AI Chatbots in Humanitarian Crises: GOAL’s Hard-Won Lessons

This transcript was generated by AI and may contain typos and inaccuracies.

Amie Vaccaro: Welcome to High Impact Growth, a podcast from Dimagi for people committed to creating a world where everyone has access to the services they need to thrive. We bring you candid conversations with leaders across global health and development about raising the bar on what's possible with technology and human creativity.

I'm Amy Vaccaro, senior Director of Marketing at Dimagi, and your co-host, along with Jonathan Jackson Dimagi, CEO, and co-founder. Today we're giving you a sneak peek behind the curtain to see how one cutting edge INGO is innovating with chatbots and ai. We're joined by Liz Devine, head of the program technical team at Goal Global, who shares their hard won learnings from deploying AI powered chatbots In delicate humanitarian settings, we discussed what it actually takes to get a pilot off the ground, why keeping a human in the loop is strictly non-negotiable, and how to secure donor funding for these experimental pilots despite shrinking budgets.

If your organization is trying to cut through the AI hype and figure out what actually works on the ground, you'll walk away with practical strategies for innovating safely. Enjoy.

All right. Welcome to the podcast. I am so excited for today's conversation. I'm here with my co-host Jonathan Jackson, as always. Hey, John. Nice to see you It's been a while.

Jonathan Jackson: Amy,

Amie Vaccaro: Lots to catch up on. And we have Erin Quinn, senior Director of Customer Success at Dimagi, focusing on our Comcare business, she's been with us a couple other episodes, so welcome back to the show, Aaron.

Erin Quinn: Thanks, Amy. Great to be here.

Amie Vaccaro: And most importantly, we are very excited to be joined by Liz Dev Divine, who is the head of the program, technical team at Goal and Goal is a longtime partner of ours at Dimagi. And we're just excited to hear about your experiences and your work in the innovation space in global health and development right now.

So welcome to the show.

Liz Devine: Thanks so much. It's really exciting to be here. Uh, chatting with you all

Amie Vaccaro: Yeah, us too. So telling us a bit about who you are and your background.

Liz Devine: sure. Absolutely.

Um, so, as you mentioned, I'm the head of the program technical team at Goal Global. Uh, goal is an Irish organization who provides, humanitarian assistance across the world trying to reach those who are, most affected by poverty and really, suffering seriously at the community level. We work in over 13 countries globally, on a range of different types of areas, from health, nutrition, food security, market systems development, and more recently is, uh, have integrated climate resilience and, and tech into our work. so we can try to address some of the biggest challenges we're facing in the communities that we're in. I've been working on and off with GOAL since 2017, uh, and I've been working in. the humanitarian sector for more than a decade, from everywhere from South Sudan, been working on the Syria response, I was also deployed on the surge team for regional Ukraine response as well Uh, so I've covered a lot of ground and spent most of my time working at the country level, where we're delivering different kinds of humanitarian responses. Now I'm really excited I get the chance to take my expertise and, uh, global experience and work with the program technical team so we can make sure we're delivering quality programming across the global portfolio.

Amie Vaccaro: That's amazing. You've got such a depth and breadth of experience in the space. It's really awesome. I'm curious to learn a bit about the team that you currently run at Goal. My understanding is it's focused more on the innovation space, like leveraging ai.

Tell us a bit about that team and your mandate.

Liz Devine: So yeah, actually, it's a bit broader of a team than, just innovation and ai. Uh, we, have the innovation fund uh, which is actually a team dedicated to, looking at new solutions and how we can apply them across the global portfolio. And we also have a range of technical advisors, like health, nutrition, food security, uh, those sectors that I was mentioning before.

and I think it works really nicely uh, because. we have the Technical advisors that we have, they're the humanitarian experts in each area, right. And our innovation team and our digital tech specialist, who oversees that. are really critical for us to identify what are the, key problems that we're trying to solve, in our humanitarian programming, and then our innovation team, can help design really innovative solutions to those problems. I feel like it's a great way for us to blend the two sides of, the coin to make sure we can really develop, uh, innovative solutions that actually solve the problems that we're facing at the ground level.

