Name: Is Your Mental Health AI Safe Enough for Your Workforce?
Uploaded: 2026-09-22T18:16:28.256Z
Duration: 55 min 14 s
Description: Is Your Mental Health AI Safe Enough for Your Workforce?

Transcript for "Is Your Mental Health AI Safe Enough for Your Workforce?": Hello, everyone. Good morning if you're on the the West Coast and, good afternoon if you're on the East Coast or further away. I just wanna thank you for joining us today as we talk about the safety of AI and mental health. And I'll go through introductions in a moment and set the stage for our speakers, but I just wanted to, share a little context. AI is moving so fast and it's really hard to make sense of, the headline news and understand what that means for us. So as employers and consultants, it's really important, to understand what to look for and how to evaluate an AI tool before we make it part of a mental health benefit we offer our teams. So our purpose today is really to talk about what what does safety mean in this context and had empower us all to evaluate solutions so we can be confident in what we provide to our teams. If you go ahead to the next thing, I just wanted to share a little bit about Sequoia and Spring Health, the two companies that have experts who are here to talk with us today. Most of you know that Sequoia is a, total compensation and benefits provider and advisor, and we help employers really make the most of their people spend, and they maximize the potential of their employees and the the potential potential of their health and well-being. We believe benefits really need to meet the real needs of their employees and that these benefits should be easy to use and access and empower employees to feel safe, throughout their lives and their careers. So our partner at Spring Health, we've been in touch with them for years, actually. It's a global mental health company and their mission is really to eliminate every barrier to mental health. They have a staggering 170,000,000 people getting mental health, support through Spring Health that helps them improve faster and it helps employers see real savings. Together, we want to make evaluating AI and mental health easy and practical. So you don't need to be an AI expert. I'm not certainly. But we need to be able to ask a vendor good questions and, make a responsible decision for our people. So I wanna set the stage a little bit. If you go further and say how what's the scope of this? Almost half of US, adults report using AI chatbots. That's that GBT, Gemini, CoPilot, and there are many. And nine percent of US, of those, have used it to talk through personal mental health concerns. And that is 23,000,000 people. It's sort of amazing to consider that translation, but that's a lot of people. And few of these tools that are being used have gone through any sort of safety testing. So, the gap is why employers need to set the standard for, help them look beyond what a vendor promises and actually ask for evidence. So I wanna talk about our speakers today. Let me introduce you to our speakers. I'm joined today by, my friend and colleague, Shannon Arens, who's the director of well-being at Sequoia. And we're here with Will Alexander, who's the senior program manager and the clinical product, at Spring Health. And Dan Harrah, who's Spring Health Health's vice president and of clinical partnerships, and he's a licensed clinical social worker. And I'm Diann Lawson. I am the impact program director for Sequoia. So before we get started, I'm interested in doing a poll and seeing what everyone comes up with as their top concern. So if you go to the polls tab next to the chat box, you can submit your answer. What is your biggest concern about AI chatbots and mental health benefits? Go ahead and, vote in the poll. Okay. Are we able to see results? Okay. Let's take a look. AI replacing human clinicians is a big one. Harmful or inappropriate responses is a big one, and data privacy and security. So these are all really, big concerns, and I really wanna make sure we get to the meat of, addressing them. So with that, Will, let's start with the basics. Can you share with us where is act AI actually showing up, today? Thanks, Diane. First, like Diane said, I'm I'm with Alexander. I am a licensed marriage and family therapist at Spring Health, and I'm on our clinical product team, meaning that I I partner really deeply with our product and engineering teams to bring the clinical lens to the things we're building and to help ensure our AI is safe. I think today, I I really wanna start with just one simple point and and that AI and mental health is not just one thing. The risk really depends on what the tool is doing and and how close it gets to a person's mental health care. Just like almost every other field in industry now, AI has entered mental health care, and it's supporting both the people that are getting the mental health and those providing it. In AI is increasingly helping to identify the right care as a a person I need, and assess the level of urgency. AI also helps clinicians take notes during sessions, summarize them to speed up their admin work. This is typically only provider facing, and the patient never really interacts with this aspect. For the more sensitive use cases, like an AI chatbot or an AI tool that talks directly with someone about what they're feeling or what they're going through. When a person is anxious, depressed, or in crisis, that conversation can really shape what they do next, and that's where, the safety needs to be much, much higher. So, I mean, AI is showing up in all kinds of different ways. They show up in journaling or meditation apps. They're showing up at general chatbots, and also they're showing up now in in clinical support chatbots. And those are not equal when we talk about AI safety. They may all use AI, but they carry very, very different responsibilities. When a tool is used for emotional support, it needs to be able to recognize risk, respond in a way that doesn't make things worse, and it needs to connect them to, human help when needed. That's the distinction that I think employers really need to understand is that the question is not simply whether a tool is using AI. The question is what the tool is asking people to trust it with. And I'll show you this next slide where we, have scored, the 10 leading general purpose models. So we have things like OpenAI, Cloud. And we used, our open source, safety benchmark that I'll talk about a little bit later called Veri mental health. And you can see the score here