Video: Important change to Note Taker transcript retention | Duration: 4164s | Summary: Important change to Note Taker transcript retention | Chapters: Welcome and Introduction (4.914999999999964s), Transcript Retention Policy (104.79999999999995s), Q&A Session Introduction (592.015s), AI Training Clarification (710.2249999999999s), Data Privacy Concerns (877.155s), Training vs Inference (1061.59s), Data Privacy Protections (1360.425s), Data Usage and Training (1596.175s), Ethics and Transparency (1824.495s), Ethical Commitment (2022.02s), Pricing and Value (2261.5299999999997s), Compensation and Data Usage (2439.79s), Transcript Editing Limitations (2682.625s), Data Bias & Ethics (2805.09s), HIPAA Compliance Safeguards (3023.81s), Data Privacy Balance (3293.04s), Quality & Rollout (3485.215s), Leadership Change Protections (3542.31s), Opt-In/Out Mechanics (3710.0950000000003s), Closing Q&A Session (3949.1899999999996s), Closing Remarks (4042.495s)
Transcript for "Important change to Note Taker transcript retention": Hello. SimplePractice. Thank you for being here today and joining us on this live stream. My name is Doug Scott. I'm the chief marketing officer at SimplePractice. Topic of this live stream today is our transcript retention policy update that we announced this week. We know a lot of you have questions around this. There's also concerns around AI and its role in therapy. And we wanted to proactively have this forum so that you could ask questions. Hopefully, we can give you the answers that you seek, and, we'd give you, an opportunity to connect with us around this topic. So very much looking forward to that conversation and hearing from each of you. First of all, this is about transcript retention policy. We're gonna be trying to answer as many questions about that, how it relates to our notetaker product, things like that. So please, enter those questions. We'll do our best to get through those. And, this is we're gonna set a little bit of groundwork here in the beginning, but for the most part, this is a 100% about answering the questions that you have. That's why we're here. And I'll be joined by a bunch of my colleagues across our clinical, technological, legal, product, teams to help give you the answers that you need. But first, I wanna start off with doctor Lindsey Oberleitner, who is our head of clinical strategy here at SimplePractice. Hi, doctor Lindsey. And I was hoping she could introduce herself and then give you just a a little bit of a grounding, set the table on what this policy change was, why we made it, and just make sure that we have a fact based to go off of. And from there, we're gonna jump in to questions and bring some other colleagues up on, on the screen. Doctor Lindsey, welcome. Thank you, Doug, and thank you everyone for being here with us today. I'm, doctor Lindsey Oberleitner. I'm a clinical psychologist myself and the head of clinical strategy here at SimplePractice. So to start off today, as Doug mentioned, we wanna make sure that we, give everyone the same information to start this conversation. So just a quick, recap and some extra details on what is happening with transcripts. So starting on 06/16/2026, simple practice, we will begin retaining transcripts. Existing notetaker customers are automatically opted out of this change. The transcripts are de identified and decoupled. What that means is that all 18 HIPAA safe harbor identifiers are scrubbed and removed from that information, and then we decouple the de identified transcript. So what that means is it's no longer linked to a clinician, to a client, to anything, that would allow it to be relinked to any clinician or client. So the decoupling is an extra step that we wanted to ensure that we took with any sensitive data. De identification and decoupling provides enhanced, privacy protections beyond what law requires to our customer sensitive data and safeguards against anyone ever seeking a session transcript, including but not limited to the case of subpoenas. So this is why we wanted to go that step further and ensure that everything is decoupled from you, in addition to ensuring that it's fully de identified. This evolution in our policy, was created thoughtfully after speaking with clinicians under the advisement of our customer advisory board and our clinical advisory boards. We've spent a lot of time in the past few months really ensuring that we've made the policy and approach that we have, reflective of the sensitive data that all of us work with as clinicians. So I know a lot of the questions that have come up have been around why are we doing this. So the goal of us retaining de identified and decoupled transcripts is so that we can continuously improve existing and upcoming AI features. Retaining the de identified and decoupled transcripts helps improve note taker quality because it lets us understand clinical nuance and context. It helps us improve note taker accuracy and the clinical language used. It can help us reduce biased or inaccurate interpretations that can be made in creating documentation and helps us recognize therapeutic progress and client sentiment. For example, a word like pain might technically be seen as negative to note taker. But as clinicians, we know that when we discuss pain in context, that can actually represent real progress and change, for a client. And we wanna ensure that notetaker is continuing to be able to better document and accurately categorize things like that. And I know as some of these questions have come up, one of the things that we know and I know as someone, who's who's helped consider how do we make data behind the scenes to be able to build things like note taker, mock sessions will never truly capture the full kind of complexity of what actual treatment looks like. So what does this mean, and how is it being implemented for, simple practice? Again, I know I already mentioned this, but existing notetaker customers are automatically opted out from transcript retention. Note transcripts will be stored unless you actively choose to opt in, and you always have the right to control your opt in and opt out. If you wish to remain opted out, you don't need any further action. You are automatically opted out fully from this process. If you are not an existing notetaker customer, but you sue choose to sign up, de identified and decoupled transcripts from your sessions will be retained by simple practice for product improvement unless you choose to opt out. And, again, you have the full choice to opt out at various levels. You have full control over your data, so if you ever decide to opt out of de identified transcript retention, you can do so. Under, again, close advisement of our customer and clinical advisory boards, we've created three ways that you can manage your session data to give you the utmost control of how you might experience this, this option. So the first one is clinician level. If you as a clinician say, I don't want any transcripts stored, for any of my clients, you can use a global toggle in your settings to turn transcript retention on or off with a single action. Each clinician in a group practice must toggle transcript retention off for their own clients. There's also a client level, choice. If you are okay as a whole with transcript retention and want to leave that clinician level on, but in conversation with your clients, determine there's an individual client or a set of individual clients you want this turned off for. You have the ability to toggle transcript retention on or off for specific clients. When toggled off for a client, it applies to all sessions that client has with you. And finally, we also know and part of working with other clinicians along the way to build all of these levels of options is that even if you and your client agree to transcript retention, there may become sessions that you decide it's not right for this session. I don't want this individual session saved. So we've also allowed a session level opt out, and that option is for you