Sidecar Sync
Welcome to Sidecar Sync: Your Weekly Dose of Innovation for Associations. Hosted by Amith Nagarajan and Mallory Mejias, this podcast is your definitive source for the latest news, insights, and trends in the association world with a special emphasis on Artificial Intelligence (AI) and its pivotal role in shaping the future. Each week, we delve into the most pressing topics, spotlighting the transformative role of emerging technologies and their profound impact on associations. With a commitment to cutting through the noise, Sidecar Sync offers listeners clear, informed discussions, expert perspectives, and a deep dive into the challenges and opportunities facing associations today. Whether you're an association professional, tech enthusiast, or just keen on staying updated, Sidecar Sync ensures you're always ahead of the curve. Join us for enlightening conversations and a fresh take on the ever-evolving world of associations.
Sidecar Sync
AI Agents for Associations: Revolutionizing Customer Experience | 56
In this episode of Sidecar Sync, Amith and Mallory dive deep into the world of AI agents and their transformative potential for associations. Amith shares insights from his keynote at Digital Now 2024, exploring the difference between models and agents, and why agents hold the key to unlocking significant ROI for associations. The discussion covers use cases like customer service and knowledge management, and how associations can get started with building their own agents. Whether it’s creating an efficient content workflow or enhancing member engagement, this episode breaks down practical steps for implementing AI agents that can automate complex processes, reduce costs, and elevate the member experience.
🔎 Check out the NEW Sidecar Learning Hub:
https://learn.sidecarglobal.com/home
📕 Download ‘Ascend 2nd Edition: Unlocking the Power of AI for Associations’ for FREE
https://sidecarglobal.com/ai
🛠 AI Tools and Resources Mentioned in This Episode:
Microsoft Copilot ➡ https://www.microsoft.com/en-us/microsoft-365/copilot
Zapier ➡ https://zapier.com
Member Junction ➡ https://memberjunction.org
Claude by Anthropic ➡ https://www.anthropic.com
Chapters:
00:00 - Introduction
02:10 - Recap of Digital Now 2024
07:47 - Why AI Agents Are Crucial for Associations
12:17 - Example of a Content Workflow Using AI Agents
21:18 - AI Agents for Customer Service: Klarna and Bell Canada
27:48 - Should Associations Combine Domain Expertise with Member Service?
37:26 - Who Can Build Enterprise AI Agents in Associations?
44:26 - Difference Between Automation and AI Agents
51:15 - Getting Started with an AI Data Platform
58:14 - Future of AI Agents in Associations
🚀 Follow Sidecar on LinkedIn
https://linkedin.com/sidecar-global
👍 Please Like & Subscribe!
https://twitter.com/sidecarglobal
https://www.youtube.com/@SidecarSync
https://sidecarglobal.com
More about Your Hosts:
Amith Nagarajan is the Chairman of Blue Cypress 🔗 https://BlueCypress.io, a family of purpose-driven companies and proud practitioners of Conscious Capitalism. The Blue Cypress companies focus on helping associations, non-profits, and other purpose-driven organizations achieve long-term success. Amith is also an active early-stage investor in B2B SaaS companies. He’s had the good fortune of nearly three decades of success as an entrepreneur and enjoys helping others in their journey.
📣 Follow Amith on LinkedIn:
https://linkedin.com/amithnagarajan
Mallory Mejias is the Manager at Sidecar, and she's passionate about creating opportunities for association professionals to learn, grow, and better serve their members using artificial intelligence. She enjoys blending creativity and innovation to produce fresh, meaningful content for the association space.
📣 Follow Mallory on Linkedin:
https://linkedin.com/mallorymejias
But what if you could do it in 24 seconds instead of 24 hours? Right, Customers would be a lot more pleased generally, especially if the answers are perhaps better than the average human. Welcome to Sidecar Sync, your weekly dose of innovation. If you're looking for the latest news, insights and developments in the association world, especially those driven by artificial intelligence, you're in the right place. We cut through the noise to bring you the most relevant updates, with a keen focus on how AI and other emerging technologies are shaping the future. No fluff, just facts and informed discussions. I'm Amit Nagarajan, Chairman of Blue Cypress, and I'm your host. Hey everybody, Welcome to the Sidecar Sync, your home for content at the intersection of all things artificial intelligence plus associations. My name is Amit Nagarajan.
Speaker 2:And my name is Mallory Mejiaz.
Speaker 1:And we are your hosts. And before we get into an exciting set of topics in the world of AI and associations, let's take a brief moment to hear a word from our sponsor.
Speaker 3:Introducing the newly revamped AI Learning Hub, your comprehensive library of self-paced courses designed specifically for association professionals. We've just updated all our content with fresh material covering everything from AI prompting and marketing to events, education, data strategy, ai agents and more. Through the Learning Hub, you can earn your Association AI Professional Certification, recognizing your expertise in applying AI specifically to association challenges and operations. Connect with AI experts during weekly office hours and join a growing community of association professionals who are transforming their organizations through AI. Sign up as an individual or get unlimited access for your entire team at one flat rate. Start your AI journey today at learnsidecarglobalcom.
Speaker 2:Amif, it has been a crazy, crazy few weeks for both of us. How are you today?
Speaker 1:I am doing great. You know it's been, as you said, nutty. I had such a fun time at Digital Now in DC last week and I also came back just out of gas. I needed a day or two just to kind of recover. I actually went to work for the next couple of days, but it was like I just realized I didn't have a whole lot to offer. So it was, uh, it was time to recharge. So it's kind of like oscillating between periods of high intensity and periods of recovery. It's important with physical exertion. It's also important for your brain.
Speaker 2:Yeah, I appreciate what you said about not having a lot to give, because that's how I felt right after Digital Now. I was still trying to work because I felt like there were things that needed to be done, but sometimes you just know and you're not on and you're like am I actually bringing that much value to anybody to myself today? So Digital Now was amazing. It went as well as I could have imagined. It was a whirlwind. I honestly felt delirious most of the event. I think I had passed exhaustion, passed stress and I was kind of just riding it out, which was a good place to be in. But it was fantastic to see everyone in person. The content was great. I just I'm on cloud nine still, I think.