Amie Vaccaro: yeah, I love the the combination of those two teams, right? Because innovation on its own without that expertise could be a whole mess. So can you share with us maybe is there an example of an area of focus right now? It sounds like you've working across many different sectors, really. Obviously there's so many cross-cutting elements that come together when you're looking at poverty.

But I'd love to just get like a tangible example of that kind of humanitarian expertise combined with the innovation expertise and like what does that look like?

Liz Devine: Yeah, absolutely. I'd say probably the, best example that we have is actually a solution that we developed Uh, It's a multichannel chatbot that sits alongside our community mobilization approach. uh, there's a range of different dimensions to it, but essentially we were trying to respond to.

a disease outbreak. there was an mpox outbreak in Sierra Leone and we already had a community-based intervention that was designed to mitigate the spread of the disease outbreak.

to try to contain it. We wanted to be able to scale that quickly and try to reach as many people as possible, and make sure that we're sending out the right information to them. and giving them the right information that they can actually act on to reduce the spread of mpox. that as rumors are circulating at the ground level within communities about how the disease might be spread or who has it, where they can get treatment and support.

we wanted to be able to build something that will allow us to uh, to address that challenge in real time.

So we developed, with one of our tech partners, a multi-channel chatbot, basically, that allows us to interface with the community, but also get the, that information, through to the sub-national level, and also at the national level, of the government response agency.

leading the mpox response. So for us, it was a, really interesting opportunity to develop something, you know, new in a very, time-sensitive situation. Uh, it Didn't go exactly according to the plan, uh, but it was one of our most recent attem-attempts at developing something that can be really useful in something like a disease outbreak. and for GOAL, the really cool thing is from the very beginning we were already thinking of how else we might be able to deploy that. So it's not like this is just a platform for disease outbreak, but there are plenty of other applications that now we're actually piloting, out across two or three other use cases and I think we have the potential to scale these types of use cases as we learn more about how well it functions in actual deployment, which is really exciting.

Jonathan Jackson: And, that's a great example and one of the challenges I would imagine you have and many of our humanitarian partners have is, you know, it's kind of the last place you wanna be testing and innovating. because you're working with such vulnerable populations with such urgency and and crisis, how do you think about doing this safely and,

Liz Devine: Yeah. Well, you know, so, there's, there's two things here, right? Uh, The first thing is about the speed, the urgency that we all feel to try to get something out quickly, right? Um, And that's actually something that we found when we were trying to develop this chatbot platform in Sierra Leone, is that an emergency is not really the best time to do that.

Um, with any kind of innovation, uh, there's definitely a lot of, bumps in the road along the way, and you've gotta refine, it, you've gotta develop the actual tech piece of the solution. and then there are all these other really important considerations.

You know, things like data protection, uh, procedures that you have to follow, uh, things that you have to do in adherence with local regulatory frameworks as well to make sure that we're Safely collecting information you know, that we're anonymizing it. And that also as we're developing the tech solution, it.

it's not like we're just gonna deploy something and let the, AI-powered chatbot just run on itself. you know, We have to make sure that there are certain kinds of decision gates that are layered in at key points to make sure that, you know, we're sharing the right kind of information and that if there's something that comes into the chatbot, uh, something more sensitive, that we have mechanisms in place to refer that to the right people so that we're, you know, we're not kinda letting the agent run away with itself, that we have these safeguards that are, already built in. so that was definitely something that slowed us down in the actual deployments of it.

Uh, I think it's fine actually, you know, if it's a little bit slower. I'd rather we wait to deliver a safe product than just try to get something right out the door. so we actually took a lot of these learnings on board when we were looking to a-adapt the chatbot platform for other use cases.