to opt out of transcript retention for any individual session even if it is on for that client and for you. I also wanna make sure, now that we've talked about some of the specifics for retention, that I share a little bit about our core approach and principles to AI at SimplePractice. So whether or not you are a note taker customer and whether or not you opt in or out to this process, we're committed to your privacy, and we have three principles that we always follow in regards to the transcript, retention and holding of data. Transcripts will always be de identified and decoupled according to all 18 HIPAA safe harbor identifiers, and we will decouple the transcripts, meaning it's not linkable, and safeguards you against subpoenas with that data. If AI augments your work, it doesn't replace it. This data is used solely to create supportive tools for clinicians that optimize your workflow and help you manage your practice more efficiently. We are not training a model to replace your role as a clinician. And, we believe in human first data governance, and our internal AI governance team is made up of clinical, legal, technical, and ethical experts who have vetted this process and been involved on every step of the way to ensure that we continue and will continue to monitor and ensure that all things that we do meet high safety and legal standards. With that, I don't wanna take too much time explaining these because, again, as Doug mentioned, the purpose of today is to answer all of the questions that you might be coming to us with. So I will pass back to Doug as we move on to the next step. Thanks, Lindsay. Yeah. So let's move into q and a here. As we do, we're gonna bring some, additional colleagues up on the stage, and I'll ask them to introduce themselves. While we're getting that all sorted out, I'm also gonna remind everyone to please put any questions into that q and a tab. That is absolutely the best place, for those questions so that they can be upvoted. And I will remind you on a temporary basis to head back over there and upvote those questions that you most want us to answer because that's the way we will be, going through. And with that, I'm gonna, welcome some of these other folks to the stage, and I'll start with Tony. Tony Lee, if you wouldn't mind introducing yourself and, let everyone know what you do here at SimplePractice. Thanks, Doug. Hi, everyone. Thank you so much for joining us today. I'm Tony Lee. I'm the chief technology officer at SimplePractice. Awesome. And, Ali, if you wouldn't mind, doing a quick intro, as well. Yeah. Hi, everyone. I'm excited to be here as well and answer some of your questions. I'm the chief legal officer here at SimplePractice. I manage our legal compliance and fraud teams. And last but not least, Justin, if you wouldn't mind introducing yourself. Absolutely. I'm Justin. I'm a principal product manager here at SimplePractice. I work alongside our engineering teams to develop the features like notetaker and many of the carried features that we've been building. Awesome. So as you can see, we have a great cross section of folks here, at SimplePractice that hopefully can answer all of the various questions that you've already put in the q and a. And there's a lot of them, so we should just get right into that. Just so you can follow along, I'm gonna share this the question up on the screen. That's gonna make all of us really small, I think. So I'll share it. I'll read it, and then I'll bring it back down, so that, you can, see from us and hear from us as we answer the question. Okay. So first up is from Lauren. Lauren, thank you for the question. It says, in the FAQ, it says this is not being used to train an AI model. Can you confirm that this will not ever change? What is the actual purpose of this? I think this is sort of a two part question. So why don't we have Tony talk a little bit about the purpose of it? And then, Ali, maybe you can talk about the confirm whether or not this will ever change aspect of it. So, Tony, what is the actual purpose? Thanks thanks, Doug. I I think, Lindsay covered the the intent to improve the quality of the features, especially notetaker today, by using this information. But I I will go a little bit deeper because I think, when you just rationally think about it okay. So you you have all these, do you not de identified decoupled transcripts. You test notetaker, some of them do great. We know notetaker is not doing great on everything. We have some cases. What do we do then? I just want to reiterate that we don't take those failure cases and use them to train the model so they do better on them because we don't actually have control over the model and we don't train models. What we do have control is the instructions we give notetaker, and that's what we're changing as a result of these test cases. Like, we can tell notetaker, try to make it shorter in these cases. Try to emphasize this. Try to be a little bit more formal, a little bit more casual. The instructions are what we can send to the model, but we're not training it as a result of this. And that's how we intend to, improve the quality of notetaker. Awesome. Thanks, Tony. And and, Ali, I was just wanted to touch on the other part of that question, which was around whether or not this will ever change. Ever is a long time, of course. But maybe you could talk. a little bit about how we see the world, and why we're why we're. Yeah. And and I know I'm gonna sound like such a lawyer when I say this. We certainly can't commit to, never doing anything. Right? Because, like, that's just an impossible standard to hold. However, what I do wanna say is, like, we are super, super committed to maintaining a human in the loop approach to how we develop AI and use AI. Meaning, as Lindsay said, we are not developing an AI therapist. We are not trying to replace you. Our goal is to help you use AI to augment your practice while you maintain full clinical responsibility, oversight of care of your patients, and all those things. We will, go to the next question and keep things moving. So, Kevin raises, I'm concerned about how the data will be used, such as selling to third parties, governments, private companies for purposes over which you have no control. Will you let us know all the ways the data is being used and how we can be sure? So, in specifically thinking about some of these questions about I I see a couple things here, which is selling to third parties. I'll pass to you, Ali. And, I think we've we touched on the last question about all about ways that the data is being used, but I'll leave that open for others after I pass to Ali first. Sure. Yeah. I can answer the question about, selling data. So simple practice does not sell client records, identifiable client information, or de identified AI transcribed transcript data to any AI third parties or any any third parties or any AI companies. I wanna be very clear that your clinical records and client relationships remain yours. So when the when information is de identified, we remove all those identifying details. We talked about that decouple from the client, the clinician, the practice, session content. You know, we we talked about how SimplePractice is planning to use the identified information for product improvement. However, that is limited to us improving our products. Right? It's not creating a data set for us to sell to outside companies. If we do work with third party vendors or technology partners, those are supporting simple practice and providing the services and they're subject to very strict contractual privacy security and more applicable business associate obligations. They do not receive customer or client information to use for their own independent purposes. I'm back. Sorry about that. Great. I I forewarn people I've been having Internet issues. It's kind of classic, just as you one would expect and hope for a live stream. So I apologize for that. If it happens again, I apologize in advance. And I'm sorry that I missed, that