Speaker 1:Well, you should be, because you did an amazing job. The team supporting you did a tremendous job in pulling together an incredible event. I consistently heard from association attendees how great of a time they were having, how much they were learning, how much they were enjoying the way the conference was organized, which is a really high bar to get over. You know those that are listening to this. Most of you are involved in planning events, so you know we realize, in serving you, in providing an event like Digital Now, we've got to meet the standard of your own professional excellence. So it's a tough bar to get over, but I think Mallory and team did a great job with it and we're excited about Digital Now 2025 as well, aren't we?
Speaker 2:Very excited about 2025. I realize now we have not announced that on the podcast just yet. So Digital Now will be next year November 2nd through the 5th 2025 in Chicago, Illinois, at the Lowe's Hotel, and we are thrilled to be bringing Digital Now to another hub city for associations. I love Chicago personally, so I think this is going to be a great event and I think my wheels are already starting to turn for content for next year, which is exciting.
Speaker 1:It was amazing and it's an amazing city. I think my wheels are already starting to turn for content for next year, which is exciting. It was amazing and it's an amazing city. I think being able to bring the event to DC this year and Chicago next year, it's a great combination. So we'll see how Chicago goes, and I know a lot of the folks in DC were saying bring it right back to DC after that. So we'll see. But I think we have a good rhythm building up.
Speaker 1:We had, you know, 230 people or so at this year's Digital Now, which is an all-time high for the conference. It started out with a very modest kind of mid-100s type of number for many, many years and now it's, you know well, into the 200s. We're anticipating significant growth again in 2025. And we're encouraging teams to attend together. I think that's where some of the best conversations that I participated in and, in some cases, just heard in the hallways were when you had two or three or four people from the same association coming together to learn together. It's a fun experience. It's a way to bond a team. It's also a way to get different perspectives from different types of people in your organization about key topics and thinking about how to attack problems.
Speaker 1:In a prior life, when I was running a software company, I used to bring a number of people from my leadership team and also what we call rising stars folks that were not on the leadership team but people that we just thought were amazing and so we'd bring anywhere from 12 to 15 people twice a year to leadership conferences.
Speaker 1:It wasn't digital now, it was outside of the association space, but what we did is we would always make sure that the team of people would spend an extra night in the hotel, so the conference would end and we'd spend an extra night in the hotel.
Speaker 1:And then the next morning, when we do some kind of fun team activity, the afternoon the conference was wrapping up and the next morning we'd block off about four hours, and our goal was very simple that team of 12 to 15 people would not leave the room until we had our one thing. What is the one thing that we agreed we would go and do based on our learning? So we might've learned 30 things each from the prior three days, and it was such a fun activity, but it was just a great way of really homing in on something that you committed to yourselves and to your teammates that you were going to get done, between that point in time each year where teams can come together, get really inspiring content on things that they're not working on day to day and share what they are working on stage, to when they have great case studies to share. But use it that way where they can come as a team, learn together, play together and then ultimately end up with deliverables come out of all that.
Speaker 2:Yep, I heard from multiple attendees that their brains were exploding with information, which is good. On the one hand, I'd rather it be exploding with information than not exploding with information, but it can be hard to leave the conference and then turn all those ideas into action. And I heard from one first-time attendee that he was kicking himself in his words for not bringing more team members, but it was his first time attending so he wasn't exactly sure what it was about. And that individual has already signed up for a group for next year.
Speaker 1:So I think it was a good event Yep. That is awesome. Well, looking forward to it. Chicago, for me as well, is one of my favorite cities, and early November in Chicago it could be beautiful outside or we could have six feet on the ground of snow, but we'll see. Either way it'll be a great event.
Speaker 2:Well, today's episode was inspired a little bit by your keynote, amit, the first keynote of Digital Now 2024, where you talked a lot about AI agents. So, before we kind of kick off the episode, I'm hoping you can set the stage for our listeners, for our viewers, on why this topic is so important.
Speaker 1:Well, to me, it's where ROI comes to play. The return on investment on AI for associations has to come or there will be no conversation. Right, technologies are interesting, but they're not the goal. The goal is to affect the business, which in turn, affects the mission. How do you advance what the association does? And that might be through improved efficiency. It might be on being able to provide better or different services than you've ever been able to provide in your community, but ultimately you have to be able to do that, or what's the point? And AI through things outside of the world of agents certainly can have a massive effect, but agents, to me, are the biggest single opportunity for associations and, frankly, for every other business on the planet to go attack over the next 12 months. So I wanted to focus a chunk of my time on stage talking about agents to make sure people first of all, understood what they were and then, from there, thinking about, like, how to go about taking advantage of agents in their work.
Speaker 2:And that's exactly the structure of today's episode. First we're going to kind of lay the foundations, talk about what agents are and what they're not, then we'll be talking about some use cases like customer service, knowledge agents and data agents, and then, finally, we're going to get practical with agents and talk about the hard questions how do we actually take action here and what can you do after listening to this Sidecar Sync episode? So, first and foremost, what are AI agents? They are systems designed to not only process and respond to information, but to take action. Unlike traditional AI models, which provide answers when prompted, ai agents can plan, act, observe outcomes and then learn from those experiences. This capability allows them to execute tasks independently, based on their analysis. So let's kind of dig into this model versus agent distinction a little bit more.
Speaker 2:So AI models are passive responders that analyze data and deliver answers when prompted. They can act like experts providing advice, but cannot act independently. They have fixed capabilities after training and don't improve without updates. In comparison, ai agents are built on top of models and can proactively complete tasks. They recognize when action is needed and take steps autonomously or semi-autonomously. They continuously learn from experience, improving their performance over time, and they can execute tasks like sending emails, updating databases, not just providing information.
Speaker 2:And really the capabilities of AI agents can be broken down into a few categories. First of those is planning they can determine the right course of action based on context and then acting. They can execute tasks independently. And then, finally, the next step being observation and learning, they can monitor outcomes and refine actions for future improvements. So that's the foundation of Meath. I'm curious to hear what you think about that as someone who just gave the full 45-minute keynote on it. But really I want to ask for my first question, the idea that we've been talking about models, models, models, models for the past few years on this podcast, in the public, and yet at the top of this episode you said you think agents will be the thing that can really transform associations. Can you talk a little bit about that gap in between what we're talking about and kind of what might be most transformative?