that the range is from 25 to 64, and and then the score is out of a 100. The most important thing here is not just the ranking. It's it's where people are failing. So it's the gap in guiding people to human care. None of these models scored above 30 in that area, which is a little scary. You can see more of the detailed, background on the scores on bearmh.com. Just a little plug. The reality of this shows that a tool, though, can sound supportive and still fail at the moment that matters most, hoping someone get to a a person. These general purpose tools are not built to to manage a mental health crisis. That doesn't mean that AI can't help in mental health. It just means, the safety bar has to match the use case. So when we're asking a question from the AI tool and get a wrong answer, in a lot of settings, and we do this every day, I was using Chagibuty earlier. It was giving me terrible answers, and I started typing in, you're you're wrong, and it was really frustrating and annoying. But in an emotional setting, getting the wrong answer can be dangerous. The tool, they missed a chance to connect someone to care. It may sound, overly confident while reinforcing an unhealthy thought, or, it may respond in a way that creates harm instead of helping. And and these are things that we've seen, tragically, unfortunately. And then, these are not abstract concerns. These are these are failure modes that we have to test for, and this is the reality of the things that that we need to test for. I I wanna clarify something, just real quick, Will. Yep. I'm I'm seeing a few, Yep. questions in the chat from John. And, John, I wanna... I appreciate you asking the question. He's asking, is... Has VERA m h been tested in the real world or just in simulation? And I think I did a bad job of answering his, his question. I know you work really closely on on Verra. But to help, John understand how we how we are testing Verra to look at the the type of safety here, I know you're gonna get a little bit more depth here, but I wanna make sure as you continue through, that you're talking a little bit about how we test it, how it is being tested, and, how it's being used in the real world by different organizations going forward as well. Yeah. Great. Thanks. To to answer John's question first, it's it's we've we've tested on simulated conversations, and that part was really to get we also brought in providers, clinicians that were able to get some kind of iterator reliability into what our judge LMS were were spitting out. So we first wanted to do that, and then we started to to to test on our on our real data. And so, we and our real on real conversations. And this is where, again, I think there's more on on the website, but, we scored some of our own internal products, and they they scored at at minimum 20 points higher than general LLMs. It's also important to note at Spring Health when when, RAI, notices someone in crisis that we we immediately, involve a a human being, a a clinician in the loop at all times. So, John, I'm not sure if I answered your question, but, feel free to type some more, and then and I'll and I'll get to you some more. So AI is already shaping the mental health benefit experience, either through a feature or tools that, y'all delivery deliberately choose or if there's something that employees are using on their own, like, or Claude or or the many that are out there. The question is not simply whether a tool works. The question is where the weather it is safe when a a member needs it most. And I'm gonna pass it back to Diann. Yeah. Yeah. Thanks, Will, for that insight. That's really, really, important to to learn. And I wanna look over to Shannon and say, you know, what employers are seeing. So, Shannon, you talk with benefit leaders every single day. What are you seeing and how AI shows up in mental health benefits, and what are the questions that you're hearing from employers? Yeah, thanks Diann and hi everybody. As Diann mentioned in the beginning, my name is Shannon Arens. I'm the Director of Well-being at Sequoia, and my role is really to work with our team of benefits consultants and our clients to support them with their overall well-being strategy. And that includes mental health and I also have a personal background in mental health where I went to grad school to be a mental health clinician, so this is also a topic that's really important to me. So you asked, you know, what are we seeing from employers or what are we hearing from employers right now? So right now we're sort of seeing this sort of two layer dynamic, take hold. So starting first on the bottom layer and we'll already mention this. Employees are already using AI on their own. Diann, you mentioned in the beginning, that half of US adults, almost half of US adults have already used an AI chatbot, and many for emotional support. So for the employers on this call, this is happening whether or not you've made a benefits decision about it. And then we have this sort of second layer or this top layer of what we're seeing vendors doing. So in my role, my team, we're talking to mental health solutions, across the marketplace. So we're seeing AI being integrated and showing up in a few ways, as a first line of defense for that in the moment support for care navigation and increasingly as these standalone AI chat bots and coaching tools and that's what employers are really considering to be part of their mental health mix. So when we talk to benefits leaders, we are hearing questions that do suggest they're being really thoughtful and also cautious about this. They're they're asking, is this clinically validated? Does it work? What happens if something goes wrong? And how do I know that this is actually safe? So these are real questions and and they make a lot of sense and I'm glad they're being asked. What I'm noticing is that employers are asking these questions partly because they're already experienced fragmentation in their mental health portfolio, and they don't want to make it worse. So earlier this year, Sequoia published our 2026 well-being trends report, and we found that 74% of employers told us they're unsure whether their mental health strategies strategies are actually improving employee well-being. And partly responsible for this is that most employers have actually layered multiple solutions. So in that same study, we found that 63 have two to three mental health vendors that they're partnered with, some have four or more. So, you know, when it comes to measuring success, you know, really the