question. But next up question is from Amanda. Amanda says, initially, we were told transcripts would not be kept. The AI program has been working great. So why now? Are the benefits of improving AI worth the risk? How that data can be used and sold in the future? It's a very big question. I think, this is one where anyone could probably answer this, but I think, Tony, if you want us retouch on some of the benefits of improving AI and how we think about, the, you know, risks associated with that being accessed on the technical side, things like that, that'd probably be a great place. And then we can we can revisit our policy around this about not selling this data. So, anyway, but, Tony, maybe you wanna start a little bit with the benefits again. Yeah. You know, I'll touch I do want I'm actually grateful for the opportunity to kind of expand on my previous answer because I'm looking down in the q and a. I see there's, additional comments. So little little I wanna talk a little bit about training means to me, and, with the understanding that it means different things to different people. So I I feel like I should baseline a little bit about what that means for everyone. And if you have a different belief and understanding, that's perfectly legit. I just wanna explain when I talk about we're not training models what it means. These large language models essentially have two modalities. One is in training mode and the other is, kind of in question and answer mode, what they call inference mode. The training mode for these models is something that happens at a big data center. It takes months and a lot of money, and it ingest the Internet and essentially builds the intelligence, the pseudo intelligence, the model. That's not what we're doing. We're using it in inference mode, which is essentially what you're doing to chat GPT. You ask it a question, you get an answer back. The way we use it is that it has no memory. It has no state. It doesn't learn. Well, l o m's don't learn anyways. They're not people. But it doesn't remember anything of what we asked it beyond answering our question. And so when I talk about we're not training it, I mean, we're using it as question and answer machine. Take this transcript. Please summarize it following my instructions. It doesn't take anything away from that. And when I say we're not training it, we're not doing the formal training model where we feed it a lot of information and it adjusts itself. There's no adjusting of the models the way we're doing it, and we're not we're not planning to do that. So, I just want to baseline on the meaning of that. For the benefits, it is really I mean, we've learned this through note takers that, it's really hard to capture just the breadth and richness of sessions to write a good note. And, you know, what we do is we build product for you and for the rest of our customers to deliver value, and we we want to serve everyone equally. And that's our challenge, and that's why we're motivated to do this. And we're trying to do it as safely and and, you know, as ideal as possible. I know it's controversial to many of you. We're putting in all the safeguards. The data once it enters our system is is a HIPAA compliant, HITRUST certified environment. We have very limited controlled access to the data itself, the identified data. The decoupling is the extra step that we added, and we're making those trade offs because we believe we can deliver a significantly better product with this test data, that we're collecting. Just to add a little, Doug, if you don't mind, to some of those questions too. I know part of it came up that in that question of we that you weren't aware of our intentions to store transcripts. And I'll say from the start and we understand that there's so much information coming at once, but we we've tried at each stage to say we would not do it until we develop the opt out. And, actually, this was a really core part to us is that we are trying at each step to ensure that we are as transparent as possible, which is why we're also announcing this before it goes live and why we have been trying to engage clinicians along the way to share what the procedure would look like and ensure that we we could bring it to you as quick as we could so that you have the choice. It's also why we were very thoughtful in this space of, like, how are we gonna do opt outs and where were they going to exist? And truly deciding At one point, as you can imagine, our thought was, do we allow one of these? Which one do we allow? And as you can see, what we came to is we want all of that control to be in your hands again so that we are transparent. It is clear. You have those choices along the way. And the one other piece that I wanted to add to that question is I'm really glad to hear that it's been working well for you and that and we hope it continues to in all of those spaces. We know we still need to do our best to be able to ensure that it's not working perfectly for everyone in all scenarios, and there's still room in which we can grow a lot. And as I mentioned when I shared the intro, what we know what I know as a clinician myself is even when I've done mock sessions with individual, it's not the same as real data. And we know that some of the gaps that we see when there are problems with notes are because it just isn't the same. I was talking with someone on my team today who said, you know, the hardest thing to imagine is, like, when you see the you know, even as AI tries to create a fake transcript of a session, it is nowhere near what our actual therapy looks like. It's nowhere near what that real process is and how complex it can be and how how many steps there are along the way to hold a conversation. So it's really valuable in ensuring the tool keeps improving and that we can give it the right information to build the notes in the right way with the right questions and the right questions and the right limits. So I just wanted to add that little bit too. Thank you. Appreciate both of those answers. And I just wanted to apologize. Nicholas, sounds like you thought I was laughing at the concerns of selling data. I very much did not intend that. It's a very serious matter. We take it extremely seriously here. It came across that way that I was somehow making light of that. Please forgive me. Not intended at all. But thank you for letting me know. Okay. So we move on to the next question, which is, what protections exist to prevent either you, your parent company, Vista Equity, or the company that owns your cloud servers, like the Amazon, Microsoft, or Google, from violating privacy as they have frequently done before from using the collected data to create an AI therapist? I think this is probably a combination again of Tony and Ali on this. So who would like to take the first. one? I can go first, Tony, and then you can jump in. So servicing, I totally understand that concern, and thank you for the question. As I mentioned earlier, simple practice does not sell client records, you know, identify identifiable client information or these de identified AI scribe transcripts to any third party. I also wanna make it very clear that we do not provide that information to this equity, to any cloud infrastructure providers or vendors so they can build their own AI products or an AI therapist. Right? Vista is is actually an investor in simple practice, and they don't operate simple practice. They don't operate the platform. They're not going to have access to customer records, customer, you know, any informations like that, any of these transcripts. And in fact, you know, those transcripts live in a, you know, simple practice controlled ecosystem, with a very, very limited access to, by simple practice employees only. We also have, you know, restrict we have technical restrictions in place for access. We also have audit logs so we can see who's in those ask who's in those transcripts and what they're doing with them. So we and that's a very, very small limited number of simple practice