Speaker 1:Yeah, I think you know the model is the brain of an AI system and it, of course, the stronger that model is, the more you can do with it. So model advancements, model improvements, are really exciting. We've talked about large models and small models. We've talked about text-based models and multimodal models and we've talked about multimodal systems. All these different kinds of combinations of things and models are super, super important. Without models, there is no opportunity to build agents. But agents are basically systems. There are combinations of models with prompts, with other types of programming built in that can kind of connect the different models together and it might be actually the same model, like a single model working, but it's a series of steps as opposed to a one-shot kind of thing. So the way I like to describe it is think about a business process you might have. So, mallory, I know within the sidecar world, after we do the pod, each week we have a professional editor take this content and clean it up and get it ready for distribution, but then after that there's a whole content strategy that feeds off of the pod's content. We'll take the pod, we'll create a blog based upon the pod itself and then you'll typically break out the multitude of topics that are in the podcast into individual blogs and then those go out, those become part of our newsletter. In addition to that, you typically, I think, are doing social media posts that kind of take little snippets out of the blog posts. Sometimes you take little snippets of the video and put them on various things. So there's kind of a content creation and distribution workflow that needs to occur. So you cannot do all of that with a model, but you can do everything I just said with an agentic system or a multi-agent system. So let's break that down for just a minute. So you take an input like a podcast episode, which is video and audio. You take an input like a podcast episode, which is video and audio. And let's say we go to an agent and we say, hey, agent, can you take this episode and do all these things? Right, I just probably blurted out 10 or 15 different things. And the agent can do that because here's the way it works. The agent says well, how do I break that down? How do I think through the series of steps? Because think about it as an employee.
Speaker 1:If you hired a fresh Sidecar employee, they came on board, they had some skills, obviously, but they didn't know anything about Sidecar and you had to teach them how to do it. They'd have to break it down step by step and say, okay, first I need to get the transcript, then I need to think about which topics might be interesting, then I need to think about, okay, how do I create a blog for each of those topics? How do I then take those blogs and get review and approval? Because a brand new employee at Sidecar would not post a blog on Sidecar's website without getting your approval. Similarly, the agent can bring the human back into the loop and say hey, mallory, can you approve this blog that I created? These are the four blogs I created from this podcast episode. Can you approve them? And you can provide feedback or just approve them, and then the agent can do the rest of it.
Speaker 1:Taking it over to WordPress or actually I think you switched us to HubSpot. It tells you how plugged in I am to things, but we're using HubSpot CMS now, and so you take it to HubSpot, post it. So this requires the agent excuse me to not only have skills, but to also have access to tools. Tools such as requesting a review, tools such as posting to a website, tools such as being able to post to social media sites.
Speaker 1:So an agent, a multi-agentic system, can do all those things, and so it's basically a way of combining things into a complex workflow, and there would be significant business value from building an agent like that. It would save you probably dozens of hours a month from having to do that kind of thing yourself or delegating it to one of the team members. So that's, to me, a good example of a multi-step, complex process that you could actually bake into an agent system that doesn't require any advancements whatsoever in models. That just requires us to stitch together these models and their respective capabilities into a full system. That's really all agents are. It's just systems that combine models, prompts and tools.
Speaker 2:And diving into that experience even more selfishly, because I want to roll out something like that. You would say that's pretty feasible right now to create what you just said.
Speaker 1:I think. So I mean you know right now it might take a little bit more programming work than what would be desirable. I think you know you could 100% build something like that right now with code. You could use a coding framework, something like a lang chain. You could use member junctions framework. You could use a variety of tools. There's a number of agent tools Autogen is another one and you could use these tools and teach them how to do these things and then also write some code to kind of connect the parts that don't work directly within just the AI system skills.
Speaker 1:I think it would take, I don't know, maybe a couple days for a software developer to like stitch it all together. So it's not like a month of work or something for developer, but it still requires some developer skills. Very soon, through tools like Microsoft's Copilot, wave 2 and the Studio that's available from that, you'll be able to do almost all of this just as a user, because you'll be able to talk to an AI to say I want to build an agent that does these things, and that AI will be smart enough to create a lot of the pieces that aren't already there. So we're very, very close to the point where business users will be able to describe what it is that they want and then have the agentic system kind of come to life purely through that kind of interaction. So the software engineer won't be needed anymore for that type of work. So to me that's really exciting because that's going to empower every business user to just, you know, basically describe their business process and then the AI will be able to replicate it.
Speaker 2:Now, you described models as the brain here, and so is the reason when we talk. I was thinking about doing a little exercise on the podcast where I list out the names of these things and ask listeners to think is this a model or is this an agent? And then I realized, well, pretty much all the things we talk about are models. We don't really. There are no mainstream agents out there. I don't want to say that as far as, like we talk about in the podcast, right, there are no GPT-4 agents, unless you create them, to my knowledge. So I'm just curious why don't you think, given that we have the capabilities and that this is all possible, why don't we see kind of agent companies popping out of the woodworks at this point?
Speaker 1:Well, I think we actually are seeing that. I mean the majority of the AI funding happening in Silicon Valley right now. If you look at, for example, the most recent Y Combinator class has a tremendous number of agent companies.
Speaker 1:I don't know if it's the majority of the companies in the most recent YC class, but it's a large number of them that are solving very narrow problems in specific domains, creating they may not call them agents, but they're AI systems, right?
Speaker 1:So agent just is the word that we're using right now. But the reason I think agent is an appropriate term to use is because it kind of combines these autonomous, semi-autonomous and manual steps together into something that, in turn, can be built upon. So you could build the agent I described, which let's just call that the sidecar content publisher agent, right? So that agent does all the stuff that I just mentioned, right, 10, 15, 20 steps. And, by the way, the cool thing is you can go back to that agent at any time and say, well, I want to add six other steps, or these steps don't work very well, so I'm going to replace them with these steps, and it just automatically happens every single time. Thereafter there's no training required, right, because it just it's part of the system. But let's say that agent becomes a part of a broader system where you kind of build like with Lego blocks, and then you have another agent that uses that agent or talks, so the agents can talk to each other in a way that I think is pretty remarkable.
Speaker 2:That makes sense. And then the last topic today too, we're going to talk about this distinction of kind, of having an individual or a professional agent versus an enterprise agent. So I don't want to jump the gun there. But, amit, you also mentioned that agents can be multiple models working together, or even perhaps one model that has a GenTech functionality built on top of it. I'm thinking, claude, with computer use that we've covered in a previous episode. Does any of that really matter to the average individual, or is the key here being able to take action?