Uh, we're currently piloting it in South Sudan for use in a cholera response. We're adding in some additional layers, of it so we can allow it to incorporate things like climate related data in the way that we're able to analyze uh, the information. but it's better that we can really take our time with that.

um, you know, slowly roll it out and make sure that we've got the build correct and that we're-- we've identified the right points where people need to be involved to validate the information before it goes to certain trigger points.

I think that's the most important thing uh, that we have to be so careful of to make sure that we're not just rushing to get the tech out because it's, a very hot topic within the sector at the moment.

Amie Vaccaro: So I'm, I'm curious, take us through like the users of this. Either the original chat bot or the call or a chat bot, like what does that flow look like and what feedback are you getting from them?

Liz Devine: Absolutely. Yeah. Well, so and just to be clear, um, so the one that we had in, that we have in Sierra Leone, we had been at the testing stage, but didn't have it deployed at the community level yet. Uh, we're in the early piloting stages in South Sudan as well. So the ultimate users, I'll maybe go through the example in South Sudan, actually, is what the future users of it will be. ultimately we wanna make sure that we have community members that are able to report in certain kinds of climate-related issues that can act as warnings, that there's gonna be some sort of, clima-kind of climate shock, Uh, something's gonna happen where maybe flooding is imminent, and that way we can expect that around that time.

Uh, they're also very likely to see cases of cholera in some of these high-risk flood zones. So that we'd have a community member could be reporting in some of those key signals, almost like citizen science. Uh, they could be reporting that into the platform. Then we can have let's say at the district level, we can analyze that information that's coming in, has AI-supported analysis of the reports.

Then at the district level, the government agents could be using this to decide, "Okay, well, now we wanna launch maybe an anticipatory action package. We wanna make sure that people are aware flooding is coming These are the risks, whether or not they're in.

high flood zones, And what actions they might be able to take to prepare themselves ahead of time and mitigate the damage, uh, of the, you know, imminent flood." and are there any other actors that are also launching some kind of anticipatory action programming Uh, and we can try to connect all those dots together. So that's ultimately the kind of information flow that'll go through it, at the, you know, the n- community level and of course, through to the higher levels at national. um, uh, it-- we can also have back-end reporting on how the platform has been deployed, who's engaging with it, how accurate and reliable is the information that we're getting, with ground truthing from the communities feeding into the platform throughout. So there's a couple of different ways that we'll have this information from the deployment of the chatbot help, inform the design of future responses as well.

Amie Vaccaro: That's, that's so cool. I'm curious, like, stepping back a little bit on the tech it sounds like is, is most of your kind of innovation or sort of like the, innovation side of things, looking at chatbots and then within chatbots, like what makes you so like. Optimistic about the, the value of chatbots in these, in these use cases.

Liz Devine: I would say that we're developing a number of different type of tech solutions. Uh, but as the community interface and comone- uh, component is predominantly our default method of engaging, in the areas that we work, we would include the use of a chatbot, in some way, shape, or form. And the reason for that is because, pretty much everyone in the communities that we're working in.

either They or maybe a family member, somebody in there either has access to a phone whether it's a smartphone or even if it's a, an older model phone, most people will have something, right? Uh, some way of communicating. And that's a very easy way for us to engage with community members, at scale and make sure that we can get up-to-date information and reliable information to and from them quickly. And more importantly, if we can reach most people that way, then we can send out our, maybe our actual staff, the humans in the loop, uh, to go reach the people who are more vulnerable and who might not have that digital access. The great thing about the chatbots is you can have both the SMS options plus the data options too, right? Uh, So I think that's also something that's huge for us in the communities that we're working, in, that we have both of these possibilities. I think that's the reason why it's more so um, you know, kind of our default method for engaging with community members in a digital way.

And I think it's, uh, you know, one of the highest value, components for us.