best they can do is look at utilization, not really kind of understand the whole clinical outcomes perspective. And then the last thing I want to mention here, to sort of this to double layer is this readiness question about AI, when it comes to how benefits leaders are thinking about this. So our same report in our data shows that 43% of employers want AI to reduce administrative burden for them. So no surprise there. Our HR and benefits leaders are very busy and 31% want to improve experience but only 9% have actually implemented an AI driven mental health tool at their organization. So we can kind of see that there's this gap between ambition and readiness and sort of decision making as well. Yeah. Wow. That's really, important data. Thank you for sharing that because it really helps inform, some of the decisions that we make. So do you think this goes beyond being a technology or a a clinical issue and it extends really to being an employer issue? Yeah. I do think that this is a, coming up as a real business issue for employers, and I think for a few reasons. The the three that I'll call out is employee trust, employer responsibility, and then employer employer reputation. So I'll break each of these down. So first, employee trust. This is really important. If an employee is reaching out for what they think is clinical support when they're struggling and then they get a response that's not what they need or worse it's not helpful, that really does affect how that employee sees the whole benefits program that is offered by their employer. So it's not just about that one tool, it's about whether their company showed up for them, when they needed it. The second, reason why this is coming up as a business issue is that employers do take responsibility for what they put in front of their people. So when you put a tool in front of your employees, you are making a deliberate choice, and that choice comes with the responsibility to really understand what is offered and what the employee experience is from end to end. And from a procurement standpoint, being able to point to what has actually been validated or reviewed before you buy is really foundational. So benefits leaders we know are being questioned more than ever about their spend on benefits and having this information to say we've done our due diligence is really important. And then that third reason that I mentioned, is that if something goes wrong it does reflect on the employer, unfortunately. So if something goes wrong with an AI mental health tool, for example, if it misses an escalation, point or delivers unhelpful advice, it does become a question about their the company's values and judgment from the employee perspective. So that's a real consideration in today's talent, environment. And then the last thing I'll say on that, there's the timing layer too. Many employers, and we know this, firsthand at Sequoia, are dealing with medical renewals that are hitting them with double digit increases. It's really putting a crunch on budgets. And so when costs are rising everywhere else, any additional line items become much more physical and more scrutinized. Yeah. Well, that is all certainly true. And what makes this so hard for employers right now? Yeah. I think it really comes down to the language that is being presented to them by these solutions. You know, I think it can be confusing. So a vendor, a mental health vendor who say or it could be any vendor that's integrating AI, that's saying our AI is safe during the sales process. It's, you know, it's not always the same or doesn't always equate to a vendor actually doing the work to make sure it is safe. A responsible vendor really should be asking themselves the hard questions, during development. So how were the how is this AI tool trained? What are the data sets that it's, being trained on? Who's reviewing those outputs? What happens when the tool is uncertain or when someone is in crisis? Again, it comes back to these escalation points. So these are the kind of questions, that you should be asking if the vendor is asking themselves first. And I think that's really important, but it also becomes why it's so hard also. Yeah. Certainly. Thanks for that. Well, I think that brings us to the practical question is, what should safe enough actually mean in the context of mental health care? Yeah. Great question. I think I think safe enough has to be really, really specific. And for us, it it comes down to four things. First, the tool has to be continuously tested. I mean, we're we're at a place where the technology, is changing every day, and that's not even being hyperbolic. It's it's the technology is moving at a pace that only we've ever seen before. So, anytime there's a model change, it that that we have to test for known failures in edge cases. So a one time review is not enough. Second, a a human in the loop. A named clinician should be reviewing flagged conversations and should also be able to contact the user when it's appropriate. Third, testing should reflect the real ways people talk about risk. People don't always use clear and direct language with their when they're struggling. So, the tool should be able to disambiguate that risk, when it appears. Lastly, tools should be measured against public's published and public standards that others can review and benchmark. These four points, are the foundation to, I think, to give employers something concrete, to evaluate mental health AI tools and why we created, Spring Health Mental Health. When Spring was founded and Spring was founded, on this idea of precision mental health, and using machine learning to, identify and get folks in the right care quickly. At that point, there was there was no shared way to access or assess safety in mental health, mental health AI, so we helped build VeriMh. So VeriMh is an open source clinically validated tool for evaluating the safety of AI conversations about mental health, so specifically about mental health. It gives everyone a standard to evaluate the mental health safety of AI conversations. The goal is, we think pretty straightforward is move the conversation from from broad promises, to actual evidence that people can review, and the vendors can review. So how how does VeriMh work? VeriMh looks at five practical things. And these things are like when we were initially coming up with trying to to to, shop the ideas and bring in experts, we're like, we we we think that, things are things that are are in our face every day, the things that we often overlook are the things