Speaker 1:The key is being able to take action and to kind of be able to invoke the agent, not just through your request on a per request basis, but for the agent to be kind of ready and available to interact with the environment. And what I mean by that is, let's say, an email comes in, being able to respond to that automatically, or someone coming on the website and performing an action on the website, which might in turn, also invoke an agent to do a number of things for you. So agents are kind of like these. You know, always on systems that can interact with the world, they can certainly be invoked by a user requesting them to do something. But you might actually have a scenario where, for example, when the edited video and audio for the pod comes back to us from our editor, that might be something you get through a Dropbox link or something like that. Well, when that Dropbox link has a new episode dropped into it, the agent could be monitoring that Dropbox folder and just automatically do this stuff without you having to ask for it. So I think there's things like that that make them like even not only more efficient but just more consistent.
Speaker 1:You know, we're people, we're full of all sorts of interesting unique flaws, and one of them is that we're not the most consistent machines on the planet. We're great at lots of other things, but you know, consistency and executing like a machine is not our strong suit typically. I know there's some people who are really good at that far stronger than I am at it but you know, and frankly I don't like that, I don't like having to do the same thing over and over again. Right, that's where I think agents can be truly remarkable in freeing us up to do things where we are stronger.
Speaker 2:Well, that's a good segue into the second part of this episode, which is talking about some business use cases for agents. So when people think of AI agents, a lot of people think about customer service. That's the first thing that comes to mind, and for good reason. Customer service represents a key function where agents can have an immediate, noticeable impact. And if we think about history right, traditionally businesses have set up customer service operations to prioritize efficiency and cost management sometimes maybe oftentimes at the expense of customer satisfaction. Cost management sometimes maybe oftentimes at the expense of customer satisfaction, and this has led to systems that we all know and love, like phone trees and scripted responses and long wait times. But agents can change this dynamic by providing a solution that benefits both the business and the customer. Agents are capable of handling complex, high-volume interactions quickly and accurately, and, of course, that ensures a better customer experience or, in your case, member experience, while still aligning with business goals. And we could talk about this all day right. But let's look at the facts, look at the data here.
Speaker 2:Klarna is a company that we have talked about on the pod before. It is a buy now, pay later, company. They rolled out a customer service agent. I believe it was sometime last year and the results have been astounding. It has resolved two thirds of all customer service chats. It's available in 35 languages. Also 24 seven. It's always on, like Amit was just talking about, it dropped resolution times from 11 minutes to under two minutes, decreased repeat inquiries by 25%. Their assistant or their agent is doing the work equivalent to 700 full-time humans. Customer satisfaction's gone way up. And maybe the cherry on top, klarna anticipates a $40 million profit improvement in 2024 alone.
Speaker 2:Another example that's relevant to this conversation is Bell Canada, which is Canada's largest communications company. They are utilizing Google's customer engagement suite, which combines Google Gemini with an omni-channel contact center. Bell Canada saved $20 million by offering instant customer answers and directing them to self-service options, and the virtual assistant that they have has handled over 1.1 million interactions across their brands. The obvious thing here, amit probably for our listeners and our viewers are that Klarna and Bell Canada pull in billions in revenue. So these are great use cases. We're seeing the impact, we're seeing the ROI, but do you think customer service AI agents are a reality for associations?
Speaker 1:I think so. I mean, the thing to think about there is you know, you said this earlier why has customer service technology existed? And the answer to that question isn't experiences for the customer. They're not optimizing for the customer experience, they're optimizing for low cost, whereas now we have technology that can do both at the same time. We can optimize for customer experience to be flawless, to be immediate, to be very pleasing because I don't know about you, mallory, but I think I'd rather spend two minutes than 11 minutes resolving a problem right, it's not super fun to be dealing with lots of back and forth. You know two minutes is great. So my point would be this that AI is so much smarter than all the other technologies we've used in this realm, that it is technologies we've used in this realm, that it is, I would say, mandatory for all associations to start finding ways to leverage AI for their service needs. People are going to expect this because they're going to get it from all the big companies. They're all rushing to this use case immediately because it's amazing. It solves both problems at the same time, improves customer experience dramatically and also lowers cost.
Speaker 1:I was talking to an association leader actually yesterday in the evening about this exact topic. He was saying that their association, which is a professional membership association, has a stated goal of a 24-hour customer service turnaround. So when they get an inquiry whether it's through their website or email or sometimes people leave them voicemails their stated goal is to respond to the customer or the member, in this case, within 24 hours. Seems pretty reasonable, right? Like especially in the world of human to human interaction. First of all, they don't really typically achieve that goal, that's their BHAG right Is to have that one day turnaround. But what if you could do it in 24 seconds instead of 24 hours, right? Customers would be a lot more pleased generally, especially if the answers are perhaps better than the average human.
Speaker 1:Because, think about it, like you know, our customer service isn't about being the deepest domain experts on demand. It's about being able to solve customer service inquiries, which is about, like payments and access to resources and guidance on which products and services and programs people might be a good fit for. It's an important role, but it's actually very limited. The domain expertise isn't assumed to be included in that, but if you could bundle that in there as well and people could really ask questions that are much deeper, like if I'm going to Klarna and saying, hey, I might be able to go to them and say, well, I had this deal. I bought this product on this website. I used Klarna, so it was buy now, pay later.
Speaker 1:It was four installments of $100 each over four months. And I'm talking to the agent about changing the payment schedule. That's a common BNPL customer service inquiry. It's like I can't pay you in four installments. Can I do it in six instead?
Speaker 1:Well, it would be cool if that agent also might be able to give me some advice about personal finance. That would help me do a better job budgeting on the home front, you know, to be able to solve that problem. Maybe that's not something the customer wants, but maybe the customer's asking for, like you. Maybe that's not something the customer wants, but maybe the customer is asking for, like you know, hey, do you have any suggestions on how I can improve this? And that's kind of a very natural vein for them to be in Right. In any event, I guess the point I would have is you know, a lot of people have talked about companies like Zappos that are known for outstanding customer service, or Singapore Airlines, or you know, pick your favorite customer service centric brand. What if we could all deliver value at that level? It would be amazing, and so that's really what I think is so compelling about this use case.
Speaker 2:You bring up a really interesting point for associations, which is the idea of the domain expert and then the idea of a customer service agent, for example, or member service agent. Do you feel like it would be a mistake for an association to go down the route of creating a member service agent that wasn't also a domain expert? Do you think those things need to be coupled together for the association success?