employees. In terms of protections, you know, we use multiple protections to limit how data can be accessed and used and especially with our vendors including, your contractual restrictions, business associate obligations when that's applicable, privacy secure and security requirements for vendors. As I mentioned, access control, audit logging, we put that those obligations on vendors as well. And then we also limit, the use of customer and client information, meaning they can only use the data we give them, for the specific purposes that we're contracting for with them for. As I mentioned, cloud providers, and Tony can probably elaborate on all this more, but they support the infrastructure used to deliver the service. They don't receive, rights to use customer or even access necessarily customer or client information on their for their own AI model or or product development. So I I'll Tony, I'm sure you can probably weigh in a little bit more there. Thanks, Sally. I mean, I think you you covered, most of what I would have, said as well. I I will just reemphasize that, in our cloud usage, we are on AWS. We do that for all of our serving infrastructure. So we already have PHI in there. We are very we we maximize our security profile within AWS using the controls they grant us and using our, HITRUST certification as well to audit them regularly and make sure that it's secured. It is a dependency. I mean, I'm not gonna say it's not. But I feel pretty good about the relationship we have with them and the security controls that we have in place, to maintain your privacy and our privacy. Great. Thanks. Next question is from Talia. So you're using us and clients to train your models and profit off your features you're offering. Where is the compensation for this? So, Tony, I think it would be great to talk about, whether or not we're training models and how we're using it, And, and then we can try to address the compensation question, as well. So, Tony, you want to talk about a little bit of the training models or not aspect. again? Yeah. Yeah. Yeah. And, you know, I'll try to be consistent that, definitely should train is almost an eye of the beholder at this point. But I I will say that we are definitely using, your inaction with the product to build new product features and to improve the current product features. So in that sense, yes, we have a feedback loop where what you're doing does inform what we need to do, and I think it improves us. And improves the quality of our product. So I think that's true. I mean, as I said earlier, by the definition that I provided for training, we I do not believe that we are training models in any formal way. But we are definitely using, your feedback loop to improve our product. I'll hand off to someone else to talk about the, I think, the compensation piece. I think this is a big topic. And, Ali, please feel free to jump in here as well. But from my standpoint, what we're doing right now is trying to build the best products that we can. And this is a reinvestment cycle, basically, in trying to build out those products. So it's based based on a lot of what Tony is saying. And right now, that's the the loop. And because it's a pretty focused use case for this, it's not something that has, like, broader compensation implications at at this time. So, a very good question, though. I think this will continue to be a very interesting conversation for a long time, going forward, about the broader question of models, what they're trained on, and ultimately what they, and and how that compensation loop works. So a totally legitimate, very important question. Okay. The next question is from, Chase. Chase asks, please explain how the retention of sessions transcripts for the purpose of improving notetaker is not, quote, training, unquote, the AI system. Tony, do you wanna just touch on that real quick? Yeah. I mean, I I see the time stamp happened before I gave my other responses, so I don't know if it's still a question. Again, we are using the the test suites we produce out of this to inform the instructions we're giving the LLM, not for training, a model by any formal definition of training, and it is stateless. It has no memory. It doesn't learn, improve itself based on those instruction changes. It just will affect the output of the note that it creates at our request. But if if I'm happy to keep going. I just don't know if the the previous questions helped. Yeah. Hopefully, that's been something that's, been addressed. But if if not, Chase, please do let us know. The next question is from Joanne. Joanne asks, the burden should not be on the practice or clinician to opt out. It should be defaulted to having the practice opt in. Many clinicians are not computer savvy. They're experts at therapy. Totally valid question. Maybe we can talk a little bit, Lindsay, about, the where we sort of drew the line on opt out and, a little bit about why. Absolutely. So, this is the reason that we thought that for existing note takers, this was very important that it was opted out. This is a behavior that clinicians are already engaging in. We did not want it to ever come as a surprise that all of a sudden transcript retention was occurring. We said from the very beginning that we would wait until we had our opt out, and we had confidence in our ability to fully de identify and fully decouple, which is where we now are. And that is why we didn't want this to be an opt in for those who are existing notetaker users. For those who are new to notetaker, this is why we do think it's so important that we continue to have these these, opportunities to ask us questions that we give the information very clearly that, you will see there's explainer guides on how to share this with clients. So that from the day that someone starts using notetaker who's new to notetaker after June 16, they have the information from the start. They can make the choice from the very start before they meet with their first patient that they can turn this feature fully off for their whole practice if that is right for their practice. But what we actually, in this process, realized it's about transparency, and we talked again and again with our advisory boards, with other clinicians to ensure that this was clear, that the options are clear, that it will be clear when you're ending a session if you would like to opt out because you now have that option. It's not something that we want hidden. In fact, there were lots of discussions about how do we make sure this is prominent enough that that choice can be made and not forgotten. Because, again, once it's saved and decoupled and de identified, we can't even match it back up. We want to be sure that someone is okay with having that transcript used for continuing to improve the product right from the start. So so we worked hard to make sure that this piece is transparent. I I hear there are many ways to do that and, and and totally hear you on that. That's actually again why with existing note takers, we knew that was a change in behavior we wanted to make sure to address, but appreciate you raising this. And as you see this and as you we always want the feedback. If it, we hope and we believe it's very clear where these options are and where they've been built into the system. We're always open to hearing feedback too. If it's not as clear and easy, as you mentioned, and and myself as well. Right? We're trained to be clinicians. Sometimes the technology side, we don't want it to be a trick. We actually want it to be simple and clear. So please tell us if it doesn't appear as simple as we believe it to be. We're always open to hearing that feedback. Thanks, Lindsay. Next question up is from doctor Kristen Ritter. Thank you, doctor Ritter. Just because you're following the law privacy wise doesn't mean this is ethical. Lindsey, I would love for you to talk a little bit about how we see this fall within the ethics, of of practice, management and and go from there. Absolutely. So I this is one of the