Fundamentals of AI, Part 2 | [Sidecar Sync Episode 151]
In Part 2 of their Fundamentals of AI series, Amith and Mallory move from definitions and building blocks into what abundant intelligence means in...
37 min read
Mallory Mejias
:
Published
Summary:
Some of the clearest AI lessons for associations are coming from drive-throughs, bank call centers, and oncology. Amith Nagarajan and Mallory Mejias step outside the big AI labs to look at what other industries are actually doing with AI, and what association leaders can borrow from them. They revisit AI tacos with Taco Bell's voice AI, now taking orders in more than 900 drive-throughs, and weigh Chick-fil-A's stance that hospitality should stay human. They break down Bank of America's published AI returns and its CTO's advice to reach for deterministic software first, then explain where decision models sit between classical code and frontier LLMs. They also unpack CureWise, whose virtual tumor board simulates a panel of cancer specialists for patients. Along the way, Amith makes the case that lowering friction to trusted knowledge, through tools like Betty and MCP, is how associations turn member service into deeper engagement.
Timestamps:
00:00 - AI Tacos, Round Two
05:48 - Taco Bell's Voice AI Drive-Through
11:22 - Is Voice AI Ready Yet?
16:31 - Should Hospitality Stay Human?
20:58 - Bank of America's AI Returns
26:55 - Where Decision Models Fit
29:59 - Proving AI ROI to Your Board
37:20 - CureWise's Virtual Tumor Board
41:00 - Closing the Cancer Guidelines Gap
43:14 - Getting Members to Use Betty
48:08 - The Always-On Virtual Member
📰 Yum Brands Flags AI as ‘Transformative to the Business’ ➔ https://bit.ly/4ffkZeC
📰 Smaller AI Models Help Bank of America, Capital One Manage Costs ➔ https://bit.ly/4t6oY2K
🎧 Sidecar Sync Ep. 72: 'AI Tacos & Amazon’s Return to Office Mandate' ➔ https://apple.co/4jqEfHB
👥Provide comprehensive AI education for your team
https://learn.sidecar.ai/teams
🤖 Join the AI Mastermind:
https://sidecar.ai/association-ai-mastermind
🎀 Use code AIPOD50 for $50 off your Association AI Professional (AAiP) certification
📕 Download ‘Ascend 3rd Edition: Unlocking the Power of AI for Associations’ for FREE
📅 Join the pre-registration list for digitalNow 2027:
🛠 AI Tools and Resources Mentioned in This Episode:
Byte by Yum! ➔ https://bit.ly/4Ayv2no
Erica ➔ https://bit.ly/4zvyE8Q
CureWise ➔ https://curewise.com
Betty ➔ https://meetbetty.ai
Model Context Protocol ➔ https://modelcontextprotocol.io
Claude ➔ https://claude.ai
Claude Fable 5.2 ➔ https://www.anthropic.com/claude/fable
ChatGPT ➔ https://chatgpt.com
Gemini ➔ https://gemini.google.com
https://www.linkedin.com/company/sidecar-global
https://twitter.com/sidecarglobal
https://www.youtube.com/@SidecarSync
⚙️ Other Resources from Sidecar:
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 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
🤖 Please note this transcript was generated using (you guessed it) AI, so please excuse any errors 🤖
[00:00:00:14 - 00:00:09:17]
Amith
Welcome to the Sidecar Sync Podcast, your home for all things innovation, artificial intelligence and associations.
[00:00:09:17 - 00:00:25:14]
Amith
My name is Amith Nagarajan.
[00:00:25:14 - 00:00:27:19]
Mallory
And my name is Mallory Mejias.
[00:00:27:19 - 00:00:33:14]
Amith
And we are your hosts. And Mallory, have you eaten any good tacos lately?
[00:00:33:14 - 00:00:56:17]
Mallory
Oh, have I eaten any good tacos? You know, actually my husband and I joke that we eat Mexican food way too much. That is just the food I gravitate toward. So I like to eat the same thing for lunch. I go through phases. It just makes things easy. So I've been having breakfast tacos with eggs and tortillas for lunch for the past few weeks. So those have been pretty good. What about you, Amit? Have you been to old Taco Bell?
[00:00:56:17 - 00:01:20:01]
Amith
You know, I have not. But I've been to some other taco shops, you know, here in New Orleans. We have some options for Mexican food. I grew up in California and California Mexican food is pretty amazing. So when I go visit my son out there, he's in college out in Southern California nowadays, it's great to hit the 24 hour taco shops that are awesome.
[00:01:21:12 - 00:01:32:02]
Amith
But here in New Orleans, I tend not to gravitate towards Mexican food because I just haven't found too many places that I really, really enjoy. There's a couple that are pretty good. But whenever I think of tacos, Mallory, I think of artificial intelligence, don't you?
[00:01:32:02 - 00:01:37:24]
Mallory
Is that so? You know, I don't always think of artificial intelligence, but please, please do tell us why, Amit.
[00:01:37:24 - 00:02:01:00]
Amith
Well, way back in the Wayback Machine, which is probably episode 72, I think. OK, I was going to say under 100. It was a long time ago in AI timescales. That's like the Neolithic era. So long ago in the history of the sidecar sync, we had an AI tacos episode. And so for some reason, things stick in my head sometimes.
[00:02:02:05 - 00:02:27:04]
Amith
And so every time I think of tacos, I think of AI now because of that episode, Mallory. I haven't been able to get it out of my head. We're now on episode, I think 161 is this one. So it's been a minute. 90 episodes or so. That's like a year and a half, maybe a little bit longer. But yeah, tacos and AI. They're one of the same to me. And the same minor on that work is all squished up. And that's how I think of tacos now.
