Custom GPTs and the Argument for Pausing Legacy Upgrades in Favor of AI [Sidecar Sync Podcast Episode 04]

Show Notes

Join Amith and Mallory as they guide you through the process of creating Custom GPTs and discuss potential applications for associations. Then, shift gears to a critical discussion on whether associations should pause legacy system upgrades in favor of AI's transformative potential. This episode promises practical insights, challenging questions, and a deep dive into the strategic choices facing today's association leaders.

Let us know what you think about the podcast. Drop your questions or comments in the Sidecar community: https://community.sidecarglobal.com/c/sidecar-sync/

• Download Ascend: Unlocking the Power of AI for Associations: https://sidecarglobal.com/AI
• Join the CEO AI Mastermind Group: https://sidecarglobal.com/association-ceo-mastermind-2024/

Thanks to this episode’s sponsors!

Betty Bot: https://bettybot.ai/ 

rasa.io: https://rasa.io/ 

Tools/Experiments Mentioned

• Amith’s Insights GPT: https://chat.openai.com/g/g-9JNGi82aT-amith-s-insights 
• Create your Custom GPT: https://chat.openai.com/create 
• MemberJunction: https://memberjunction.com/ 
• Microsoft AutoGen: https://microsoft.github.io/autogen/docs/getting-started  
• Microsoft Azure Bot: https://azure.microsoft.com/en-us/products/ai-services/ai-bot-service 

Articles/Books Mentioned

• Ascend: Unlocking the Power of AI for Associations: https://sidecarglobal.com/AI  
• The Open Garden Organization: A Blueprint for Associations in the Digital Age: https://www.amazon.com/Open-Garden-Organization-Blueprint-Associations/dp/1599329557 

Social

• Sidecar's LinkedIn: https://www.linkedin.com/company/sidecar-global
• Amith's LinkedIn: https://www.linkedin.com/in/amithnagarajan/ 
• Amith’s LinkedIn post on legacy upgrades vs. AI: https://www.linkedin.com/posts/amithnagarajan_associations-ai-ams-activity-7129274837520318464-NSA8?utm_source=share&utm_medium=member_desktop 
• Mallory's LinkedIn: https://www.linkedin.com/in/mallorymejias/

Hello everybody and welcome to the latest episode of Sidecar Sync. I'm excited to be here with you this afternoon. There's so much going on in [00:01:00] the world of innovation and AI. My name is Amith Nagarajan. I'm here with my co host Mallory Mejias and can't wait to jump into the episode.

We've got a lot to talk about today don't we?

Mallory Mejias: We do.

Amith Nagarajan: And before we jump into the meat of the content, I wanted to take a moment to thank our sponsors for this show. For this episode our two sponsors are rasa.io and Bettybot AI. Rasa.io provides AI newsletters for associations. Rasa.io has been doing this for a number of years now working with a couple hundred associations in the market, providing an amazing, highly engaging newsletter product that brings up engagement considerably through personalized content.

 Our second sponsor this week is Betty Bot AI. Betty provides an AI assistant for your association to provide anyone who's interested in your content with expert answers on demand. Betty is [00:02:00] capable of ingesting all of the content from your association, including your past conferences, all of your articles, journals, publications, and so on, and through that knowledge is able to answer questions at the expert level. For anyone, any time, check out more at Bettybot.ai.

Mallory Mejias: Thank you to our sponsors. I'm really excited to get this episode kicked off. Last week, I was pretty inspired by our conversation around custom GPT. So I sought out to create my own and Amith, you did as well, right?

Amith Nagarajan: Yes, I did. It's an interesting experience to say the least.

Mallory Mejias: Absolutely. And we want to share what that experience was like and what we created. If you didn't catch the last episode, custom GPTs are custom versions of chat GPT that you can create for a specific purpose. Anyone can build their own GPT. No coding is required. You can make them for yourself, for your company's internal use or for everyone.

Before last week's episode, we weren't able to play around with the tool for ourselves, but since then, we [00:03:00] have each created our own custom GPTs. First, I wanna give you a little overview of what the creation process was like, and then we can dive in on the GPTs that we created. So when you go to chat.openai.com/create, I believe we'll add that in the show notes. You can create your own custom GPT. There are two tabs there. One that's called “create” and one that's called “configure.” In the create tab, you can actually interact with the GPT builder through chat. Just how you interact with chat GPT.

In the configure tab, you can manually build out your custom GPT. And that was the route that I went. I'm curious to know, Amith, did you go the create route? Did you go the configure route?

Amith Nagarajan: You know, I went the create route, so it's actually a really interesting thing to point out that. Software tools for a long time have had capabilities to configure. You know, that's obviously a thing you do in most software products is you set [00:04:00] up different settings. And what's interesting about the Create tab is it's essentially the power of the AI just having a conversation with you about what you're interested in doing and then doing the setup for you.