pieces when we're thinking about the like, there there's two sides of this. Right? There's the ethical practice that we need to adhere to, and then we know that every bit of how we manage data as clinicians is then a really important part of our own ethics and our ways of practicing. So the way that we built the system to give the control to the clinician is what we believe is the core aspect of this. We know ethically, we are responsible for ensuring that we build a product that is accurate, that we build a product that does not, have bias, that does not lose it. We want to continue to commit to making the best product we can so that it doesn't lead to problems clinically down the line. But at the, on the other end of that story is how clinicians interact and make these decisions with their patients for their own practice. There are many considerations. And in that, that's why, again, to go back to that choice piece, we thought this was so core, that a clinician may have reasons that, ethically, this does not match the type of practice they have, how they intend to practice, and they have the choice to use or not use notetaker. They then have the choice, if using notetaker, to choose whether or not a transcript is, stored and de identified and decoupled. There are multiple spots along the way where we want to ensure that clinicians can practice the way that they believe aligns consistently with how they want to practice and how they adhere to their own, practice guidelines. So that element that was so core to us that we are transparent because the other piece that connects between us and clinical ethical practice is that you have all the information you need as a clinician to make those choices and to hold these conversations with your patients. So we are continuing to ensure that we give you the information we can that even goes beyond what would be core as a practice management to offer sessions when we can and trainings and education to offer tools about how to share this information with your patients' clients so that you can ensure that you're adhering to your ethical practice. I'll also say that we've gone beyond also to ensure that you have tools in sample consent of how you might choose to discuss and document this in your practice. And again, you you may not choose to use that. That's your choice, and we wanted to ensure that choice was core to everything that we did for clinicians with this. Thanks, Lindsay. There's another question from doctor Ritter. What is stopping you from selling this to major AI companies? You're already showing that you're ethically and morally corrupt for retaining sensitive information in the first place. So why should customers trust that you would ever do the right thing in the future now that you own that information? Well, first of all, I I think I'll I'll answer this question. If anyone else wants to chime in, please please do. I don't believe we are ethically and morally corrupt. What I wanna get to is our commitment and our alignment with independent private practitioners, to build thriving businesses, to provide the best care that they possibly can. That is why we exist as a company. That's what we get up every single day to do is to try to find ways to help you. That is behind this policy as well, as well as the way that we're building AI tools and trying to provide choice and optionality, for the the many, many private practices that that we work with. That is what we're doing, every single day, and this is all part of that higher level goal. So our goal is to maintain your trust, to build on that trust, to work together, to help you thrive in the ways that you want to. And, and that's our goal, period. And, we're gonna keep trying to do that every single day the best that we possibly can, for for all of the clinicians that we work with. And I apologize if, if that comes across to you as as ethically or morally corrupt, but we we are genuinely committed to supporting you in every way we can. Okay. No other comments on that? Then we will move on to, the next, the next question. This is from Joanne Sell. Joanne, I'm sorry, but here's what it sounds like to me. Are you taking our private session data for free and using it to improve your AI? And have you possibly increased our prices down the line for AI? So I I'm gonna try to parse this question to be, are we planning on increasing our price for AI down the road or the implication that this will be, you know, be getting a price increase? What I would say is is, pricing is obviously a very important and sensitive topic. As we build, new tools and hopefully are creating incremental value for you, whether it's time savings or the ability, to to, be more present, any number of those things. There are costs associated with some of these things, whether it's our our teams that build them, the the tools and technology that run them. And so in those cases, we may occasionally take price increases. I can't say that we won't, in the future. But the most important and guiding light for us on everything when it comes to this is making sure that we are delivering value to you and that we're giving you the optionality to choose to pay for those things or not whenever possible, unless we think it is universally required and or necessary for folks. And that's why we have both tiered plans that people can choose from on the, EHR side. That's why notetaker is an optional add on. We know many people don't want to use AI at all, and so that's why it's not been rolled into our core packages. So, as we build new products, it's possible that those things may be optional add ons that have a price associated with them. But that is there's not a, some master plan around that specifically to this. This is a 100% about improving the quality of the products that we have, in your hands today. Okay. Next question is from Nicholas Carlo. Nicholas asks, can you clarify on not training a model? The data is clearly being kept to train the model. Clinicians should be compensated for training your model. I think we've touched on some of these things. Again, so, Nicholas, if we if we feel like we have not answered this question, I think we've talked about compensation. We've talked about the definition of training a model. I think, we'd be happy to continue this conversation with you. For what it's worth, I'd be happy to, provide everyone here my, email address so you can reach out to me, and I can try to, connect to you depending on the right, the question, like, to the right person. If I don't have the answer, my I'm putting it in here right now just really quickly. So, Nicholas, let me make sure I spelled everything correctly. That's my email address. And everyone else, please feel free to send me an email, on this regard, and and with your questions, and I'll we'll continue the conversation that way if we want, if we're not getting to your questions here. But, Nicholas, please in in chat, please feel free to, comment if we have missed some aspect of this you'd like us to clarify something. So I'm gonna move on to the next question, though. Talia's asked if you are using our work, how will you be compensating us? You cannot just take our work for free. I think we've we've talked about compensation here. I I understand your perspective on this. So right now, there is, you know, it's not a compensated aspect of this. This is something we're using to try to improve the product that, we're providing you. I'm gonna move to the next question at this point, which is, what guarantees is SimplePractice is from Amanda. Thank you for your question, Amanda. What guarantees is SimplePractice able to provide clinician clinicians that the data of our session transcripts are not going to be sold to other companies in the future or to create another company to provide AI therapy? Ali, do you wanna touch on this again? This is related to a question from earlier, but it's just wanna make sure that we we touch on it. Yeah. I mean, the the the identified transcripts that we're gonna be