[00:02:27:04 - 00:03:04:21]
Mallory
Oh, man, all jumbled. If you have been with us that long, we appreciate you. In the first AI tacos episode, I'll give you all a second to figure out which of us do you think suggested that episode. It wasn't me. It was Amit. Amit often comes to me with like these fun ideas. We had like a football episode. AI tacos was certainly a standout in the history of this podcast. But on that episode, if you recall, I asked Amit if he had ever had a Crunchwrap Supreme from Taco Bell, which I feel like is a universal phrase. I thought everybody knew what a Crunchwrap Supreme was, but he had not heard of it and never had one. And I think, Amit, you still have not, right?
[00:03:04:21 - 00:03:25:10]
Amith
I am still deficient. However, this time around, I have since been duly educated by you on what a Crunchwrap Supreme is. Right. But I have not yet gone to Taco Bell since our last episode on Taco Bell. So maybe I'll have to have to go do that. It'll bring back memories of high school for sure. But I've been to Taco Bell since high school, but just not in a while.
[00:03:25:10 - 00:03:34:24]
Mallory
I have also not been to Taco Bell. Definitely. I don't do a ton of fast food, but now I'm tempted. Sometimes I'll make a Crunchwrap Supreme at home if I'm feeling really inspired.
[00:03:34:24 - 00:03:49:06]
Amith
But just remind our listeners for those that may have been or maybe as, you know, unfortunately, not enlightened as I was once upon a time, you can enlighten our listeners as to what is a Crunchwrap Supreme exactly.
[00:03:49:06 - 00:04:46:19]
Mallory
OK, I will do my best to do it justice with words because it's really something you have to experience. Also, I still think everybody's going to know what this is, but maybe I'm wrong. So it is a soft tortilla on the outside and then inside. You can do whatever meat you like. I think typically they do ground beef and beans. And then on top of that layer is a crunchy taco shell, basically like a tostada in a round format. And then on top of that, you can do lettuce and sour cream and tomatoes, onions, kind of whatever toppings you would typically put on a taco. And then they wrap the whole thing in that soft tortilla and what do you call it? I guess press it in a pan to where you get a little bit of a crust on the outside. So it's kind of like a burrito, but it's really different because you get the soft, you get the crunchy, all the toppings. Maybe they put queso on it, too. It's just it's really good, everybody. And honestly, it may not be directly related to A.I., but I guess as we're going to talk about in this episode, maybe it is.
[00:04:46:19 - 00:04:49:11]
Amith
I think there's like a Dorito flavor version of it now.
[00:04:49:11 - 00:04:55:17]
Mallory
Ooh, that would be good. That feels sinful, though, I mean, I don't know if I can order that in my adult life.
[00:04:55:17 - 00:05:02:03]
Amith
Yeah, well, you know, you can still go buy cereal at the kids cereal aisle and have and enjoy that. But
[00:05:03:14 - 00:05:47:00]
Amith
I promise this is relevant to the world of A.I. and associations very indirectly, perhaps. But, you know, I'm kind of hungry now. So but the world outside of associations and one of the reasons I bring these seemingly silly topics to Mallory for consideration for the pod is because I think it's really important that we cast a wide net as a community and try to learn as much as we can from not only our our own community members within the association world, but from people in all walks of life and all sorts of industries. And so I'm always looking to learn from other sectors and we try to bring some interesting ideas from other verticals, other industries into our own world of associations, which is really what this episode is about, isn't it?
[00:05:48:08 - 00:07:48:05]
Mallory
Exactly. And to continue that thread, I feel like a lot of our recent episodes have been centered on some of the big A.I. companies, Google, OpenAI, Ananthropic. And those are the companies, of course, that are making this progress possible, but they're not the only ones doing impressive work with A.I. like Amith just said. So in today's episode, we're zooming out to look at what some other industries are actually doing with it. We are starting with fast food, revisiting that A.I. taco topic with the voice agent they have now taking orders at the Taco Bell drive through. Then we're going to look at banking, where Bank of America and Capital One are publishing real numbers on what A.I. costs them and what it returns. And we close with CureWise, an A.I. cancer navigation platform built by a founder who has cancer himself. So we're starting off with A.I. Tacos round two. Bite by Yum is the technology platform Yum Brands built in-house to run its restaurants, which include KFC, Taco Bell and Habit Burger. One system handles ordering point of sale, the kitchen screens, inventory and scheduling, and more than 25000 Yum locations use at least part of it. This is what we covered back in episode 72, the Crunchwrap Supreme episode. At that point, the A.I. was working behind the counter on things like recommendations and forecasting, but it has since moved to the speaker box. Yum says Voice A.I. is now live in more than 900 Taco Bell drive throughs with a target of more than fifteen by the end of this year. Taco Bell has roughly 7,700 U.S. locations, so that is close to one in five. On the Q3 earnings call on November 3rd, CFO Ronjith Roy said that stores running voice A.I. saw a 6 percent improvement in overall drive through satisfaction, roughly a 1.5 percent sales lift driven by larger orders and better team member retention, all measured against a control group of stores without it. Amit, are you surprised by these Taco Bell results with Voice A.I. in the drive through?