that we need to really focus on. So does the chatbot detect potential risk? Right? So people, like I said, don't always speak directly about risk. So they'll come into a chatbot and say, you know, I'm really I'm not I'm feeling like I can't do this anymore. Can the chatbot identify that that's potential risk? And then does it ask the right follow-up questions when it needs more informations? So does it detect risk in those in those ambiguous and edge cases? And then does it confirm the risk? At that point, does it guide the person to human care? AI is not ready to handle, a human in crisis. So does it guide it to a person? And then does it hold a supportive conversation in that moment? I think for that is really important to us. And and does it and at the same time, does it remind people of the limitations of AI and avoid feeling potential potentially harmful behavior? These are, I think, pretty simple questions on the surface, but they're all really, really critical when someone is sharing something difficult, and you can't be really good at one, and fail the next, particularly when when we are meeting people in their most vulnerable states. So VERA test these questions in repeatable conversations across a 100 personas, and it doesn't rely on a demo or or a broad statement that a product is safe. I'm gonna turn it over to Dan, who's actually gonna walk us through what this looks like, in practice. Yeah. Thanks thanks so much, Will. What's what's so interesting to me about developing the the standard for for VERA was how we started and and what we wanted to sort of develop initially. We thought about the highest level of risk. And by the way, my name is Dan. I lead an area called clinical partnerships here at Spring Health, licensed mental health clinician. Started my career, long time ago in inpatient psychiatry, dealing with folks who were sort of the sickest of the sick, folks who had had, issues with, severe chronic mental illness or status post suicide attempt, things like that. And and generally speaking in mental health care, the first thing that you always wanna do is assess for for safety. And what we built at VERA when we built the safety standard, the first safety standard that we went to look at for VERA was identifying that highest risk situation, which is the risk of of suicide, for for individuals utilizing a mental health tool. Or, interestingly, VERA can actually be applied to non purpose built mental health tools. So things like, CHAT GPT or CLAWD, etcetera. You can actually run that VERA standard on the source code and take a look at exactly what Will just highlighted. Is it good at identifying the risk? And then if risk is identified, how quickly can you get somebody to human care that is a qualified human care or a licensed clinician? Those are those are the really big, guardrails that we built with Verra initially, and we continue to develop Vera into things for, like, risk from others. So for example, you're in a a domestic situation where there might be a threat of violence or there is violence occurring, things like that. So we're looking to continue to develop Vera as a safety tool, but I wanna bring it back to... Okay. Spring Health is a purpose driven AI product. Right? And what we're trying to do is ensure that when we do build AI bots into our product, that it can actually pass that standard that will establish... That, is aligned with the guardrails that that will establish through Vera. So what we did is we actually applied the Vera framework and the Vera testing to our our specific source code here to make sure that we are hitting that high bar and and sort of outperforming, significantly outperforming those typical chatbots. And there's a couple of reasons that I'm gonna go into in a little bit more depth here in a minute. But what we built was something called Guide. And Guide is essentially built within our platform, within the user interface to help guide members to exactly the right care for them at any given time. So a member can actually interact with this chatbot, and it's really important to us that we let the member know that, hey. You're talking to a chatbot right now. This isn't a live human being, and here's what it is here to do. What it's here to do is to help you get to the right care for you, but it is not here to replace the care of a licensed clinician. That's not what we're claiming guide is here to do. But it can definitely help members, and it can help our clinicians as well, the providers within our network, to deliver better care to individuals and help identify what could the right care be for an individual. So guide can do some of that subclinical work like breathing exercises or point you to digital resources. And then in the event that, you know, you may need to schedule with a clinician, a coach, or a medication manager through our platform, it can help you with that scheduling and some of those administrative burden pieces. And then we've also built AI components into the platform that help our providers with note taking, with next steps, right next steps, clinical, and identifying member risk and things like that. So what I want you to remember on guide is that it's not there to replace clinicians, but it's there to make our clinicians more effective. But getting a little bit into the safety conversation on guide and and making sure that, that we're doing all of those things that we sort of built the framework on Verra around, I wanna show you on the next slide just a real world example of of how, the safety layer within what we built at Spring Health, would work. So you're seeing a few different screenshots here. On the left hand side of the screen, you're seeing a member making a very direct statement. I've been thinking, about it a lot, actually. I have a plan. I was thinking tonight. As Will Alexander mentioned earlier, most members are not that direct all the time about about what they're doing. But for us, this would flag very quickly within our system the opportunity to speak with a licensed clinician immediately. We say, hey. Based on what we just heard, you could see guide interacting with the member. We surface our 247 support line, and that actually goes to a licensed mental health clinician, a master's level mental health clinician immediately. It's not simply going to 988, which is a great resource, by the way. It's our national suicide, crisis hotline and certainly a great resource. But what we're doing here within Spring Health is we're actually connecting them directly to licensed clinicians