retaining are not being used to create a separate dataset that we're going to go off and sell to any third party and particularly any third party to make a product that's going to compete with you all. Right? We are super, super committed, to you and to helping you with your independent practice. And so we're not there's there's no plans whatsoever to to do that at all. You know, in terms of guarantees, I mean, I think we can just say our our mission is our guarantee. Yeah. And I just wanna like, in the chat, Andrea mentioned that, you know, can we move on to other questions that are not related to past questions to cover more questions here? Lots of thumbs up and exclamation points there. So we'll take that under advisement, and I will, try to keep, moving through the questions that we have here because there's lots more questions in here, and so I appreciate that comment. And the next question up is from Connie Chen. During the seven day window, you have full access to view the complete transcript while editing your draft note until you sign and lock the note or seven days pass, whichever comes first, end quote. Can I edit the transcript if I see that it is not accurate to the session? This is a question that I think Justin could give us an answer on. Sorry. Let me come off, mute there. Thank. you. Helpful. Yeah. So can I edit the transcript if I see that it is not accurate to the session? So one of the things I wanna bring up with the process of this deidentification deidentification and decoupling is that it would be impossible for us to be able to link that back to you or your client in any way. And so when, our tooling, and I believe doctor Lindsey mentioned this, that we're following all of the best practices that HIPAA has provided all technology companies, in this regard is that when we go through this process of that deidentification, it would be impossible for us to even link it back to you. And so there isn't a way for you to either see it or uncover it or any of that because we just can't bring it back, to either the client or you as the clinician. And that is intentional, not just because we're following a checklist that HIPAA gave, but it's also a safety and privacy aspect for you, the clinician. Did. that answer the question fully, Doug? I I don't see it still up on the screen, so I wanna make sure that I'm I'm I I think I think it did. But if if not, please, re, hit us up in the chat and let us know if we've somehow missed the mark on that, and we'll we'll try to address. And Connie says that I met within the seven day window. So does that do you wanna modify your answer at all, Justin, there? Yeah. The the session content is always available for you to view within the you'll see that in, like, I'm I'm referencing the web aspect of that page for you to view. We will still go through the same retention policy with, your session content documents. So, it will be removed after that seven day window or when you sign and lock your note. But you can always go back and view and see what's in there. You can't change anything. That's the way the system currently works today with that. However, we do offer things like refinement, in order to provide context if something's wrong with the draft that got produced within notetaker. Thanks, Justin. A next question up is from Andre. I have a few ethic equity and ethics questions, but I'll name my top three for now. How do you evaluate whether retained data disproportionately represents certain patient populations? What steps are you all, simple practice, taking to prevent AI tools from amplifying disparities in health care? Lastly but not least, are communities whose data contributes to improvement able to see benefits from that contribution? Lindsey, do you wanna take a crack at these? Happy to start this one, and this is an area I personally care deeply about ensuring that we continue as a field to ensure that we are doing all we can to reduce the bias, that that can occur because of the data that we are built on. Right? Because of the things that AI can, you know, learn from our own historical information that that does hold bias. What I will say is this was a real debate that we had to have, and we, gave up the ability to know more about who the transcript was on to ensure greater protection of privacy and decoupling. So could we have done this even better if we knew details about age, about race, about socioeconomic status. Absolutely. We could understand where biases were happening in that data. However, in that balance that we tried to strike in every piece of this decision making process, which is which truly has taken many, many conversations internally across many teammates to determine where do the benefits outweigh the risks. And in this one, we decided that decoupling was the most important. So what I will say, what that still allows us to do though is what we know right now, again, that our best effort right now when we wanna actually test internally is using mock transcripts, using made up transcript. Now, what it allows us to do is have a greater breadth. So we don't know the distribution. We don't know what transcripts rely on others, but we know we can test more often to understand, with greater, complex clinical data across individuals and from diverse backgrounds to ensure that we aren't seeing particular errors come up in certain spaces. What I will say is that's why so from a data perspective, that's where this assists us is a greater breadth of data, that we're able to to use and to understand. This is also on us to continue to where do we get information about the field? Where are there notable biases that we need to be able to ensure, don't get able to ensure, don't get replicated from just knowledge in the field, non transcript related and something that we take very seriously as well as we continue with any products. So and I guess as that ties into the third part of the question is because we can't directly tie into anyone's data and know their background, there isn't a way for us to directly show those benefits to those individuals. But we do hope, again, as a field, we're able to as as a product, we're able to, improve where this information is being based because you're right. There's bias on who tools, who are assessments, who are models of treatment are even built on. We wanna do our best to ensure that breadth is something that we continue to to learn from. Thanks, Lindsay. Okay. Next question, just to keep things moving here is, you said you don't own the AI model, so you aren't actively training it. How do you verify that the AI model being used to generate documentation is HIPAA compliant if you don't own it? What legal safeguards do you use the record holding entity have in place with the owner of the AI model that provide the parameters for when, how, and where the data is accessed and processed by the AI model? So, this is both a technical question for Tony and probably a contractual or legal question for Allie. So, Tony, do you wanna talk a little bit about technically how these things are related? Sure. And, we're not, we're not shy about sharing what we're using in the back end. We've done it before at these calls. We use, as I said, we're we're primarily served by Amazon Web Services, which is, you know, one of the big industry providers for SaaS services. And we use a service they provide called Bedrock, which essentially, puts a a wrapper around third party large third party models, like Anthropics Cloud and, OpenAI. But it does it in a, they create a temporary instance. And what they do, they they guarantee make guarantees around, not saving any state about bringing this down and not sharing it with other providers other customers. Sorry. Which means you don't have any risk of data leakage or mingling with other customers that are using it. It's ours, and it'll go away, and there won't be any state left. So it's a little bit traceless. And, you know, that that is, that's why we use Bedrock. And you do get to choose the third party model