Speaker 1:So it's a great question. I think that if it's easier, for whatever reason, to just the customer service part first and to not do the domain expertise initially, that's fine. Nobody expects that right now, but wouldn't it be great if you could outperform the expectation and have that available at the same time? So it's not that you shouldn't roll out just customer service, but if you're able to do them at the same time, I think that offers kind of an all-in-one email address that people could email and all-in-one kind of web chat. This is something that I think could be truly an amazing resource for people to not have to worry about where to go. So I'm excited about that combined use case. But by all means, if people are going to phase this and do one piece maybe just the knowledge part first or maybe just the customer service part first that's totally fine. I don't think it's a mistake at all. To me, the mistake is doing nothing.
Speaker 2:Fair point At Digital Now 2024,. An interesting point was brought up a few times for me and that was the reminder that many associations have organizations as members, not individuals, and so if an association has, let's say, 200 organizational members, maybe customer service or member service is not as big of an issue for them. That's kind of a guess on my part. Would you agree with that?
Speaker 1:It can be, it depends. So sometimes the membership at the business level would mean that certain transactions are much lower volume, kind of going from your intuition a little bit deeper into that. The kind of the renewal cycle, for example, if your members are companies rather than people, would probably be simpler, easier in some ways. At the same time they're higher stakes, because if you only have 200 members and they each pay you quite a bit of money in order to be members, you need to make sure that the experience is really seamless. The other part of it is that the way people usually measure whether or not they're going to join to begin with, and then certainly whether they're going to stay with an organizational regime type membership, is by looking at the value accrued to the individuals that are part of their company. Are those people taking advantage of the member benefits that are available to them because of the organization's membership? And if the answer is not really, then the case for a renewal is much less compelling. So if you have things like a knowledge agent available on tap to be able to provide day-to-day, week-to-week support and assistance in a way of essentially value creation far beyond the expectations that currently exist, and you can show those who are actually responsible for the decision of do we renew or not, that they have hundreds or thousands of employees who have benefited from your service because it's so amazing. That is definitely a relevant thing. It's just it's not necessarily solving a problem in quotes, in that the problem isn't so much that people are asking tons of questions and it's taking a long time to get a response. That may be the problem, but often it's that they're not asking you anything. They join and then it's radio silence. You're like, hey, we've got all this great content, we're experts in this field, we've got all this cool stuff. We'd love to connect you with other members, but the phone's not ringing and that's a really clear indicator that something is amiss in terms of the value chain back to the end consumer.
Speaker 1:And so if you are an org-level membership type of model, you have to really think about that and say how do I reach deeper into the organization's individuals? Right? Because ultimately there is no such thing as a company. A company is, it's basically a work of fiction. You know, there is no such thing as a company, there is no such thing as a government, there's no such thing as a country. The only thing that exists is physics and biology derived from that. So we, as humans, have made up all of these things in our minds, which, of course, is how our whole world is constructed, and it's awesome. But ultimately the decisions are made by people.
Speaker 1:And so if there is a groundswell of demand for your service, because it's so fantastic that the individuals within the membership of that organization that's paying are so used to having that service and it's so woven into the fabric of their day-to-day work, it's very hard to remove you. Your switching cost, so to speak, is extremely high, and so your association is going to benefit a lot. So I think customer service is super relevant in those organizations, just not necessarily in the way that people might be thinking, because when people think of customer service, the first image that comes to your mind is an irate customer on the phone screaming at you about some complaint. That is what customer service means to most people and, yes, that exists and that's important to handle professionally and quickly. There's all these other dimensions of service that are not necessarily in that classical customer service vein that are super important. It's the unanswered calls because the calls that were never made, because people who don't want to subject themselves to your phone. Tree of hell.
Speaker 2:Wow, yeah, that's a great point, amit. I recently joined this networking group in Atlanta because many of our listeners and viewers know I moved to Atlanta a few months ago from New Orleans and I didn't go to any of the activities for a few months. I didn't really check out their community. I kind of just signed up and then let it fall to the wayside and I had this moment where I realized, oh my gosh, I'm the unengaged member, it's me. And that was just really insightful to kind of think about. You're right, maybe it's not always the phone call, the irate customer, maybe it's all these things that you offer that are not being transmitted to your members. So I thought that was an interesting kind of meta moment I had in my life.
Speaker 1:Makes a lot of sense. Yeah, people join for all sorts of different reasons. Sometimes, like in your case, you moved to a new city, you wanted to connect with other professionals who might have similar interests, and you know, but nothing happened. You signed up and then it was incumbent upon you to engage. Right, you were the actor required to take a step forward.
Speaker 1:But in the context of agents, by the way, the other thing these things can do is actively reach out to people, right?
Speaker 1:So, rather than just sitting there and waiting to respond, you could set up an agentic system that is looking for people like Mallory, who is not being a good member, right, and Mallory hasn't done anything with this association in Atlanta, and this agent could say hey, mallory, listen, you're a new member and when you joined, you told us you had just moved into town. There's a couple people here that might be interesting for you to connect with, that are near you. So let us know if you'd like an introduction and you could click a button in the email that says yes, and they say hey, mallory, I'm introducing you to so-and-so and this is, you know, a connection. You know, that kind of thing is the center of what many associations do, but they only do it upon request, and so that, I think, is another opportunity for agents is to always be aware of these kinds of situations, to look for these patterns of lower engagement or complete disengagement and do something about it.
Speaker 2:Yep, and for the record, since I like to be an A-plus student, I did go to an event recently, but I'm thinking if I'd gotten a text to my phone that said exactly that like, hey, mallory, click this link, click the link and it'll RSVP you for this event next week, I would have been more likely to take action.
Speaker 1:So yeah, associations. Another thing that I would share is a lot of associations are thinking constantly about how to not be annoying, and that's like me paraphrasing. It's not like what I hear exactly from associations, but what I mean by that is, you know, associations are saying you know, oh, you can't possibly want me to send another text or another email. My members are already telling me I send them too much stuff. So the problem is not so much that you're sending them too much. Usually, it's that you're sending them stuff they're not interested in.