[00:07:48:05 - 00:09:38:17]
Amith
Not at all. I think that it's first of all, I'm definitely going to have to go and find a Taco Bell to order something and check that out. But, you know, we've been really excited about Voice A.I. and more broadly just real time A.I. for quite a while. We've been experimenting with it and building apps and doing things for a while, you know, well over a year now and the technology just keeps getting better and better. And when people first experience what current voice A.I. can do, they're really surprised because I think Mallory, a lot of people kind of when they think of voice A.I., they think of Siri or maybe Alexa, which at the time, you know, 10 plus years ago when those were introduced, they were groundbreaking and amazing what you could do, but now are pretty limited in comparison. The new Siri is a bit of a step up from the old Siri, but still not nearly at the level of what Claude or chat GPT can do and certainly what you can do yourself as an association with voice A.I. So I'm not surprised in a couple of thoughts. First, when people know they're talking to an A.I., I think they might feel a little bit less shame in ordering exactly what they want. So if you wanted that milkshake, you know, you maybe don't want to ask for it from a human that you don't really care about talking to a robot. Maybe that's part of it. I don't know. I'm not a psychology person, so I don't know if that's the thing that I probably would do that. So I'm you know, my sample size of one perhaps might order more that way. The other thing I think is perhaps people will feel like they get a more consistent experience because part of the challenge is, you know, all of us get tired. People have different levels of training, different language skills. And so, you know, if it's hard to interact with the human you're talking to because of a difference in accent or something like that, that can make for a higher friction, less pleasant experience for both the customer and for the employee. So I'm not surprised at all as my short version.
[00:09:39:20 - 00:09:52:18]
Mallory
I know we rolled out Grace, our audio agent on the Sidecar website. We actually had Grace join us on the podcast before. That's a very interesting episode if you want to check that out. Amit, do you have any updates on what Grace has been doing lately?
[00:09:52:18 - 00:11:20:24]
Amith
Yeah, you know, Grace has been hanging out on Sidecar.ai if you want to go talk to Grace. She's available to chat with you. Grace is awesome. Grace can answer any questions you have about Sidecar, can walk you through like what the Sidecar learning hub is about, what our that's our core education platform and offering here at Sidecar where we do our association AI professionals. So Grace can help you even sign up because Grace has interactive tools. So Grace has continued to do what she started off doing. We launched Grace back in I think it was March or April. It was earlier this year and she's had thousands of sessions. So it's been a hit, I think. People who've actually interacted with Grace have given us really positive feedback. And we keep making improvements. We also rolled out a voice AI called SID inside the actual learning platform, the Sidecar learning platform itself when we relaunched it back in September. So it's been over two months since we launched LXP, which is our learning experience platform and SID specifically, which is a far more sophisticated AI audio agent because SID is actually designed to be a tutor. So SID can actually help you learn, pull up whiteboards, illustrate concepts for you, create custom lessons and learning paths on the fly for you based on the thousands of hours of content that we have and learning up. So it's pretty cool. So yes, Grace is awesome and we've continued to evolve and extend the technology from there.
[00:11:22:07 - 00:12:33:21]
Mallory
I want to take a moment and look at how Yum got here. In August 2025, Taco Bell's Chief Digital and Technology Officer Dane Matthews told the Wall Journal that the company was rethinking voice AI after a run of viral failure videos, including a prank order for 18,000 water cups. At that point, the system was in more than 500 drive-throughs and had taken over two million orders. The company said it would coach teams on when to use voice AI and when a human should step in. Rather than pulling it, Yum nearly doubled the footprint and kept a person watching. For context on whether Yum is ahead, McDonald's ended its IBM voice AI test in 2024 and is now piloting a Google-built agent called Archie in five US restaurants. Chick-fil-A, interestingly, has reportedly stayed out of AI order taking entirely on the grounds that hospitality should come from people. I want to circle back to that. But Amith, my question for you is Taco Bell struggled with some viral mishaps with their voice AI. And yes, that's comical and maybe easier to overlook with fast food, but perhaps not with an association who's built its reputation on trust. So what do you say to people who say maybe the technology is not ready?
[00:12:35:01 - 00:14:18:20]
Amith
Well, I think you can take that view and then you can probably just acknowledge to yourself that you're going to end up being a bit behind or perhaps a lot behind. So if you're not willing to take any lumps at all as you explore any technology, you're most likely a late adopter and that comes with its pros and cons and some associations are very much in that camp. And I respect that. I don't agree with it, but I respect that. What I would point out is that even if you're a late adopter supposedly implementing a more mature version of a technology, you will still take your lumps because it is new to you. Therefore, even if you wait for the next generation and the generation after that of voice AI as an example, your organization will have no cultural adaptation, no organizational knowledge, no individual training through experience in how to use these tools. And so you will still take some lumps when you roll it out, even when it's a far more complete, more well-tested solution. Now, this isn't to say the precaution of the wind and just go adopt whatever the latest is. I'm not suggesting that either. I simply think there's pros and cons and the negative side of waiting is often not really well enough understood because people just assume it'll be smoother and easier later because they're underestimating the importance of the change management side, the human side. Chick-fil-A has a really good point. My take on their point is that the human part is always important. Even if you're cool with AI doing some aspects of member service or customer service, the human integration in that process takes time to think through. You have to retrain people. So there's work involved here. It's not just as easy as plugging in a new piece of software and turning it on. So I think there's a number of important things to unpack here.
[00:14:20:03 - 00:14:23:12]
Mallory
Are you starting to see associations experiment more with voice AI?
[00:14:23:12 - 00:16:30:09]
Amith
Yeah. We're working with clients right now who are rolling out things like inbound contact center where they can essentially stand up telephone lines that are phone numbers that people can call into and AI agents can answer. That connects to the whole question of, well, what is the agent capable of doing? If it's just like a friendly chatbot that can talk to you, there's limitations to that value. And it has to be, of course, trained on your knowledge, which is helpful, but can it actually take action on your behalf? So if someone calls your member services hotline and talks to a human and says, hey, you got my registration for the conference wrong, you misspelled by name, or I have a dietary restriction. If all they're doing, if all they're able to do on the phone is say, well, go to this website link that would be an uninspiring customer service scenario for the member. You want the person on the phone, once you've been appropriately authenticated, to handle it for you. You want it to be as simple and clean and easy as possible. Right. So the idea would then be that the agents have to be able to have a governed access to tools so that they can do things on your behalf. But to answer your question more succinctly, yes, people are starting to roll it out. Customer service member service is really the obvious one because this is a weak spot. Most associations are not world class at customer service. They're not Chick-fil-A. They're not even Apple in terms of their perceived customer service quality by their members. So it's an opportunity for improvement. It's also an opportunity to relieve pressure on some of your staff who are probably overqualified to do 90 percent of the phone calls, meaning most of the calls coming in are rote FAQ type things. They're very important to the person calling, but they don't utilize the intellectual capacity and the knowledge of the people that are taking the calls. And so AI, I think, has an opportunity to create just a tremendous amount of value in this case. So we have people experimenting with it and the results are really positive. It's still very early. I would say voice AI is probably a year or two behind people's adoption of text based AI on their websites.