within our platform. And if the member elects not to call, because we say, hey. Call us. Press option 2. You're gonna talk to a licensed clinician within thirty seconds. Right? So this is the AI connecting somebody immediately to a licensed clinician. And if the member elects not to call, we actually have a licensed clinician proactively outreach to that member in that situation. So, essentially, when the tool identifies the risk within our platform behind the scenes, we're sort of stratifying the the risk level based on what we're hearing from the member, to Will point, verifying the level of risk. This is a member who's stating they have a plan, think... And, like, a plan... Immediate plan and and getting that crisis response to happen immediately in the moment with a human licensed clinician in the in the loop. So we're gonna ask directly. We're gonna have our AI chatbot ask these questions directly and then get somebody connected very, very quickly to a licensed clinician who can help deescalate the situation and ideally get the person, on a good care plan for them, which may or may not include emergency psychiatric services depending on the assessment of the licensed clinician in the moment. Very, very often, we're able to deescalate the situation and get them connected to an immediate same day appointment with the therapist to help build out that safety plan and make sure that they're working toward, you know, feeling better very quickly. Yeah. Thanks, Dan Harrah. That example makes all of this so much more concrete. So if you're evaluating an AI tool and you don't need to understand every technical detail, but you should be able to get some clear answers to basic questions. So, Shannon Arens, I'm gonna turn it over to you. Let's start with the first question employers can use with vendors, like, which is, you know, how do you test for safety? Yeah. So, I'll kind of answer this question from sort of the consultant's perspective of what employers should be asking of vendors when you are either evaluating or you know at the stage of the RFP and evaluating a number of different solutions. So, I'll kind of highlight a few questions that you should be asking. And I think the most important thing is that, when we get into the world of AI and, you know, products and engineering, really important that the vendor is explaining this in plain English. And also not in marketing speak. They're being really, straightforward about how they're doing the testing. So that would be first question. Ask about testing. So, your RFP or you're asking this in the sales process, you're asking these questions. What kind of situations are they testing for? Are they testing for the hard conversations, not just the easy ones? Are they testing when the model changes, you know, they make product updates, or did they just test it once at launch? So getting the the vendor to really explain what they're actually testing, is really the first filter here. The second question I encourage you to ask is is how, how do they go about finding problems? So if they do find an issue, what happens next? So Diann just walked through us sort of that their, very clear protocol for what happens when an escalation is detected. How quickly can they fix it? What are their timelines for responding? Who gets alerted? A good, answer isn't that this never happens and hopefully vendors aren't saying that because that that would not be practical, but really thinking of answers like here's how we catch it, here's what we do about it. And then that third recommendation I'd have for you to ask is what are the guardrails? So what, when does a human review a conversation? What triggers that review? How quickly is it reviewed, and so really thinking about you know what are their what is their bar for safety, and how they're able to answer that. So those would be the questions I'd recommend when when evaluating a solution. Great. Thank you. And, Will, can you share with us can you show independent results and what a good answer would sound like? Yeah. I think I think, again, it's like, a good answer includes evidence. I think so much what Shannon just said, like, we we we want we don't just wanna set a sales deck in a sales pitch. We need to see, the evidence, you know, how the tool is performed against a a published standard. It should explain, like, what the results mean and where the gaps are. And the point, like, really, I think an important point is, like, that every tool doesn't need the same score. The point is that buyers should be able to see the evidence and understand how the the vendor is measuring safety. Yeah. Right. So if you take these, three questions away from today, take these. How do you test for safety? Who reviews high risk conversations? And how can you show independent results? Those questions will not tell you everything, but they'll certainly help you get past the broad promises and go into some details that actually matter. So what I wanna do now is turn to questions from today's attendees, and I'm gonna go to the q and a. And feel free, if you have a question, type it into the q a tab, and we will, go ahead and answer those directly. Yeah. I I actually... And and, Diann, I see a great question from John. And, John, I I stalked you on LinkedIn a little bit, by the way. And so you certainly are highly, highly credentialed in this in this area and, have a have a great point of view. And perhaps we could work on a webinar together at some point. But John asked a question about considering running our AI, so Spring Health's AI guide through an independent evaluation. And, you know, given that we built the the scale and the AI that we're talking about here, absolutely, we'd be very open to that, and this is a question we get a lot. I'll often have benefits leaders or academics such as yourself say, hey. Is this the fox guarding the hen house here? You know, Spring, you're talking about Vera, this independent, this, independent open sourced evaluation benchmark for AI safety, specifically around suicide risk. But would you allow for a third party to come in and also evaluate what you would do? Absolutely. We would open that up. We want Verra as well, the standard in Verra to be super open source. And we sought a lot of feedback from industry experts to even get to the point of what does good look like for for suicide risk assessment and safety. And so be happy to send you as a follow-up a lot of the work that went into that in evaluating the VERA benchmark as well as taking a look at the VERA benchmark, online. We