that they will use for you on your behalf, and we have picked Anthropics Cloud for a number of reasons, quality being one, but also for their ethical AI stance and their sustainability mandate as well. And I can go more into that too. But those are some of our, technical controls. Ali, do you wanna speak to the contractual legal side? Yeah. Absolutely. So, we all of our vendors, but particularly those who, you know, have PHI involved and when they're touching PHI for the process purpose the purpose of processing, the AI model, we have very, very strict contractual requirements around them. They're required to sign a business associate agreement. They're required to have all the security and privacy controls in place, to meet HIPAA and exceed. And in our case, we actually ask them to exceed HIPAA obligations. So we have a ton of privacy and security obligations in those contracts. And often we require above what the law requires. Thanks, Tony and Ali, for that. I do wanna note that, we're getting very close to the top of the hour, and, many of you may have places to go. So what I wanna say, we're gonna keep going trying to answer some more questions here for a while. And I know people do need to leave, but I would, just note is that, we will be trying to summarize. We're not I'm sure we won't get to all the questions today. We will be trying to summarize our responses here. We will be posting that out to places like, our Facebook community. There's a simple practice community on Facebook. Please go check that out. There's a link in the docs tab here. We'll also be publishing this to the trust center, as well as our blogs. So we will we will certainly try to do everything possible, to to answer these questions in totality as much as possible. Appreciate some of you have to leave, given the timing. We appreciate you being here and asking the questions and and bringing, bringing those to us because that's the only way we have, the best possible chance of answering them. So we really appreciate you spending the time, in your, very important days. So thank you for that. But and with that, we will we will keep going. There's some important, links and and other information in chat, so please do reference that, as we go. Alright. Next question is from Charlotte. Charlotte asks, in doctor Oberleitner's video, she said retain de identified session data helps notetaker align with, quote, your clinical judgment and draft notes in, quote, your specific voice, while also saying the data is de identified, decoupled, and cannot be linked back to the clinician or client. What data is used for clinician specific personalization versus general AI improvement? How can both be true? I think this is probably mostly a question I mean, obviously, doctor Lindsey, feel free to to comment here, but I think Tony might wanna chime in on this as well. Do you want me to start it? You wanna go, Tony? I can start off, Lindsay. There's a let me see here. Well, okay. When we're talking about so I don't remember the exact quote. So I'm I'm not gonna debate that or argue about that. But, when we're looking at as a a test set, the de identified data, de identified transcripts, it will be in a large mass. Part of that is because we do wanna capture breadth and diversity, so we are looking at a large sample set. And but within that diversity, there will be individual voices, and so that will be captured. And that's the whole point of it. It's not we're not trying to create a generalized voice. We wanna try to capture all the voices that were isn't within that set. I think that is what is implied here, but maybe I'll hang off to Lindsay, her fellow. Yeah. I think part of this to explain is that what we want is not that right now, you can imagine when we first think about how the prompts to AI were written and how these things were asked to be right and to be able to write a note from a transcript. A lot of this was kind of generally, like, pull forward information that fits these parts of a soap note or these parts of whatever is relevant. What we know we need to continue to improve on and continue to hear the need to improve on is how do we actually ensure that we are able to ask those questions or write those prompts in ways that pull the right information forward from any transcript to any person. So this does not I will, again, much like I said, with the information about bias, we we had to make an active choice. There was a point where we said we know we could write better notes. We could understand better. We could have history like we need as clinicians to show progress over time if we had transcripts coupled to your client and you. We did not think that the privacy risk overweight that. So what happens is making better notes means that then, as you know, note or as you may or may not know, but notetaker does use your prior notes to use your style to then match. And we can continue to improve the product to get better at using your note to improve your next note, which is something that it does. So that is the way in which I would explain this information. We actually made the active choice that we can't see and we can't track progress outside of what those notes say, because the transcript won't stay tied. So you're absolutely right. It is a balance between those two, and I hope that explains it in the video. It was a really short description to try to get to that information. So I really appreciate you bringing that up. I will make sure to tag to Justin if there's more you wanna share too being deep in it. But if that covers it, just wanna make sure to give you a chance to. No. I was prepared to jump in with what you just touched on, but the whenever we run, like, an output for a draft of a note or we're doing, like, the treatment planners or the client summaries now, we are leveraging context that's available from the EHR. One of the things that, this initiative allows us to do is it allows our data teams to work even faster with a lot more clarity so that we can make sure that as we're rolling out, whether it be a better features for notetaker or additional features that are gonna support you in your work, that we can have confidence that the quality across the board is gonna be at the standard that we want, to roll out with and that we can improve that quality over time. Thanks, everyone. Alright. We will keep going here for a little while longer and try to get through a couple more questions. I know that, we're starting to see people leak out of this. So, if we don't get to your question, my apologies, and we will try to follow-up in other forums, to get you answers to your questions. This is for Andrea. Most of the safeguards discussed today so far rely on simple practices, current policies, and leadership. What protections survive a change in leadership, ownership, investor priorities, or business model? I think, Ali, that is probably one that is, for Yeah. Yeah. I can take this. And it's a super fair question because, you know, we want we don't want our customers to rely only on trust in our current leadership. So what I would say is there are several protections that are built into how simple practice operates that are intended to continue regardless of changes in leadership, ownership, investor priorities, our business model really. It's legal and contractual obligations. So, you know, our terms of service and business associate agreement, vendor agreements, privacy and security laws, continue to govern how customer and client information may be used. You know, for protected health information specifically, there are strong very strong business associate agreement protections and strong laws in HIPAA, that include limits on permitted uses and disclosures and safeguards for PHI, breach reporting, restrictions, all sorts of stuff like that that can survive the termination even of a business associate agreement. As I mentioned previously in this session, with respect to sort of ownership and investors, you know, this equity partners is an investor, but