[00:16:31:18 - 00:16:51:08]
Mallory
So Chick-fil-A took the stance that hospitality should be human, which is not super surprising, I feel like, given how they approach customer service at their restaurants. Do you think association, if an association feels like this is core to their culture and their mission as well, that they need to take a stance on that? Like, what are your thoughts around hospitality being human?
[00:16:51:08 - 00:20:44:23]
Amith
Chick-fil-A is such an interesting case study here because their customer service for a fast food chain really is outstanding. They consistently are rated as the best in terms of customer service, customer satisfaction. I think there's a food quality component there in terms of how they train. But a lot of it is that when you go in there, you have a pretty easy experience compared to a lot of other places, even when they're super busy. You know, if you've been in a long line at a Chick-fil-A drive through, you know, around lunch hour or something, you know what I'm talking about. But it tends to move fairly smoothly most of the time. I would say that some companies that have a deep culture around customer service and they really are truly excellent at it, it's one of the differentiators. And so if the food quality at Chick-fil-A, for example, was terrible, all the customer service in the world wouldn't make a difference. But if you combine a good food product with really differentiated customer service, it can mean something to the buyer and create a loyalty level there. And so I think that they're right to think very carefully. If you're Taco Bell, though, or if you're McDonald's, not exactly known as exemplars of customer service out in the world to most. No disrespect intended, but that's just generally how those brands are not perceived in my experience. I don't think there's as much to worry about because there's really only only upside for the most part. I don't think associations are necessarily in that category. It's great to have pride around your member service quality. But oftentimes this is an area where associations struggle to keep up, especially during the peaks and valleys. And if you think about like a restaurant chain, yes, there's seasonality. Yes, there's variances based on time of day and day of week. But there's predictability around that for the most part. Associations do have predictability with their peaks and valleys, but it's much more difficult to deal with. What I'm referring to is like conference season or renewal season for credentials. If you're in a credentialing body and you know that at a certain time of year you have basically everyone or a third of your members wanting to renew their credential and they have to submit documentation and we know there's going to be issues with it. You kind of go crazy for a while, or if you have a big conference coming up, you know, associations tend to almost shut down their whole staff in terms of anything else right around when they're going into peak conference mode. And so these enormous stress points for staff are great opportunities to consider, hey, maybe we deployed AI at that point to test out how can we relieve that pressure on our staff at these key moments in time to improve quality of responses to customers. Sure. But it's also improve our staff's lives. That's another thing to think about. So there's some opportunities here. It's not doesn't need to be all or none. I think Chick-fil-A will eventually come around to this because I think the experience people expect will be faster and more efficient. And if McDonald's and Taco Bell and other brands are zipping people through their drive-throughs because it's so efficient to get orders out because of the AI, I think Chick-fil-A will have a really hard time. I also think that there's a little bit of theater to this because customer service forward brands might want to say this, but Chick-fil-A also has an excellent app. And so most of the time when I go there, I don't even actually ever order at the counter. I just like literally walk in and grab my food. It's the same argument. You know, Starbucks's CEO of the past had argued how people want community in Starbucks. They want to go into their Starbucks. They want to get to know their barista. They want to have a conversation. And that made it true for some people. I will tell you, it's not true for me. It's not that I have anything against getting to know anyone. It's just I'm busy. I want to go get my coffee and move on. So I think there are different personas and I think that associations have a lot of room to experiment here. So I think Chick-fil-A will probably come around to this at some point.
[00:20:44:23 - 00:20:57:00]
Mallory
Okay. Well, hold on to that prediction, Amith. Everybody, as we wrap up part one of the episode, Amith and I are giving you permission to go try out some AI tacos if you really want to or an AI Crunch Wrap Supreme.
[00:20:58:05 - 00:22:24:11]
Mallory
Moving to our topic two for the day, let's look at banking. At the evident AI symposium in New York in October of this year, executives from Bank of America and Capital One talked about how they keep AI costs under control. Bank of America is one of the few large companies publishing actual return numbers, which is interesting. CEO Brian Moynihan said in September that roughly 140 AI uses cost the bank around $400 million and generated about $800 million in benefit. The bank plans to double its AI budget next year. Their consumer assistant, Erica, has handled more than $3 billion interaction since 2018 and the bank's CTO has said that without it, they would need 11,000 more people answering calls. Now, the next part may sound a little bit counterintuitive. Hari Gopal Krishnan, Bank of America's chief technology and information officer, said one of the biggest mistakes he sees, including at his own bank, is rushing to AI as the solution when deterministic software would do the job perfectly well. Deterministic software is ordinary rule-based code that gives the same answer every time, the kind of thing that has been running in the background for decades. So this is the CTO of a bank doubling its AI spend, telling people in a way to use less AI. Amith, can you help us understand that concept of do more with AI, but also do less or do it strategically, I suppose?