can send you the links there, on those pieces. But we actually wanna welcome that type of scrutiny. We wanna welcome that type of scrutiny that comes from academic peer review, that comes from independent third party validation of these tools, and we would actually encourage the industry to do the same. We we don't really believe in saying, hey. We have a proprietary, method for evaluating the safety of our AI product. We think that it should... We should invite that level of scrutiny, and we would be very open to it. And I would love to kinda continue that conversation. Yeah. That's great, question and answer. I appreciate that exchange. And so there's another question, that I wanted to bring to the team is how can we convince founders and executives with purchasing EAP plus AI, it's not redundant and it can be helpful in decreasing employee leave of absence and higher medical premiums, etcetera. Who can respond to this question? I'm happy to take a first stab here, but then I would love, Dan or Will to chime in on really kind of the deeper ROI, conversations that can be had here. So we work with our employers all the time on really putting together the case. And I think what Shannon's asking here is kind of a dual solution, approach or perhaps maybe AI here is an add on. Just trying to interpret the question here, but that's kind of how I'm interpreting here. But I think the best thing here is thinking about the different ways that employees are seeking help and not everybody might be ready for either in person, face to face, mental health support. You know, while we've been talking about mental health for a long time, much more openly, there is still a lot of stigma related to this. And I think this is where an AI tool, or an AI chatbot can really help those who maybe are still, experiencing, you know, the the anxiety or fear related to that stigma. So I would I I guess what I'm trying to say is put together the case to say this is how we can meet the needs of all employees no matter where they are with their mental health, in their mental health journey. Tiffany, you also asked about decreasing employee leave of absence, higher medical premiums. Actually, I learned this from Spring Health, and so they can probably speak better, but I'll take my first stab at explaining it. They've done a lot of research into think, looking at how, those who are struggling with their mental health, there are also impacts to physical health as well. And so oftentimes when an employee might be experiencing a mental health crisis or something fairly acute, they might go, to the emergency room or urgent care as their sort of first line of treatment because they don't know where else to go, which then increases cost to the employer because there are lower, more cost effective ways such as a mental health solution to address these mental health, needs. But I'll pass it to Dan and Will. I'd love for you to stand on that. Yeah. Shannon, you did great. You did really great. That... That's absolutely right. Where we see the most of the savings... And by the way, Tiffany, this is such a fabulous question right now because my goodness, our benefits leaders up against it in terms of every conversation that they have about vendor purchases right now, about investing in employee well-being, and really needing to show that CFO, show that executive the value of every single solution on the on the market. And I was actually at a dinner with some consultants about a week ago, and they were like, Dan, we are staring down 15 to 20% increases in LOA and disability in medical spend in stop loss coverage for catastrophic high cost claimants, for both medical and behavioral health spend. But what what Shannon brought up is is really kind of the key piece here. It's the the connection between physical health and and mental health. And I I I can explain it in a really, really simple way. If you are struggling with a mental health condition, literally everything in your life gets harder. Right? And so that includes taking care of your physical health. And there is a great bolus of research out there that shows that when an individual has a chronic physical health condition, something like high blood pressure or diabetes, for example, COPD, something along those lines, that when they have a comorbid mental health diagnosis, that it actually increases their claims cost over the course of the year by anywhere from three to six x. So when you're seeing this trend, this, like, extreme trend in physical health cost based on people just spending more on their their physical health, physical health. We know that one of the very, very largest drivers of that increase in trend is untreated or undertreated mental health support. And so what we've actually seen with our data, and we've actually published this in JAMA, which is the journal of the American Medical Association, that fast access to high quality mental health care actually significantly drives down the overall spend within your health plan. And so what we do at Spring Health to kinda help make that case a little bit more convincing to the CFO is we put performance guarantees in place where we'll actually review the total claims cost for a customer, and we'll look at members who use Spring Health versus members who don't use Spring Health who have been matched on medical risk scores. And we look at the overall medical cost of those individuals. Individuals who use Spring Health, even when you take out the cost of the Spring Health Health benefit, the the difference of the cost in the medical plan actually makes up for the cost of providing this excellent member experience, which gives you fast access to high quality care. So that's where a lot of our research has been, and we're really happy to work with you on any type of business case that you'd like to make on this. But it's it's one of those rare benefits where you can actually do something that's really good from a member experience perspective, but it also will benefit the organization from a cost containment perspective. Yeah. Those are really deep and helpful insights. I I so appreciate that. And I just wanna take another question while we can. Is, what should a smaller team focus on if it does not have a large procurement function? Yeah. Diann, I'll I'll answer this one. So we at Sequoia, we're working with groups of all sizes but really our bread and butter are the smaller groups and so we come across this a lot where a lot of our clients don't have a large procurement function. In fact, sometimes