Amie Vaccaro: Yeah, that makes a ton of sense. And I know that, a lot of listeners of the show they're also technologists and practitioners and so, I'm sure they'll wanna know a little bit more of the how, like how did you set this up? What kind of tools are you using, how are you evaluating responses?

Can you take us into the nuts and bolts a little bit here.

Liz Devine: Sure.

Uh, So let me preface this by saying um, I am not a tech developer, so I'm gonna borrow some, of the detail from, uh, from our digital tech specialist and our, tech partner. Uh, but for this chatbot solution, uh, one of the things they do use is to have, the conversations come through.

WhatsApp, and it allows us to keep the overall cost of the platform down. Uh, the messages then get relayed through Twilio before being anonymized, analyzed, and reported in through Monday, uh, which we have, uh, free licenses for since we're humanitarians. It's really very useful for us to use WhatsApp as there are, you know, multiple modes of communication. You can use text, voice, pictures, web links, and, uh, And so on.

And the automated AI analysis of the conversations we get generate even more insights. I think, for that, I think those are two of the really key components for, us that are allowing us to make this platform accessible and usable at scale.

Amie Vaccaro: thank you for that. I'm curious since you're quite optimistic and you're like really blazing this path forward, sounds like you're, from what I'm seeing, like you're definitely far along on this path perhaps further than others. But you're probably also coming across people with misconceptions about chatbots.

Like what are some of those misconceptions that you hear from folks about deploying a chatbot in these settings?

Liz Devine: Mm-hmm.

I think I would say the biggest misconception is that you know, people think a chatbot, for the sake of a chatbot is gonna deliver some kind of change or solve the problem. there's huge optimism and interest throughout the entire humanitarian community to do something with tech. And I think chatbots are a very easy thing that people can refer back to and say, "Oh, well you know, we've got a chatbot, we've solved the problem, We're using tech." and I think people forget about the fact that. I mean, you can't just introduce a chatbot for its own sake There has to be a really clearly defined. Problem that you're trying to solve by deploying that chatbot. And I think we you know, we have almost a tendency to lose track of what the problem is we're trying to solve just to get that tech solution out there. and, you, run the risk I think, of seeing uh, you know, a proliferation of chatbots, and then there's seven thousand numbers that commu-community members might need to know when in reality we should have, you know, one single number, that, anything can be triangulated through. you know, something that like you know, here in the US we've got nine ni- nine one one or we have three one one for more social service types of, concerns. um, We should be working towards something like that so we have, you know, we're working with governments to identify more, single, channels rather than create a whole bunch of numbers that people need to fi- to know.

in order to access any kind of information or support.

Jonathan Jackson: and of the challenges I know that we're seeing across partners is chatbots can be amazing for some use cases and then still not great for others. Or the content is too specific. And so as you think about that, need to have a one stop. Shop from a user experience perspective, you don't want end users.

I remember 10 different numbers, but some of these content areas or programmatic areas are kind of off limits to the ai. How do you think about balancing, that breadth versus which use cases you can be really confident in are working well?

Liz Devine: Yeah, definitely. I mean, there's actually something that we've seen with even, with this, uh, this chatbot for deployment, um, that we've been working on for. Mpox, is for cholera as well. Uh, You might have people coming in with really.

targeted requests for information, maybe they're looking for services that are beyond the scope of what you can you know, respond to and address with the Mpox or the cholera response, for example, so for that piece, we're looking at the referral pathways that we need to pay attention to. Uh, we've identified and mapped out in the same way that, in the humanitarian sector we do partner mapping, or service mapping. We've done that already, and that's a, layer, right?

Uh, So we have a secondary agent behind the, chatbot that can go through and based on the up-to-date information that we have on the service mapping, can share information about that.

Uh, So that's, how we'll look at that, and say, "Okay, so this is the very clear scope that we address with this Mpox chatbot." if you get questions, excuse me, about other issues, we, can refer you to this other service. And then if you get something that's more sensitive, we can link that.

