well. So we sort of plan our roadmap on a six month rolling basis and go from there.

Curt Widhalm 34:16
Where can people find out more about SimpleP ractice, all the wonderful work that you’re doing, h ow to influence this product roadmap?

Ian Knox 34:27
Yeah, it was a couple things. We can obviously go to simplepractice.com and check out everything we’re up to there. We have live streams happening all the time, so go check those out, so you can just turn up and ask us questions directly. We also, I’ve been just giving my email out, ian.knox@simplepractice.com Always happy for people to email me, and love that when people do. I’m sure Megan would appreciate the same. So.

Megan Toomey 34:52
Yep, Megan,.Toomey , t o o m e y@simplepractice.com. Also, we community too, we love to see dialog in c ommunity, and get great ideas from folks there in Facebook.

Curt Widhalm 35:03
And we will include links to all of that, p ut that information in our show notes over at mtsg podcast.com and make sure that you follow up with us on our social media. Join our Facebook group, the Modern Therapist Group, to continue on with this and all of the other conversations until next time. I’m Curt Widhalm with Katie Vernoy and Ian Knox and Megan Toomey from Simple Practice.

… 35:27
( Advertisement Break)

Announcer 35:28
Thank you for listening to the Modern Therapist’s Survival Guide. Learn more about who we are and what we do at mtsgpodcast.com. You can also join us on Facebook and Twitter, and please don’t forget to subscribe, so you don’t miss any of our episodes.