there are just one person, doing all the things in HR and benefits. So really I would say lean on what you've already got, lean on your broker, lean on your medical plans because they have the procurement functions, that are doing a lot of the vetting, when they think about their partnership. So definitely, you know, use their consultation, ask them if they've done security reviews, ask them if they've gone through compliance vetting. All of these questions, your broker probably can answer if they've if they've leaned on, the solution as a partner. And so one thing I'll say too at Sequoia is we have our well-being bundle program. If you're not familiar with it, this is where we have 14 vetted partners, one of which is Spring Health Health, that we've already done all of the work for you. And in fact, many teams don't even need to go through procurement to purchase a well-being bundle. So I'll just throw that out there. There are other partnerships and medical plans that to also to have partnered with solutions like Spring Health Health and and other similar solutions who have also done the vetting. So I think there's, especially in this medical and high benefits renewal, environment we're in, leaning on which you've already got included and, easily accessible to you is a great strategy. Yeah. I just wanna. plus... Go ahead. I I just wanna plus one what what Shannon said there. Consultants are your friend, you know, in in these types of situations because you have experts like Shannon, you have experts, who can help you there. And there are often partnerships like Spring Health, for example, we have partnerships with Guardian, we have partnerships with Workday, we have partnerships with Alight. So there are plenty of vendors out there. You can ask those existing vendors. Hey. Do you have a partner, and are they a high quality mental health solution? I happen to believe that Spring Health is the highest quality mental health solution as it relates to, like, our published research and results. But, certainly, look at those existing partnerships because it might be a little bit easier to contract and paper through those existing partnerships. Yeah. Nice. I'm I'm already this is fantastic. I'm feeling more confident myself already. I wanna go to another question is if, how can you, as an employer, tell whether clinically validated is a real claim or just a marketing phrase? I'm gonna point that one to Will. Thanks, Dan Harrah. Yeah. No. I think you have to look at what what what are people what's the evidence. Right? Can you point to clear evidence that you are are doing the work? People are selling people are very good at selling what they are building. They and they will give you a great sales deck. They will give you all the vibes in the world, with very little safety backing. And so that's that that is, like, what's clinically validated? Did you publish something? Can you point me to the research that that is saying that this is helpful? These are the questions that I would be asking. Yeah. Yeah. I go ahead, Dan Harrah. I was just gonna say, I think another question is, you know, who are the licensed clinicians that are working on it? Is there a Will Alexander there who's a licensed mental health clinician who helped build this thing? Or, you know, have you been published? I think it's, like, a really, really key one that that will hit on there. Like, it's one thing to have a a third party white paper or an internal white paper. It's another thing to go through the process of academic or medical peer review on any of these claims that are occurring. So just dig for those receipts, and I think those are really, really key receipts, that that level of academic peer review on some of these outcomes and claims that that companies are putting forward. And I didn't mean to interrupt you there, Diann. I'm sorry. No. I mean, totally. I'm glad that you did because you shared, once again, a lot of helpful and important information. So another question is, you know, how can an employer tell whether clinically validated? Oh, we already just talked about that. But if you're not comfortable with an AI feature, can you limit it or can you turn it off? Yeah. It's a real... It's a really good question. We like to put the choice in the hands of both our... So at Spring Health, for... Just for one example, you know, we do have AI features within the platform, and we wanna make sure that folks understand what the AI is and isn't doing. And so, for example, part of what drives our AI features and our ability to, give session summaries and takeaways, for example, to an individual consumer or client of of mental health, is the member and the clinician have to agree that they are open to the session being recorded and that it would be used exactly for that, for session summaries and takeaways, and and to help have memory and connection on the care plan and the treatment that's going on over time. If either the clinician or the member is uncomfortable with being recorded, you can still use the platform. You just say, no. I'd I'd prefer this not to happen. They won't have access to some of those cool benefits and those features, but they can still use the platform. So to us, it's really important that we give people choice and awareness when they are utilizing one of these tools so that they know what's going on and that we're not, sort of hiding the fact that we're using AI, but that we're very open on which of these features are AI. And then they... It's always up to them, and it's always their choice whether they want to whether they wanna use or, take advantage of these functions. Yeah. Great. Thank you. I just have another question here that I see is, how do you explain AI features to employees, without losing their trust? You know, actually, as Dan Harrah was talking, I was gonna jump in. So I'm actually so glad this question was asked because, Dan Harrah made a lot of really great points about, like, knowing like all of the precautions that are put in place so that the member knows exactly when they are interacting with AI or being recorded by AI. And I will say as benefits leaders as employers, I think knowing that that feature exists is really important for you to communicate to your employees. Where, employees are in a fairly, high stress environment right now where they're being told to use AI to be more productive, There's a lot of pressure to use AI so there might be moments where especially when they're dealing with their mental health, they they want that choice. And so that is