[00:22:25:15 - 00:23:22:06]
Amith
Absolutely. So, you know, the keynote I gave a digital now a few weeks ago at this point was about sovereignty or sovereign AI. And there are three principles that we talk about sovereign AI. One of them is to own the land you build on, needing to have a platform you control, where you control the data, you benefit from the accrued context and memory over time. The second principle is to rent the intelligence rather than buying it, which is to get the best model you can for the purposes you need, because they're depreciating so quickly, don't buy them, rent them. And then the third principle is what we call control the leverage. And that's exactly what their CTO is talking about. So controlling the leverage means use AI and use leverage with AI where it makes sense. If you have a business process where, let's say you're doing something on a routine basis over and over and over again, and it's something that classical software can do just fine through rules, as you described, Mallory, there's no reason to say hi.
[00:23:23:09 - 00:26:55:03]
Amith
Oftentimes, classical software would have a gap in it where it didn't have the ability to, like, let's say, reason over something or deal with unstructured content like text or an image. But classical software is typically good at dealing with, like, structured data and more consistent types of information coming in and making a decision like, did this person have enough CEU credits to renew their credential? Yes or no. And if that's a hard and fast black and white rule, you don't need to run an LLM to say, did Mallory submit enough CEU credits to earn the renewal? It's just true or false based upon the data. So that's a good example where AI there is actually worse than using no AI because AI is not 100% the same every time. It's non-deterministic, which is what you want from AI. That's how you actually get more creativity out of it, for example. And so it's a really good point and one that I think a lot of people skim past. It's like saying, hey, we have this new tool and it's this amazing tool. Let's just use it for literally everything when we have lots of existing tools that are also really, really good. And by the way, AI can generate deterministic software. And that's a technique that I think is really powerful where you use AI to generate code that will, once tested, you know it will work the same way every time. And that's a really important aspect of the solution as well. Mallory, I do want to point out one thing that I thought was really interesting about Bank of America, just pointing out that their assistant, Erica, handled, you know, three plus billion calls starting in 2018, which is well before the era of large language models or LLMs. So they had a version of this that was handling some requests back then. I'm guessing if you looked at the curve of interactions, it's accelerated immensely in the last few years as the technology has gotten better. But the point I wanted to make though, is they're saying, well, we need 11,000 more people answering calls to handle this, which is probably, you know, a mathematically true statement. You divide the number of calls or the number of inquiries by the number of humans needed to handle it. Sounds about right. However, most likely this is one of those Jadon's paradox scenarios, where as a resource becomes less expensive, people use more of it scenario. And so what happens here is if customer service is better, perhaps people will want to use it more. Whereas if customer service is a giant pain in the ass, I'm not going to call the phone number very often because it sucks. That's terrible, right? But if it's awesome, if it's pleasant, if it's easy, if it's informative, maybe I'll call it a lot more. In fact, in one of our agents is called Izzy, which is a customer service member service agent that does this kind of stuff. The sales pitch is actually don't try to reduce the number of tickets you have. What you should actually be looking for is an absolute explosion in the number of tickets. Because yes, we want to answer them faster and answer them well. But if we're doing this well, people should contact you more often. And so it changes your perspective. Now, would that mean that if you then say, "Hey, we did really well with this. We had, you know, our association received 42,000 inquiries in the last six months, we would have needed 35 more people to handle that or 50 more people to handle that." Yes, that's technically true. But the point is, is that AI actually created demand for something that satisfied a latent customer need that you wouldn't have solved with more people. And that's the key point, right? That's the strategic differentiator. And that's probably not even like captured in their economic value upside calculation of the 800 million that you mentioned.
[00:26:55:03 - 00:27:25:06]
Mallory
Yeah, that's a complete flip of the relationship. So that's really interesting to consider. I mean, I don't want to get too much in the weeds here, but I had this question pop up in my head. So maybe others did, too. You gave the example, you know, does Mallory have enough credits to renew and how that could just be a really simple, like deterministic workflow. We have talked about on the podcast before decision models, which essentially make just simple decisions like this. Can you talk about lightly the difference between a decision model or like classical deterministic software?
[00:27:25:06 - 00:28:38:16]
Amith
Yeah, so a decision model is still AI. So it's non-deterministic. It's capable of dealing with more nuance and more variability. So in the case of a decision model, it's something like I want to evaluate the qualitative aspects of Mallory's response to a question. So Mallory responded. She gave me this answer. How should I rate her answer as, you know, good, bad, great, et cetera? So a decision model can very quickly give you a non-deterministic answer to a narrower question. Whereas an LLM like Claude or Chapsheet PT, you can literally ask it to do just about anything you can imagine. Whereas decision models are also AI, but they're models that are trained to do very narrow things. They're quite useful though, because decision models allow us to say, hey, listen, there's this intermediate space. You have classical software, which is what I'm calling deterministic software. It's just a term I think people can hopefully kind of relate to better. So the software you used to use it right throughout history of time with Witech, you have classical software that does rules based thinking and it works with structured data well. It doesn't work well with unstructured data and it doesn't really veer outside the rules. It shouldn't actually unless it's got a bug.
[00:28:39:24 - 00:29:58:24]
Amith
Then you have on the other side of the spectrum, something like Claude Fable 5.2 or Chapsheet PT, just announced, GPT 6.2, Astra and Sol, which are amazing as well. Gemini has great models. These are brilliant, brilliant models that can do just about anything you can ask them to do. That's on the other end of the spectrum, totally non-deterministic. And it's something that can solve basically any problem you could throw out at this point. But somewhere in the middle is a decision model where it isn't deterministic. It's still AI, but it is trained to solve a very narrow type of problem. Like here is a piece of text out of these 500 possible labels, which labels are relevant to this text, like what's called the classified problem. So you can say, hey, like here is a paragraph that is the summary of an abstract. Which tags from our taxonomy should we apply? Decision model can indeed do that. Decision model can do things like, you know, this user has requested extra time to complete their certification. Should we grant it or not? If that's a fairly simple decision making process, decision models could nail that for you. So they're high velocity, high scale, low cost. So that's why they've kind of blown up because it's essentially a narrower use case still on the AI side, but somewhere in the middle between classical software and AI.