So that way it's like, okay, well, these are the government services that are available for that concern. Um, if you get a really serious concern that comes in, or maybe somebody is reporting, an issue of say, violence or, you know, sexual violence or something along those lines, immediately have that trigger to a human in the loop.

So if something came in that's serious and sensitive and it needs to be addressed, but it's well beyond what the the ch- the Mpox chatbot can do. for example, that's where we have these trigger points that are really important to keep people checking in on those types of comments, to make sure that things aren't slipping through the cracks and that, especially if it's something-- somebody who uh, maybe needs, a very sensitive type of care or support, um, that we're able to at least activate some kind of pathway for them to get it, whether it's from us or from the other government, or partners that, uh, actors that we're working

Erin Quinn: Can you talk a little bit more about like the, the human side of this, so. it is obviously sounding to me like you know, the chat bot is kind of chat button day to day, but then you realize that there needs to be this additional layer of some human looking at this helping pull out whether or not the system is identifying these really critical cases that might need a human eye.

But I'm curious a little bit what that looks like and even just kind of the ratio of. How many humans do you need per deployment to make sure that you're kind of maximizing everything you can out of ai, but making sure that you've got that human touch.

Liz Devine: Sure. So one of the first things that we're looking at, um, you know, when it comes to the human in the loop from our team, uh, is that for each individual pilot of a project, there's an administrator of the project that is, certainly gonna be uh, you know, the person checking in and more so defining with our digital tech specialist where the human in the loop points need to be for that individual project. It's not a hard or fast rule, and, so this is where I'd be hesitant to give an exact ratio on this but because I do think it depends on the complexity of the bot, if, you have a single-use chatbot, uh, and the only thing that it can respond to, for example, is disease outbreak. that limits the number of kinda checkpoints that you need to have along the way. But if you're adding layers, like you know, what I had mentioned before about referral pathways to other services linking it into other kinds of um, you know, government services that are available, or to other platforms. or even if it's starting to cycle into other parts of the chatbot, and if it's covering more ground, or triv-- triggering different layers of analysis um, or if it's, gathering uh, a lot of information about, you know, rumors that are spreading on the ground, for example, there will be a couple of different categories, almost tiers of checkpoints that need to be in place. so I would say. you know, not a hard or fast number, but it will certainly increase know, the number of, people in the loop increases as the complexity of the chatbot increases.

Erin Quinn: Yeah, that makes a lot of sense. And probably the sensitivity of the topics as well, I would assume.

Liz Devine: definitely. Definitely.

Amie Vaccaro: I'm curious, do you have any stories about a time when a deployment or a, an effort just did not go as planned and what you learned from that or how you adapted?

Liz Devine: Well, I'd definitely say when we tried to uh, then deploy uh, the chatbot in the middle of an mpox response in Sierra Leone. it was just, you know, we were wildly optimistic about the timelines. it would take to, to get this up And running. And I would say as our. first, real chatbot that we were trying to develop, We, definitely underestimated the amount of time it would take to make sure we were following all the right regulatory frameworks, That we were making sure we were taking the appropriate data protection considerations in place, uh, had the right agreements in place, that we also had some external people take a look, and make sure that we were doing this safely.

so I think, in the end, it ultimately took much longer for us to deploy, and develop the solution and get it on the ground, um, Or, get it into, the testing stage, before deployment to the communities. And that's something that I think we really learned and took on as a lesson, that surprisingly, an emergency is not actually the best time to try and develop a new piece of tech. You have to do that advanced planning Um, and We understand more or less i-in the countries that we're working in, that we're going to be facing a disease outbreak in one of our countries of operation, you know, probably multiple times over the course of a year, So we can do that preliminary work to make sure that the solution's developed that we have all the, you know, the right, data protection considerations already.

in place. Uh, we also-- we've got all these kind of internal processes that have to be done Um, so we can do all of that ahead of time So that all we have to do when, you know, an actual emergency, um, is is happening, that we just have to adapt the solution to, to this new use case. and, and that's actually in the end, that's what we wound up doing, to pilot the chatbot for use in cholera in South Sudan, because we already had the idea in place. We had some of that, Um, you know, that groundwork already done, and we just had to, had to adapt that to a new country context. we had our internal regulatory framework resolved. We knew what steps we had to follow to do that. and so it's been much easier layering in these new integrations that we have to improve the chatbot rather than starting from scratch in the middle of an emergency. or crisis.