they don't operate the platform itself, and they don't have access to, any customer records or any of these de identified transcripts. And, you know, just simply because of the fact that they're an investor. And then vendor restrictions, I would say, is a really important one. Right? We enter into contracts with our vendors, that put very strong restrictions on place, including, restrictions on them being able to use any data that they receive and training their own models or things like that. Those contracts general you know, will generally survive any sort of change in leadership or change in ownership or anything like that. And, you know, I've lied to sort of end by saying, you know, we're committed to transparency. So if any practices material or change, you know, we will be sure to update you, in in a way that you can be informed and and take action if needed. Thanks, Ali. Appreciate that. Broken record here, from Leila. I hopefully, I pronounced your name correctly. These are all important ethical questions, but, please, can we get into the nuts and bolts of what our options are, what the options mean, and what we need to do? I still don't know what opt out versus opt in means. I also don't understand if opting out is potentially more ethical. Why isn't that the default setting? Thank you. Thanks for the question. And I know we've talked a little bit about opt in and opt out, but this is a very important question. So I I think it's worth touching on, again, both from a mechanical standpoint and from an ethical standpoint. Lindsey, you've touched on these. So maybe you could just touch on them again because it's possible that, Leila was not able to to hear your prior answer. Absolutely. So, again, just to to say there are two different sides to this. If you are an existing notetaker user, you are automatically opted out. And what opted out means is that no transcripts are stored beyond that period for which they're already stored for your use, which is until you lock a note or seven days. But no de identified decoupled transcripts are saved if you opt out of that saving. If you are not a current notetaker user, when if you when and if you decide to use note taker, you are automatically opted in, but you can make that choice from day one, and we will guide you in in the marketing tools about notetaker, all the information about getting started every step of the way, we ensure you have the information about where to go to make those opt outs. And, again, opt in means we are saving those de identified, decoupled transcripts to improve notetaker, and to improve products. So, if you choose to opt out, it means nothing gets saved past that time, that it's saved just for the your purpose and use. If you choose to opt out, nothing gets stored. Nothing is saved on our end after that either lock the note or seven day period for which we don't use or see or, touch those transcripts unless you actively ask for our help to do so. That, so I hope that I I don't I don't wanna give there will be a lot of walk throughs and images for you to be able to see as this comes out. So I think that kind of, like, visual, which I know I find really helpful of, like, where is it, what does it look like, so that you know opt in as we save it after that time period, seven days are signing. Opt out means nothing is stored, past that time. If from a kind of technical standpoint, Justin, you wanna add some more, clarity to that, I just wanna make sure it's clear that you will get, very clear guidance on on how to do those pieces. Yeah. I I did wanna chime in. Thanks, Lindsay. I appreciate the question about just the mechanics of this. So there are the three levels of opting out. You can do this at the clinician level. You can do this at the client level and even the individual session level. Where those are at in the product, and this will be available both web and mobile, are in extremely intuitive transparent places. And so in your settings, like the global settings at the clinician level, you will see information about where to opt out. We have tooltips included that explain what this means. And then all of this is also available through our help center as well. At the client level, each individual client you have, you will see those toggles as well. And then within the session itself, through those notetaker files, you'll see whether or not, this is retained and if you don't want it to be. And so at each place, we're trying to make it as intuitive and transparent as possible for you to be able to control, what's in. If something is opted out at a higher level, that's gonna cascade through, to the lower levels. And so you won't need to manage this, at, like, say, every client or every session level if you're already opted out at a higher level. Awesome. Thanks. We are getting close to when this, livestream will will be wrapping up here. Just technically, it's set to end at some some point in the next four minutes. So we're gonna try to get through, one or two more questions if we can, and then I'll just wrap us up as we get there. So, let's go to the next question here, which is from doctor Ritter. Do you understand the threat of AI to therapists and patients, and do the benefits outweigh the cost for therapists and patients, not you as a company? Are the benefits really only improved transcriptions? I think we've talked about where the benefits that we see here, and, hopefully, there's some of the ethical considerations. So I'm gonna skip this one for now, but thank you for the question. And the next question is from Kimberly Stewart. Please confirm the historical data prior to 06/16 will not be used for this transcript retention project. Yeah. Tony? Yeah. I confirmed, technically, not not possible anyways. But, if you're looking for more of a a legal stance, Ellie can come in as well, Tony. but, No. I mean, no. if they're if they're simply not available, then I think that's pretty much the answer. So is that right? Yeah. They've been delete deleted. We. Mhmm. have we we we are not pertaining anything today, obviously, as we're waiting for the opt out, the identification, decoupling process to be robust and ready. Okay. Great. Let's see if we can, I think I think that we're basically probably at time because I'm I wanna make sure that I reinforce a few things to people? First of all, we only have two minutes, two minutes left in this. So first of all, there will be a recording of this that is available to people. In about one to two business days, you'll receive an email, with that as Janelle just noted, in the chat. So, you can certainly review this, and we'll make this available also, to to people to view. I believe that will be available probably in our trust center, but please someone help me confirm on that. And, also, I I encourage you, we have help center articles. Please do visit our our trust center on this. There's lots of articles about how we go about, all of these topics that you've, you've addressed today and, have many very valid concerns about how we're doing it. And we will continue to try to do our best, to serve you as, as to the best of our ability and in the the most transparent way we can so that everyone understands, how we're approaching all of these very important topics. There's an FAQ page that Isabella just linked in chat there, around our transcript retention policy. And as always, thank you very, very much for spending your time with us today, bringing your questions and concerns. Again, having this dialogue with you is super important for us. It's important for us to hear directly from you and understand exactly, what's going on in your minds and as you receive some of the decisions and and, policies that we put out there. So, again, thank you very much for your time. We appreciate it. Please do, reach out to us with any further questions, concerns, and we will be back soon, to hopefully have more conversations about this and other topics as you see fit. Thanks everyone and thanks to all the panelists for being here today.