[00:29:58:24 - 00:30:48:00]
Mallory
That definitely makes sense. I want to talk a little bit about the mechanics behind their cost control. So Bank of America runs a routing layer called orchestra that uses a small classifier model, meaning a simple model whose only job is to sort a request into a category. Some requests go to a large model, some go to a cheap one, some never reach a model at all. Capital One also did something similar for its car buying assistant, compressing a large model down into a smaller one trained for that single task and reported five times faster responses. Amith, on this topic at large, Bank of America can name a dollar return on AI spend, and I would say most organizations probably cannot do that. But many association leaders feel like they need to be able to do the same to justify AI spend to their board. So what do you say to those leaders?
[00:30:48:00 - 00:31:59:18]
Amith
I think you need to zoom out and ask yourself the problem you're trying to solve. So a lot of people are saying, I got to do AI stuff. Let's go to AI. And I'm not against that feeling because of course we here at the sidecar sink have been trying to get people to think about AI more deeply for years. So getting going is important, but you shouldn't necessarily go out there and just start investing in projects just because you heard from a colleague or friend that an AI agent for knowledge is important or an AI analytics agent is a good idea or any of these things necessarily. You should have an idea of what you're trying to achieve. The problem with that, of course, is a lot of people don't know what they want to achieve with AI until they get some experience with it. So we're kind of in this circular logic of sorts. Education, of course, can help. And obviously we here at Sidecar are all about helping with that. But there's lots of ways to educate yourself on AI that can help you open up your creative aperture a bit and understand what the possibilities are. You can also partner with an organization to come help you think through this. But the idea is to focus on the outcomes. If you were to say, hey, I've got this investment and it's really great AI tool, but nobody's using it, it's not actually moving the needle for us in terms of ROI.
[00:32:00:19 - 00:33:00:14]
Amith
That perhaps might be somewhat predictable in the sense that maybe you didn't have a really clear thesis for why you needed the tool to begin with. Just like you wouldn't say, hey, I haven't replaced my AMS in a few years. Let's replace the AMS. The question would be like, well, why do you want to replace the AMS? What's the point? Is the current AMS literally sinking? Is there something you want to do in your business that the current AMS doesn't support? There should be a business reason behind any AI tool that you go and select and implement or something that you build. I think a lot of people are missing that. They're missing that front end step. This does not need to be a six month long committee driven bureaucratic nightmare type thing. It just needs to be a fairly brief exercise. And we call it phase zero over here, but you can call it whatever you want. The idea is to actually zoom out and think about the business and then map the priorities that you go execute to that. It sounds so simple, Mallory, yet I see very few associations doing that now. They're in the mode right now. Let's just implement a pilot. Let's see what happens.
[00:33:01:15 - 00:33:23:01]
Mallory
So if an association does have this business goal in mind and they have decided a way to use AI to get there with the associations you're working with, Amith, are you seeing them be able to produce tangible number results or we reduce this by X percent or we, it's things like that. Or is it a little more abstract in terms of results?
[00:33:23:01 - 00:34:01:08]
Amith
No. When you have a clear set of goals and you start off with these outcomes and say, hey, we're going to improve member experience. Well, great. Well, what does that mean? You have to ask the question enough times where there's a very clear, you know, measurable outcome that says, did we actually achieve that? So what does that mean? Well, we want our wait times to be, you know, our average response times to go from one business day to one business out. Or we want to be able to handle three times as many calls or we want members to rate us in a certain way on an NPS survey. You have to have the instrumentation. Sometimes you don't have such things, but you have to have the instrumentation and a baseline to measure against in order to actually objectively say you've made an improvement.
[00:34:02:08 - 00:35:33:15]
Amith
Measurement can be a fallacy because if you're measuring the wrong thing, it doesn't necessarily help anyone. But if in the absence of measurement, you can't really tell me that you've made an improvement or not made an improvement. Most associations aren't looking to slash costs. Some are, but most of the time they're looking to drive more member value. That's generally the thing people talk about, but they don't really know exactly what that means. They have an intuitive sense of what it means, but they haven't actually taken the time to say it means this outcome. And if we achieve this outcome, we will unequivocally say we have improved member value in 2026 or 2027. That exercise probably can be done in a week or two and get you a much clearer idea of then what you should go invest in. The other piece that's missing is education. So a lot of associations still are in the mode of let's educate a handful of people. Let's lean on technology team. Let's let a handful of people who are more technology focused do this. This isn't everybody needs to get together and learn this problem. You will not advance your organization from a change management perspective if you don't take the time to train everyone. Plus, I view it as your obligation as a leader. I've said this many times on the pod when I've been speaking in various places. A leader who fails to train their team on AI is committing leadership malpractice because you're not comparing your people to be competent and qualified to do any job in the future. So that is something that's still missing so deeply. Or people have checked the box and said, hey, I did whatever training. And we're like, well, once you do that, it's like, oh, you see you're in a half ago. It's like nothing's changed in the eye since then, right?
[00:35:33:15 - 00:35:38:04]
Mallory
What if they tell you they're listening to the side crossing every week? Do you think that's enough?
[00:35:38:04 - 00:35:53:24]
Amith
I'd say that's a great appetizer. OK. OK. I don't think it counts as the full, you know, the full meal. You still have to go and figure out how you actually drive learning. And a lot of that's, you know, getting people at every level of your organization actually doing stuff.