So. To me, it was more of a curiosity. I'm usually pretty good at using things like the configure tab, and I did go and check that out afterwards to see what it put in, but I think it's, it's quite telling of what we might be seeing in a number of products over time where these AI agents will help us with the setup process.

Mallory Mejias: Pretty funny to me because we flipped roles there. I assumed that you would use configure and that's what I used. I'm surprised that you use create so we can both kind of talk about those experiences in the configure tab, the one I used, you have to come up with the name for your GPT. You can add a short description of what it is, which will be showed when you click on the link to your custom GPT. You can also add in instructions. What does this GPT do? How does it behave? What should it avoid doing? From there, you can build out conversation starters. And then in the configure [00:05:00] tab, there's an area where you can upload files that you want your GPT to use when it's answering questions.

You can select the capabilities that your GPT uses. It looks like as of now, the only ones that are available are web browsing, Dall-E image generation, and code interpreter. And finally, you can add actions, which is when you let your GPT retrieve information or take action outside of GPT.

I did not play around with this part at all because I wasn't too sure what to do with it the custom GPT that I created was called the Sidecar email assistant. I was hoping to create a custom GPT that could really quickly help me generate emails in the fun tone of Sidecar. I figured I could do this pretty easily by training the model on prior Sidecar emails. So I found eight emails that we had sent in the past.

I could have done more, but I did this in a really manual fashion of just copying and pasting these old emails into a Word doc, which I then uploaded into the custom [00:06:00] GPT creation process. From there, I told it that it was an email copywriting assistant, a B2B expert, and I put in a few conversation starters as well.

Which were, “I want to write an email about an upcoming event that we're having,” or “I want to write an email about an upcoming product launch.” I tested this out really briefly, and the email it created was great. It, to me, seemed like it was written by any member of the Sidecar team, and I thought it was a pretty neat thing that I created in about 10 to 15 minutes.

Amith, I'm curious to hear what custom GPT you created.

Amith Nagarajan: Yeah, first of all, I think that's just a fantastic example of something pretty narrowly focused Mallory that you created. Just as a tool for yourself. And it was a way for you and the Sidecar team to basically be more efficient, but also perhaps, you know, get some creative help in ways that, you know, might be interesting coming from the AI.

One quick thing I want to point out about something you [00:07:00] said a moment ago was the instructions you provided to your custom GPT that it is a B2B expert and an expert in email. And also, of course, you trained it on examples of the types of emails you want to create it in the first piece of that, in terms of the instructions you gave it on the role that the AI needs to take on.

That's a part of the prompting process when you're manually working with ChatGPT or Claude or Bard. That's really important. And a lot of people omit that where they just ask their question of the AI and the interesting thing about it is the AI you might say, well, I'm asking it a question about email, it should just automatically know that it's an email expert. But interestingly, the quality of the answers you get from the current state of these models is dramatically improved if you tell it that it's an expert in what you're about to ask it, which sounds strange. It's totally counterintuitive. But that's a great thing that you included that, and that's probably a big part of why you saw great results in your custom GPT.

 So to answer your question on my custom GPT is, I've written [00:08:00] two books in the last few years. One is Ascend, Unlocking the Power of AI for Associations that a lot of our listeners are familiar with. And the other one is perhaps less familiar to most of the listeners of this show.

It's called The Open Garden Organization. And it was essentially a book that I wrote for the association market in 2018 that argued for a Having a different frame of reference for audience growth, not just to think about membership, but to think more broadly about how to engage people that are interested in your topic of expertise.

So there are two bodies of work that I've created over the course of the last several years, and in both cases there are, I think, instructive in terms of things associations could be thinking about to improve their businesses and obviously reflective of what my brain thinks. And so I created a custom GPT called Amith’s insights and this GPT, I posted it on LinkedIn.

We'll include it in the show notes as well. And that GPT basically knows how to answer questions the way I would. And it's pretty good. I uploaded both books, manuscripts, the PDFs [00:09:00] to the GPT configuration. That was something I couldn't do through create through the conversation.

At least I didn't figure out how to do that. But then I'm pretty sure you can do that. I just did it in the second tab and then after that I provided a bit of instructions. I said, I'm an author, I've written two books. I care deeply about the association market work in a style where you are encouraging yet also aggressive in pushing the people who are asking you questions to take some risks to try things that make them uncomfortable, things that, you know, kind of are my style of how I work with the association community, while obviously being empathetic to the challenges that we all face in this sector.

So I thought I did a pretty good job. You know, when I asked you questions, it didn't feel like super generic ChatGPT. It definitely still felt like ChatGPT overall when I interacted with my my GPT, which is obviously what it's based on, but it had a little bit of my personality in there and it definitely had a lot of the knowledge base that I provided it.

So I was pretty pumped about it. I spent a similar amount of time. It was probably 10 or 15 minutes.

Mallory Mejias: Really cool stuff. You mentioned briefly prompt engineering, and I [00:10:00] think that's something worth diving into. I assigned my custom GPT a role of being a B2B expert email copywriter. You assigned your GPT kind of a style, maybe being more aggressive. When we do that, what happens on the back end? Why is that necessary?