Speaker 1:And so therein lies another opportunity in the world of AI, which is, you know, understanding the relevancy of content using personalization, which, of course, combines with this conversation. Because if that agent reached out to you every day and said, hey, mallory, come to this event, mallory, come to this event, you would very quickly report them as junk and not respond. You'd block the number, right. But if it was intentional and quality and relevant and was done at a good cadence, you'd probably find it really valuable, especially if he had joined but you hadn't done anything 30 days later, right, something that's a little bit more thoughtful, and we have in our hands the technology to do that, and the underlying technology is basically free. It costs money to put this stuff together and create solutions, but the underlying technology, these models, are fully democratized. Like very powerful models, these smaller models are available at basically zero inference cost.
Speaker 2:So at this point, hopefully in the episode, you are sold on the potential of AI agents, but implementing them raises some important questions. First, as I mentioned at the top of the episode, we need to consider the distinction between professional or individual agents and enterprise agents. Professional agents are often designed to assist individual users or even departments, providing support with tasks like scheduling or document generation or maybe data insights specific to a user's workflow or, in the example at the top of the episode, maybe having an agent that helps Sidecar generate blogs from our podcast episode. In contrast, enterprise agents, of course, are built for larger organization-wide functions, handling more complex workflows, managing large amounts of data and engaging with your members at scale. These agents are, of course, more challenging to implement, but deliver impact across multiple departments.
Speaker 2:So where does that leave an organization or an individual user looking to get started? Is it feasible to build these agents in-house, or does an enterprise level solution require external vendors and specialized platforms? And then, what about entry-level tools like Microsoft Copilot? Can they meet the needs of individual users or is a more robust solution necessary for enterprise-wide impact? So my first question for you, amit, which you kind of touched on, is, in reality, who, within an association team could build an enterprise agent. At the moment, this is a theoretical association team. Are we talking the most technical people only, or should we be more flexible with that?
Speaker 1:So I think the folks that are decidedly non-technical need to lead in terms of what the business cases are. They're the ones who need to lead in terms of what the business cases are. They're the ones who need to say, hey, this is where we're stuck, where we have a massive backlog where an agent can help, or this is where we aren't stuck because the phone's not ringing right. Those are as big of a problem as when you're overloaded. I think the business people are best situated to identify the opportunity areas. But, to answer your question, the technical folks, either internally or through vendor partnerships, are really important at this stage to help you, first of all, architect these things well. You don't want to cobble them together in a way that's going to break over time. You want to be very thoughtful about data privacy and security. You want to be thoughtful about scalability so that what you build can increase in scale over time, not just in terms of volume but in terms of complexity, where, very quickly, you're going to have an agent, another agent, many agents. You have teams of agents and you want to be able to have traceability or an audit trail where you know what these things are doing. You want to be able to manage them. So there's some degree of an enterprise management type of process around it. I wouldn't put a ton of heavy weight around it because that's going to slow you down, but I do think that there is a role either for a really strong internal IT team if you have one or for a vendor partner. Or even if you have a really strong IT team, they're probably busy, so a vendor can definitely support you. Busy, so a vendor can definitely support you, and it's just important to have someone that's really familiar with doing these things, because AI agents and AI in general, of course is a fairly new discipline. There aren't a lot of people out there who are really deeply expert in it, so I would say getting some help, at least to get you started, might be a good idea. There are tools that are out there, like you mentioned and we talked about in this episode and priors, like Microsoft's Copilot Studio. The Google product you mentioned earlier that Bell Canada took advantage of, is available too. There's a number of them out there. It's just going to become easier and easier and easier.
Speaker 1:So I'm optimistic that this is going to be fully democratized when I say that the cost of the technology the underlying technology is near zero in cost. I'm talking about the models of the technology. The underlying technology is near zero in cost. I'm talking about the models. So access to LAMA 3.1, the middle-sized model, is an incredibly powerful model. It's more powerful than the original GPT-3.5. It's one of the points that I made in my keynote at Digital Now last week. It's a dramatically smaller model but it is actually smarter than the GPT-3.5 that powered the chat GPT moment, so to speak, back in late 22. So we have this technology available and it's amazing, but we have to still kind of architect these solutions, assemble them, and that takes some technical know-how Over time. I think that will change.
Speaker 1:The one thing I'd throw into the mix that I think would be a cool area for folks to potentially play around with is with tools like Zapier.
Speaker 1:So if you just think about what Zapier does for a minute I know you use this tool as well, mallory, but a lot of people, particularly a lot of marketing folks, have learned how to use Zapier where it's kind of democratized access to APIs, where even it will say hey, when someone registers for a course in my LMS, let's put them into my CRM, my HubSpot right, you did that yourself when we implemented our new LMS for the Sidecar AI Learning Hub, and in the past not too distant past actually that would have required a programmer to write code custom code to do that and the world of Zapier has democratized kind of interconnectivity between APIs.
Speaker 1:Well, zapier has done a lot of work around AI where there's, you know, prompting the ability to build Zapier solutions using AI, and I think they're an amazing company. They've done a lot of leading work in this area for what I'd call consumer grade systems integration, and so I would say, check out what they're doing. There's another tool. I haven't used this in years so I'm not even sure if it's still out there, but it was IFTTT. If this, then that. Have you ever heard of that one?
Speaker 2:No, I have not.
Speaker 1:It's kind of like an even simpler, more consumer grade version of Zapier. It's what it sounds like. So if this happens, then do that right. So it's like if I get an email, then delete it. That's my favorite if, to just get rid of all my email. But you know, so it's basically a form of automation, but these tools are all becoming AI enabled where you can kind of create agent like behaviors in them, and so there's a lot of cool things you can do.
Speaker 1:The other thing I'd mentioned before I wrap these kind of opening comments and how you can get started is pay attention to what Claude is doing in the world of Anthropic. You know that's basically what computer use is. This new capability that's in public beta. With last week's announcement of the newer version of Claude, 3.5 Sonnet, which is still 3.5, by the way, just like it was a few months ago they might have wanted to call it 3.6 or something to make it a little bit easier to understand, but anyway, it's 3.5 sonnet new, and this one has a public beta feature, as we were talking about last week, which is all about computer use, and what's remarkable about this is you can provide a prompt to the ai to say, hey, you take over my computer and do all these different things.