[00:35:53:24 - 00:37:20:07]
Mallory
I want to move to topic three, our last topic for the episode, which is CureWise. This is an cancer navigation platform that launched publicly in June of this year. The founder, Steve Brown, was diagnosed with a rare blood cancer in early twenty twenty five and built the first version for his own treatment before turning it into a product. It sold directly to patients right now at forty nine dollars a month. And here's how it works. A patient can connect their medical records, which the company pulls electronically from thousands of US health systems rather than making people upload files. This is so fascinating to me. The system then runs what it calls a virtual tumor board. A real tumor board is a meeting where oncologists, surgeons, radiologists and pathologists review one patient's case together and agree on a plan. Top cancer centers do this as a matter of course, and many community practices do not. CureWise simulates it with several agents, each taking a specialist role, then looks for where they agree and where they split. It also matches the patient against the database of active clinical trials, runs answers through chat GPT, Claude and Gemini to compare them and catch fabrications and generates a list of questions the patient brings to their actual oncologist. So I mean, a virtual tumor board to me, like I said, fascinating. Can you think of any association specific examples where being able to simulate a virtual discussion with agents playing different roles could be useful?
[00:37:20:07 - 00:39:33:03]
Amith
Yeah, I think I mean, first of all, I love the idea and the idea of being able to bring together different types of expertise to create diversity and thought. In this case, to better, you know, to better increase the probability of an outcome that that everyone wants. So I think for an association, there's a couple things that that comes to mind right away. But first, what if you put together a virtual board of some of your key members to get their input on what they think makes sense for you to do and use this to help guide strategic decision making. Perhaps there's still hold good use for actual real conversations with real members, of course. But I do think that the AI can be quite good at simulating personas. And you can tie this to the data you have in your system as well. So let's say you were to just use chat for this, but it could probably do a reasonable job at this point, because the models are so smart and they have such a broad training data. But they would not know much about your association or your members. So instead of that, you could say, hey, within a secure, governed environment, I'm going to spin up some agents that will study some of our most prolific members, some of our most engaged members, as well as some of our members who are least engaged. Then we're going to go off and study the publicly available data beyond our own internal systems, their LinkedIn profile, their social media, any web posts and things like that we can find. And that becomes the persona baked into this virtual individual that becomes part of the board. And then you have this board have conversations. And so I think it can be very, very powerful. This has essentially become agents, right? You can talk to these agents whenever you want to. I think there's tremendous opportunity around this. And in solving problems like you think about like a board or a committee, associations form lots of those. Think about abstract submission. I talk about this fairly regularly. Associations often have a choke point around how their human committee members do reviews of inbound proposals to speak at a conference or submit articles for a journal. And so it's possible to utilize the same approach to get like a diverse AI opinion about a particular proposal.
[00:39:33:03 - 00:40:21:19]
Mallory
Mm hmm. I want to talk about to the problem that CureWise is aimed at, which is pretty well documented. A 2025 study of more than 300,000 lung cancer patients found only 62 percent received treatment that followed national guidelines with academic and high volume cancer centers doing better than and community programs doing worse. Those guidelines come from the National Comprehensive Cancer Network and Alliance of Major Cancer Centers that publishes the standard of care. The knowledge exists and is public, but getting that knowledge to the exam room is the gap. I mean, I wanted to share that stat because I feel like this is highlighting that the hard part here is not the AI necessarily. It's we have the knowledge, it's public and it's still not reaching people, which I feel like is pretty relevant for associations. So what can you say about that?
[00:40:21:19 - 00:41:48:08]
Amith
It's staggering. I mean, this isn't a question of is the science sufficient to cure more patients. You know, 38 percent of the people didn't receive treatment that followed national guidelines. The national guidelines are probably a step behind what the very latest is on the cutting edge. So we're not even talking about necessarily the most cutting edge ideas. It's just the national guidelines. And at 38 percent of cancer patients in the United States in, you know, as broadly of a study, various lung cancer aren't getting the treatments they could be getting, should be getting. That's both tragic and an immense opportunity to solve that gap because this is a resource problem. Community based hospitals and health practitioners don't have the resources of Johns Hopkins or Mayo Clinic. And so how do you level that? Well, this could be a tool that levels the gap or at least at least takes part of it out. So I think that's quite exciting. And this similar thing exists probably in many other fields. Right. This is something that's so visceral when you hear it. 38 percent of the lung cancer patients in that study didn't get the treatment that they were that they should have gotten per national guidelines. It immediately, like, you know, affects you. But I think that the same thing is true in many other fields. So this is a clear opportunity for AI. And I'm really excited to hear about this particular tool. I'm sure there's others like it that are probably doing very similar things.
[00:41:48:08 - 00:42:26:08]
Mallory
I know we have a lot of experience with associations who are feeling this and want to make the content that they have. More easily accessible to their members, more findable on their website. But I'm curious from your perspective, Amith, when you see associations roll out, I'm going to use the name Betty because that's one of the companies in our family of companies. I'm sure there are different ones. But when you see associations utilize Betty, how are they able to get their members to actually use the assistant? Because I feel like that's kind of a secondary problem is, OK, now our content is easily accessible. But now how do we get people to actually use this tool?