Amie Vaccaro: yeah. I love that model of taking what you've already. Built and then iterating on it for the next context versus like from scratch. And I'm working on AI from like a marketing context, and I'm always, you think these things are like super fast and easy, but they're not especially because it involves humans and human review, right?

And humans become these bottlenecks that these systems that can move so fast.

Liz Devine: yeah, just to add to that also too, what you end up with is not gonna be the same thing that you started with, because you have, so many people involved in the process, and the idea grows as, as the uh, as the uh, as the emergency grows, as well. Um, so trying it out on the ground and seeing which things work and which things didn't work. it's really an iterative process.

Amie Vaccaro: I'm guessing that folks listening to this might feel a little bit of maybe some healthy envy of wow, that's so cool that goal's able to do this. We don't have any budgets set aside for this kind of thing. I am curious like with, just so many changes in budgets and funding happening.

How are you able to make the case for investing in this experimental technology at Goal and yeah. Any advice for our listeners there?

Liz Devine: Absolutely. I mean, we've seen just some huge shocks in the sector over the last couple of years and, you know, the, the sector as a whole, the budgets are shrinking. And every time we turn around there's donors that are reducing their humanitarian, funding for subsequent years. So I think we do all have to be really conscious, of how we're spending the money to make sure it's used the most wisely. That's actually the reason I think there's going to be a stronger case for using AI uh, for things like chatbots and trying to be a bit experimental in the tech that we're developing, because we're still expected to solve just as many problems as we did last year or the year before. but with that, uh, you know, a significantly smaller budget and the problems that the communities that we're serving, those aren't going away. So if anything, we need to be willing to take a few more risks and try out, and develop different kind of solutions uh, so that we can continue to address as many of these needs as we possibly can, even with the shrinking landscape. What I found in our experience is that there's a lot of uh, maybe it can, be difficult to find the right funding to apply for, uh, to be able to develop these kinds of, experimental technologies.

But the donors have been really receptive once you figure it out, you know, once you figure out, that good use case and you've developed uh, something that you've got a good pilot, Uh, you know, that you're proposing. it's not a huge amount to, spend on some of these innovations. you can do a pilot for thirty, thirty-five thousand, right? it's not an insurmountable amount of money to be able to allocate. but there are also these really, great tech challenges and funds that are available for these pilot projects.

Uh, so I would say even to try.

to encourage your, you know, your fundraising business development team to find some of these, uh, new, new tech funds, new, um, you know, new tech challenge fund opportunities.

And if you have one, um, or, you know, whoever's doing your donor sourcing is to really search out some of those tech innovation grants because they could be really, uh, very good funding, and they're specifically geared towards these experimental types of tech. The other thing, and this is what I would recommend, is to make sure that you really clearly define the problem you're trying to solve.

Because with a lot of uh, different kinds of of these funds, the vaguer the problem that you're, that you're defining, the vaguer the tech is going to be. and the donors are, whether it's, uh, you know, these tech.

funders or, you know, more traditional donors, they wanna see really tangible results and ideally tangible results that they're going to be able to scale across the rest of their funding partners.

So if you can really zoom in on the exact problem that you're trying to solve, I think you'll probably be in a better position to develop a solution that number one, can actually be built, but then also, uh, number two, is something that can be scaled for other partners to benefit from as well.