Speaker 1:So let's say, for example, you have a really old AMS that does not have an API, does not have a web interface, doesn't have anything, but it's usable as a computer application on your computer, on your Windows desktop. Well, you can run Claude's 3.5 Sonnet computer use beta and have it automate things in your archaic, ancient AMS, even though that thing doesn't support any modern technology. Once Cloud is able to do that, then you can actually wrap all of that into an agent, and right now you do this through just prompting. You'll then be able to like wrap all that into a piece of an agent where you actually have an agent sitting on top of your ancient AMS or whatever ancient software you've got. A lot of people have some really old software out there and then you can kind of connect it with other things, right. So there's some really amazing stuff out there that you can do as a consumer, even if all you're doing is kind of like testing things out and preparing your mind for a future where you can create these kinds of automations.
Speaker 2:Brief detour for a sec because you bring up an interesting point. I pulled up this IFTTT. If this, then that company, it still exists and it does exactly that it. Basically we have triggers set up and then it can take action, like set a calendar event when I received this type of email. Now, that's automation. Can you explain how an AI agent is different from that?
Speaker 1:explain how an AI agent is different from that. Yes, the AI agent basically puts the thinking in between those steps and wraps the steps. So the automation is like hard rule-based. So in Zapier you say it's the same thing, right? Zapier says, hey, when the AI learning hub has a new student, that event triggers the creation of a HubSpot CRM record or the updating of a HubSpot CRM record. Right, it's basically an if this, then that scenario. But you've put in hard rules around it.
Speaker 1:So what you could do with an agent is say something that's a little bit broader in nature, that says you know, what should we do with this new customer? Let's think about this customer and how can we best engage this new student. Okay, this new student named Mallory signed up for the AI Learning Hub. Let's get some more data on Mallory and let's think about what it is that we could do to best serve Mallory. And the agent might have several possible tools.
Speaker 1:So it goes from being deterministic, where there's a pre-coded set of steps that can execute, which could be one step, like the example I just gave, or it could be 500 steps with branching logic.
Speaker 1:That's called a deterministic code path or a deterministic set of steps.
Speaker 1:Essentially, it's just like a predetermined set of steps and AI is non-deterministic. Ai is like you and I in the sense that it's kind of coming up with the best steps to do based upon a variety of factors. That's the way these neural nets work and ultimately that creates more choice and more opportunity where that agent could kind of you know intermediate and say, oh well, this new customer has come in. What's the best piece of content I can send them and when should I best follow up with Mallory and all these other things that you can do that are way more in the way of like thinking and reasoning, and then have access to all the tools in Zapier where you can say, oh well, I have in Zapier, I have my CRM, I have my LMS, I might also have my marketing tool, my email tools, my Google Sheets, my Microsoft stuff, and just it basically means that the agent can be set up to use all these tools. That's one of the reasons you see people like Zapier building agentic type features into their platform.
Speaker 2:That makes a ton of sense. I've been questioned on that distinction before myself, so I am glad that we got that on the podcast. To go back to the question on who can build agents within an association team, if we're talking about customer service, member service agent or perhaps a knowledge agent, like we discussed I'm going to keep putting you on the spot here. Do you have like a number in your mind, amit, of like you would need a team of this many technical people to build and run an agent at your association?
Speaker 1:Yeah, I don't know if there's a specific number, simply because it depends on the scope of what someone's trying to do. If they want to do something very simple, there are these consumer-grade tools. They can probably figure out how to set up on their own. It's still early days in the sense of like it's not a click click, click done kind of scenario, Not yet. It will be probably by digital, now, 2025, that there will be consumer, you know, consumer grade agent tools that will just blow your mind in terms of what they can do and connect to enterprise systems.
Speaker 1:Right now, I think to go into anything a little more complex, either we need to have, you know, some level of internal technology staffing or a vendor partner that can work with you on it. So, but it does not need to be a big lift. It's not like implementing an AMS or something. It's not a gargantuan project, it is. You can do really small projects, Like even if you're doing something at the enterprise scale 30, 45, 60 days, small investment of dollars and time and you can get a basic agent up and running for your first use case. And that's the most important thing is to do your first one, you know. So I'll give you.
Speaker 1:I'll give you one example is that you know no-transcript of any kind on top of that platform, and so the member junction team is at work, working with a bunch of different associations right now, setting up agents for them and these projects. They typically start off as proof of concepts. They're typically about 30 to 45 days to get started. They're pretty low cost, so it's not a big big lift. It's not something accessible yet to every association on the planet with a series of a handful of clicks, but that's where this stuff is going, which is what I'm most excited about. But for organizations wanting to get going now, the barriers are not that high.
Speaker 2:And MemberJunction that you just mentioned, and I think you said this is open source, so association listeners can go and figure this out on their own, potentially right now.
Speaker 1:Yeah, it is an open source platform. The whole point of that was, you know, we anticipated several years ago that there was a need for a neutral ground, essentially a data platform that the association completely owned, because, remember, whether it's models or agents we're talking about that data is the fuel for doing anything with AI. You have to have your data, and data tends to be disparate across a variety of sources. Right, you have an AMS, an LMS, a CMS, a financial system. You have your data, maybe in a bunch of spreadsheets too, so your typical organization has data scattered about, and traditional data warehouses tend to be very inflexible. Even if you have one, they tend to be very highly structured, they're very difficult to change, very expensive to change, and you need something that's fundamentally more flexible than that, and so the whole idea of an AI data platform, sometimes called a common data platform or CDP it's the same concept. It's this idea of having a read-only repository of all of your data coming together in an environment that's designed to be constantly evolving and AI native, meaning that the AI is actually what's managing all of that data and making it make sense.
Speaker 1:So that's what we built into the Member Junction platform. It's our gift to the association community. It's 100% free software. The Member Junction team does have professional services they offer on top of that, but any vendor partner can work with Member Junction. So if you have an existing relationship with a trusted partner of any variety, they can take Member Junction and use it. There's a whole bunch of documentation available for it. The whole goal is to make it a universally accessible tool that everyone can use in the community around the world.
Speaker 2:So when we're talking about enterprise agents, almost an essential first step right is getting some sort of common data platform in place. You could do that through Member Junction with another vendor you're working for or through MemberJunction's professional services, and then, and only then, could you roll out an enterprise AI agent. Does that sound right?
Speaker 1:That's the way I would approach it, because it's kind of like, you know, a step-by-step approach where the foundation is the data. You don't need to do it that way. You could say, hey, I'm just going to build an agent that doesn't have a data platform and just directly talks to APIs, directly talks to my AMS API, so you can do it without a data platform. I really recommend the data platform as the first step because, first of all, it's pretty easy to put in place. Again, it's not an AMS level implementation, it's pretty simple. But the nice thing about building on top of that foundational layer is then you're building on bedrock, you're not building on sand, and when you're building on, like you know, a mixture of a variety of APIs that are constantly changing, it's really hard for those agents to be reliable.