[00:42:26:08 - 00:44:26:18]
Amith
Yeah, I mean, you know, the friction is always the thing that's the barrier between people and what they seek. And so when you think about this, people often default to a good enough answer, even when they know there's a better answer or a better resource available. You know, so if you're like, hey, you know, I really like this particular ice cream shop, but it's 14 miles away from my house and there's a lot of traffic and I know there's going to be like a 30 minute line. But there's a pretty good ice cream shop just down the street. It's like a two minute walk from my house, a little bit less expensive, and there's usually not much of a line there. Lower friction, not as good of a product. Maybe you go to the place that's nearby. And that's the same thing that's happening with Google and chat GPT compared to your association. That's what's been going on for decades now. And with AI, not having an solution at all means you have a real problem because chat GPT and Claude, you know, aren't like a lesser ice cream, but they're like an almost as good answer in many in many cases. They might even be a better answer if you're being truthful about it. I would say associations do have the opportunity to have the best answer and to also make it so that they have no friction because tools like Betty make it possible for people to come to your website. You can make this totally available for the public or you can make it, you know, something that's behind a member login. But answering questions very easily and also helping solve problems. It's not just answering questions like you do on Google, but it's like digging into iterating with and actually even producing reports and artifacts. Things like you would do in like a Claude or a chat GPT environment. But doing that in a governed, sovereign workspace the association provides also provides safety and a feeling of safety and actual safety to the members, which they might be concerned about. And certainly in some of these fields to do that in a public AI like a chat GPT. So I think associations are finding success. My experience in this subcategory is in the world of Betty as well, since that's one of our companies. So keep that in mind that there are obviously other ways to do this.
[00:44:27:22 - 00:46:36:00]
Amith
We are seeing people have significant increases in frequency of engagement, but as importantly, depth of engagement. The average session that Betty users had is over 10 minutes. The average website visitor to an association is under 30 seconds. So clearly people are sticking around for a reason. They're having multi-turn conversations and they're coming back way more frequently. So there's something there. And what it is, I think, is it's the brand trust of the association because people know the association has good stuff and it's low friction. It's easy. So the combination of the two is really attractive. If I can get the best answer and it's no harder than getting a good enough answer, I want the best answer. The other thing I'll mention is there's opportunities for associations to play in the broader ecosystem. So once you set up an AI knowledge agent like a Betty, you can connect those to external tools like chat GPT through something called the Model Context Protocol or MCP. If you've heard of it, you can create an MCP for your agent and allow your members to have access to it so that they can have their cloud or their chat GPT automatically talk to your Betty or your whatever your your AI agent is so that they can just kind of stay where they're at. But they're still benefiting from your content. And the key to this is that you're not handing over all your data to chat GPT or a cloud. Those tools are simply talking to your MCP server and they're getting back a specific answer to a specific question. And you could argue at scale that that potentially creates a knowledge training opportunity for those tools, which is a reasonable concern to at least think about. But you're also what you're focused on there is lowering friction further. I think of cloud and chat GPT as the front door. So that's the way I describe to people is people are going to be in those tools just like they're on Google. That's great. You need to be there. You need to be part of that environment, but you don't provide everything there. You provide that as a front door so people can access the very high level stuff. And then you provide links that take them back into the full experience that you can provide in your association.
[00:46:37:05 - 00:46:42:02]
Amith
I think that's the key differentiator between the two approaches is that they're not separate islands. They're connected.
[00:46:43:13 - 00:46:54:21]
Mallory
And I want to double click on something you said earlier, which is, you know, when customer service or member service is good, the depth of interaction goes up. And I feel like those stats you just shared with Betty show that.
[00:46:55:23 - 00:48:54:02]
Amith
Totally. You know, we ask people, how often do members contact your member services? And if you divide the number of cases by the number of members and just kind of do like an average math and look at it independent of individuals, it's a pretty low number actually. So you look at like a number of distinct individuals that actually have contacted you. It's a small sliver of your membership. Usually that's usually the case. And we asked the Bush will, why aren't all your members contacting you regularly? They're like, well, they don't need anything. We've already solved their problem. They already are a member. They've already signed up for the annual conference. So like, do you think the member sits around thinking about the world from that lens? Like that, have they done the things the association wants them to do? Of course not. They're out there delivering care to their patients. They're solving engineering challenges. They're working on scientific problems. They're in their field doing their job. Um, and so the question is, is like, what can you do to be more useful to them on a high frequency basis? And if you're able to unlock your knowledge and said, Hey, listen, if you imagine a virtual member, that's the smartest, most eligible member in the history of our association in our discipline is available 24 seven, 365 to answer any and all questions and have conversations with any of our other members. That would be an interesting resource. That would be probably quite popular with most people or with a lot of people. And that's exactly what you're doing when you're, when you're tapping into not just your, your knowledge in this context, but also structured databases. So the whole range of associations have databases of information like benchmarking data from their industry. And they're the holder of this unified set of data from across a multitude of firms in their space. And it's possible to set up AI agents that work against that. So there's, there's tons of opportunity here. I think the association has to be very thoughtful about making sure that friction is relatively low and that they're playing nice. in other ecosystems, but at the same time, not doing what I like to call backing up the semi truck with all their data and dropping it off at Sam Altman's doorstep in San Francisco.
[00:48:54:02 - 00:49:08:03]
Mallory
Don't do that. Don't back up the semi truck. Everybody, what you should do is go and get some AI tacos. In today's episode, we revisited AI Crunchwrap Supremes and the voice agent now taking orders at more than 900 Taco Bell drive-throughs. We looked at Bank of
[00:49:08:03 - 00:49:13:18]
(Music Playing)
[00:49:24:11 - 00:49:41:10]
Mallory
Thanks for tuning into the Sidecar Sync podcast. If you want to dive deeper into anything mentioned in this episode, please check out the links in our show notes. And if you're looking for more in-depth AI education for you, your entire team, or your members, head to sidecar.ai.
[00:49:41:10 - 00:49:44:16]
(Music Playing)
In Part 2 of their Fundamentals of AI series, Amith and Mallory move from definitions and building blocks into what abundant intelligence means in...
Live from digitalNow 2025 in Chicago, Mallory Mejias speaks with leaders from the American Paint Horse Association, the Missouri State Teachers...
Amith and Mallory trace the Spirit Airlines data deal, unpack OpenAI's Astra model solving open math problems that had stumped researchers for...