really helpful to call out to them that you are offering them a solution that gives them the choice. If they want to use AI, they can. If they want to turn it off, they also can do that. And so it just comes down to benefits communications and being really thoughtful about what you're what you're sharing with employees. Gosh. That's so, so true. Just building and and maintaining trust with your employees is critical, as we transition and ramp up AI and mental health solutions. So I can really appreciate that. Thank you for that. Is there any other comment from other speakers that wanted to comment on Shannon's response? Okay. So here I am, I I'm an employer, and I need to know what to ask during an RFP or a renewal, something that is not being asked today. Who can share with us on that? I I think it's I think it's really, really important that when you're looking at the claims that are being made by by vendors here, that you ask for sort of a a high bar, a really, really high bar on safety. You know, how how do you ensure... I mean, I think about my daughter. My daughter is gonna be 10 years old in October, and I do not want her interacting with a tool that has not been tested by folks who are qualified to test that tool. So I think asking questions about the safety bar in this environment, because we're not talking about, you know, a a vendor that is helping you to improve your flexibility, right, or or stretching. We're talking about individuals who might be in crisis. We're talking about often youth or young people who might be in crisis. That's by suicide is the the number one is the number one way that that kiddos are are passing away right now. I mean, we have a crisis in this country. And so we have a duty of care to get people to really high quality care because, you know, suicide is also very, very preventable. If if and when you can intervene in a timely way and get somebody connected to a mental health clinician, this is something that is is critically, critically important. So I would just be asking those questions on, like, how do you ensure safety? How was it built? Do you... Have you thought about this from a from a safety perspective, or are these marketing claims that that you're taking a look at? What does the actual member experience look like in those highest risk situations, and what has the vendor done to ensure that it's been built with purpose thoughtfully by clinicians and that it's continuously monitored. So I think safety is... You know, I I started by saying, as a licensed clinician, you start with safety. That's number one. Verify that somebody's safety, then you validate the feeling. You figure out what their goals and their... What their goals are. But when you're evaluating a vendor, I I would really encourage you to work with your consultants, work with any... Work with clinicians, that that you're actually, that you trust to actually ask those hard questions about safety. Yeah. Gosh. You know, I recognize and appreciate how that it goes beyond the employee, but really takes it home to the fact that many employees are also parents, and they're looking out for their kiddos and the heartbreaking, share about suicide being so, common and prominent in the, in youth, just really points to the importance of all of this on so many levels. So I just want to go further and say, but how can, patient data be protected when using this platform? It's a really good question. It's an important one. And that was, I think, one of the biggest fears in our poll at the very beginning of this talk is data privacy and protection. Yeah, it's really critical. It's incredibly critical that whenever you're using any benefit, especially a mental health benefit, that it's super clear that we will not be sharing any any data, any clinical notes, or any of those pieces with, with anyone without your permission. So the the standard that we operate at here at Spring Health is the the same standards that are, related to any type of, care, any type of health care that occurs. So the HIPAA standards and and pieces like that. So what that means is, you know, we're not using this data to train our model. We're using this data within within the context of the member provider relationship and within the context of the treatment planning of that individual member and provider. And so it would stay just as it would stay within a medical record for an individual at that level of security. So there are ways for us to to test at those individual levels on the safety of of the tools and how how they're interacting, but we're not taking your your data and using it broadly and in any type of identifiable way. I think that's really critical. And, Will, I'm not sure if there's anything that you you would add there in terms of some of the safeguards that we put in place. Yeah. I think that the only thing I would add is that is that we have a and and some a vendor should have a really, really clear, like, role based access of system. Right? So only the the people that need access to patient data or PHI have that, and everyone else has to jump through a thousand hoops to get access to that. So I think that would be something that I would I would lean on in, and and I think Spring Health does very, very well is that only the people that need access have access, and even getting that access for those folks is, is difficult in a way that it, I think, is is the right way. Yeah. Gosh. Thank you all so much. I think this wraps up today's program. But thank you to everyone who shared with us these important insights and context and background and how we can move forward with this. I feel like I've learned a lot, and I hope, everyone here has. So thank you for joining us today. As you know, like, AI will continue to move quickly. The practical next step is to ask for evidence, to ask for clear safeguards, and and have a real human accountability before a mental health AI solution is brought to your team, your workforce. And state I'm telling myself to just like stay on top of this and keep learning and keep growing because that's how we all become better together and become really smart about mental health in the context of AI advancements. And so I just really hope that we all leave with a sense of a stronger, set of questions that you can ask your vendors and have conversations and that this, will continue to evolve to a place where you feel confident and capable of bringing the very best safety solutions to your team. And I'm gonna say to your families too. So thank you so much for joining today.