Speaker 2:Agent Builder, which I played around with just a bit this morning before the podcast. It's available now. I know we've mentioned it on a previous episode recently. Available now, but the agents are not able to take action just yet. That is something that they're going to roll out a little bit later. Realistically, amit, do you think this Microsoft Agent Builder will only be for kind of those individual agents that assist a person with their workflow departmental agents as opposed to an enterprise agent, or do you see that being the path with Agent Builder?
Speaker 1:So I'm a big fan of what Microsoft is doing in general in this space and I'm bullish on where they're heading in general with their AI strategy. I think the co-pilot agent tool will eventually be far more robust than what you experienced. You're right right now. It's really basically like building a custom GPT. That's essentially what it is. The nice thing is is it's part of your team's environment, so you build an agent and then it's just available within Microsoft Teams. It's like kind of like having a custom GPT that lives in Microsoft Teams, which is awesome. It's super, super useful To your point.
Speaker 1:It doesn't have tool use. It doesn't have the ability to take action. You can create agents in the world of Microsoft Copilot that are built in code rather than through the agent builder. So if you have a software development capability or, again, a strong relationship with a vendor partner, you can build Microsoft Copilot agents in code that do anything, because once you are in code, you can call any other model you want to. You can do deterministic things in the software. You can call tools. But that's not quite there yet and, like the sizzle in the video marketing that's out there for this stuff is just unreal. How awesome. It is right. It's amazing. It's not quite there yet in terms of click a few buttons and you're there, but I think it will be probably again by digital. Now 2025, we'll do our predictions episode for 2025 in December, I'm sure, but that's gonna be one of mine is that there will be consumer grade agent tools that will allow end users non-technical end users to build very sophisticated agents as part of their enterprise.
Speaker 2:I wanna say we definitely had an agent prediction or a few. I wanna say the prediction was agents will become mainstream. Do you feel like that is the case?
Speaker 1:world. There's probably a case to be made that they're becoming mainstream, but associations tend to. On the one hand, they do tend to be a little bit behind on technology generally compared to the broader market. But some of that's actually perception more than reality, because people look at the leading companies and say, yeah, associations are so far behind, like Nike, in terms of e-commerce or Apple or Amazon, it's like, yeah, of course you are.
Speaker 1:Those are like some of the leading brands in the world and, by the way, if you look behind the screen for those companies, they have a lot of hamsters on wheels too. So I guess my point is is don't feel that bad, just take some action, because in general, there's a lot of businesses out there commercial businesses that are run just as archaically or worse than some of the slowest moving associations. That's not a suggestion that you should be like yeah, ok, we're cool, we're good, we're chill, let's just kind of hang out. But rather that you know associations shouldn't spend all their time like thinking about you know, poor, poor me.
Speaker 2:We're an association, we don't have tons of resources, we're behind everyone, because you kind of are in some ways, but you're kind of not, and the most important thing is to not worry about any of that stuff and just look ahead and start taking some action, and AI is leveling that playing field for sure. My last question on this topic, amit, is kind of a fun imagination question, I guess, but let's envision this world where an association has multiple agents that serve really important business functions and then, at the staff level, every individual maybe has their own few agents the Mallory agent that handles all my emails and scheduling, and then maybe we've got a content agent and a marketing agent, so on and so of oversee AI agents in an association. Or, dare I say, do you think an agent is going to oversee all the agents in an association?
Speaker 1:There was something recently I read about mid-level managerial work and a lot of like. It's essentially like a decision routing engine is what a lot of mid-level management is. You know you get input, you make a decision and you share an output. That's, you know, obviously a dramatic oversimplification, at least in some companies. But I do think that a lot of that kind of work will be agentic decision-making, certainly the low stakes stuff with human in the loop for higher stakes kinds of things. I think it's going to free up a lot of time. I think it's going to result in a significant amount of job loss across the broader economy, which is obviously an issue, but I think it's also going to create an incredible amount of additional economic output, which hopefully will result in new opportunities, new value creation, new businesses and new employment opportunities as well. I don't really want to guess too far ahead in terms of how that's going to shake out, but if you just think about what a lot of these people do in mid-level management jobs, you know it's absolutely going to be replaced by AI, because the agents are consistent. In some cases they're actually a lot more thoughtful in terms of really evaluating the data, because, you know, think about it. If you have 50 decisions to make in a day, if that's your job and you're overwhelmed and you're tired and maybe you've got stuff going on in life, you know you're maybe going to like work through those decisions a little bit faster and not go as deep on thinking about each one individually as much as you might want to, and you might not have perfect recall of every decision you've ever made in the past to learn from them other than kind of the general heuristics of decision-making we draw upon. You know from our past experiences.
Speaker 1:So the idea essentially is is that you know we are going to see these things everywhere and we should start with the lower level primitives, which is the capabilities for particular functions, and then they're going to start assembling in ways where you have these different agents, as you said, and you're going to assemble them into teams and then teams of teams and ultimately that's going to result in entire organizations that potentially could have a very, very high degree of automation.
Speaker 1:I think the human in the loop part is going to be important for higher stakes decision making. I think that the human control over these things for creativity is going to be important. Control over these things for creativity is going to be important. There's a lot to be discussed there around the philosophy of it and all the other impacts of this, but that's how these things are going to go. Is they're going to start assembling because you have the fundamental capabilities to do this today, even if AI advanced 0% in terms of the model's fundamental capabilities, from where we are today for the next 36 months. In the next 36 months, we will see radical advancements in what agents can do, even with zero improvement in models.
Speaker 2:Well, everyone, there you have it. That is our AI Agents Deep Dive for Associations. Thank you for tuning into this episode and we will see you all next week.
Speaker 1:Thanks for tuning into Sidecar Sync this week. Looking to dive deeper? Download your free copy of our new book Ascend Unlocking the Power of AI for Associations at ascendbookorg. It's packed with insights to power your association's journey with AI. And remember, Sidecar is here with more resources, from webinars to boot camps, to help you stay ahead in the association world. We'll catch you in the next episode. Until then, keep learning, keep growing and keep disrupting.