Amie Vaccaro: yeah. I love that. I think it's really interesting of just like the funder might be totally different, like who's gonna fund that innovation versus that, like scaling, right? And just being clear about that. And also just getting really clear on the problem you're trying to solve. And again, I'm like focused on marketing within Dimagi, but I'm taking away some insights here from my own work too, of there are times when I'm just like experimenting, but I'm like, what is my goal?

What is the problem I'm trying to solve? I don't know. Getting clear on that can be really helpful. Awesome. . So looking five years ahead, what role do you think AI should play in chatbots as well in global development? And where should we be drawing the line about what they can achieve?

Liz Devine: that's a really great question. Uh, I Firmly believe that AI should be very deeply embedded in how humanitarian, and development organizations are analyzing data, identifying risks, modeling scenarios especially, and I also think there's a huge space for AI to be used to generate recommendations.

I think that's where we can really have an... you know, where AI and tech can really have a transformative impact on what we as humanitarians do. It's that key point about processing huge volumes of information.

No human team can actually do that at the level of urgency that we need in a humanitarian crisis.

I think we can really use AI to help us surface insights a lot faster than traditional systems.

But with that being said, and I know we've touched a little bit, uh, on this earlier, where for me, the line is that, and, and I don't think that there is a space for this to move, is that we have to have the space for those humans in the loop at the very key decision points, and, we have to be very careful about where those decision lines are.

We're, we're working here to save lives, and if we're deciding what type of emergency response to launch or which communities to prioritize or where the resources are supposed to go, those decisions carry a huge amount of responsibility, and we need to make sure that they have a lot of accountability.

People need to be the ones making those decisions, validating, uh, the decisions, and to be held responsible for those decisions. and tech, I mean, they can do the recommending, but they're not, you know, they're not, the ones, and they shouldn't be the ones to make the decisions for us.

Jonathan Jackson: Yeah, definitely agree and I think we've seen that in a huge variety of use cases all across our work internally at Magi and our partners work is, there's certainly no. Replacing any human decision making at this point and remains to be seen if we'll ever get to a point, in the future.

But certainly for now, the human judgment is absolutely critical.

Liz Devine: I fully agree with that.

Uh, We have to just make sure that we're safeguarding along the way uh, as we're building out these solutions.

Amie Vaccaro: Yeah, a hundred percent. All right, Liz. Well, we really appreciate your time today. It's been an honor to, to get to learn from you and get a glimpse into the incredible work that you're driving at Goal. So thank you for your work and for your willingness to, to share that with our listeners as well.

Jonathan Jackson: Yeah. Thanks Liz. Thanks Aaron.

Liz Devine: Thanks so much everyone. It was such a pleasure to speak with you today, and I'm really excited to see, how much interest there is in driving forward innovation throughout the sector. Thanks so much for hosting the conversation

Erin Quinn: Yeah. Thanks Liz. Thanks Amy .

A huge thank you to Liz Dev Divine and Erin Quinn for joining us today. And to you, our listener for tuning in. This conversation really highlighted the balance between rapid innovation and responsible deployment. Here's a couple key takeaways from Liz's experience.

First. The middle of an emergency is probably not the best time to try to innovate. Do the heavy lifting on data protection and regulatory compliance before a crisis hits second, fall in love with the problem, not the tech.

Don't deploy a chat bot just to check a box. Ensure it's solving a clearly defined real world problem. And finally protect that human in the loop. While AI is incredible at processing large volumes of data and surfacing insights, complex and sensitive situations must remain in the hands of accountable human decision makers.

That's our show. Please like rate, review, subscribe, and share this episode. If you found it useful, it really helps us grow our impact. And write to us@podcastatdemone.com with any ideas, comments, or feedback. This show is executive produced by myself, Ana Bhand and Michelle Lencia are editors. Natalia Gki is our producer, and cover art is by Sudan Chant.

A final note in the spirit of transparency, we use AI to assist with guest research, copywriting, and post production, so a small team can produce a high quality show. All AI assisted content is reviewed and edited by humans, and we retain full responsibility for what you hear.