Amith Nagarajan: Well, think about the AI model is having this massively, you know, broad set of world knowledge and having been trained on billions and billions of pieces of information. And so by doing that, what it's doing is homing in on the pieces of content that the AI has ingested effectively into its neural network as part of its training process that are most related to the way you want it to answer because it can answer. Remember, like a lot of the early examples of chat GPT were kind of, you know, on the one hand toy examples, but interesting, like you know, answer in the style of Shakespeare answer in the style of, you know, some famous actor or create a rap in the style of a particular artist. All really cool. But they point out the [00:11:00] fact that the information that the AI is providing the user in response to the question can actually be styled literally unlimited number of ways. And so by providing it that kind of direction essentially, you're giving the model instructions on the parts of the neural net and parts of its training, essentially, to utilize in forming its response.

Mallory Mejias: That makes sense. I know for the custom GPT I created, I was a bit worried beforehand, and I couldn't find this information out there, so I wanted to share with listeners, in case you have the same concerns, that when you're creating a custom GPT, you can create one just for you, one that you can share with others, that can only be accessed via a link, and then one that can be shared with the public.

So I, of course, Only allowed mine to be shared with those who have the link since it is creating emails on the Sidecar style. Amith, did you create a public facing one?

Amith Nagarajan: I did, because mine was intended to just kind of be a you know, experiment in front of the world kind of thing, so I shared the link with LinkedIn. I don't know if ChatGPT or if they give you [00:12:00] statistics on how much usage your GPT has received. I'm guessing that they'll, they will provide that to you at some point.

Probably not user level information, I wouldn't imagine, because that would be a privacy issue. But just knowing, like, the number of people who've asked questions, and the most common questions that have been asked, and things like that, would be kind of interesting for the GPT creator. But yes, I did share it publicly, and for me, it was an interesting exercise to say, Hey, as an author with two books out there could I provide more value to my readers on these subjects?

Especially if you think about, like, both books being combined into one GPT. I think it's kind of interesting, you know? So, yeah, that's what I did. I put it out there publicly. Now, that doesn't mean the manuscripts of my books are available publicly. So, I trained it on these two PDFs. As a user of my custom GPT, you cannot get those PDFs directly.

But those PDFs are used by the GPT to answer questions that come in.

Mallory Mejias: I mentioned earlier the action setting under the configure tab. I clicked on this one briefly and then immediately clicked off because I realized I didn't know exactly what I was [00:13:00] doing here. Can you talk a little bit about that?

Amith Nagarajan: Yeah, and I don't think OpenAI knows exactly what you can do with it either because it's very much an early stage feature. But Actions is when you cross the line from bot to agent. So this is when your custom GPT can really start to do things for you. And I'll give you an example that extends what you've built.

So Mallory, you've built for the Sidecar team, this Sidecar email assistant. Well, imagine if you could say, hey, I like that email, go ahead and schedule it to be sent out tomorrow to this campaign. And let's say that the agent knows that for you, you're using HubSpot or using Salesforce or whatever your CRM is, and it has been previously authorized to create the email in that system for you.

That would be the next step it could take, right? That's the form of agency. So that would be a natural action that you could add to your GPT. I don't know if there's a HubSpot connector, but I'm sure there will be for all major systems, you know, and this is kind of like the world of Zapier, where, like, the whole world connects in.

So that's really exciting. I [00:14:00] think you have to be very thoughtful about this because, you know, when there is auto generation of content that doesn't mean that you just post it and go with it. Just like, you know, teamwork in a human team where one person might create a piece of content, someone else would typically review it if it's going to go out to an audience.

The same thing is true for these agents. A quick side note on that is in a prior episode, we talked about multi agent frameworks and Microsoft's auto gen open source tool, and that type of tool actually provides a lot of additional power because you might have one GPT working with another GPT, which at the moment, custom GPTs from OpenAI are pretty simple.

I envision that they probably will do this type of multi agent thing in the future, where you might have the sidecar email assistant, which is the writer. You might have another one called the sidecar editor. And the Sidecar editor has been given all these rules about things to look for, problems to watch out for.

And the Sidecar editor agent... Interacts with the sidecar email writer agent, [00:15:00] and together they form a final draft. I would still recommend that the era that we're in, that there's always a human in the loop that reviews and approves the content before it gets fired off. But that could really improve the quality of the outcomes.

I imagine OpenAI is going to have that built in. I'm sure the other major providers in the space will as well. And there are frameworks that can make it so that you are vendor agnostic, like. The autogen framework is really cool for this because you could do what I just described without being directly wed to a particular company like OpenAI.

Mallory Mejias: Have you thought about adding any actions to your Amith's Insights GPT?

Amith Nagarajan: I haven't really thought about it deeply because the experiment to me was just a way of testing it out because I wanted to see how custom GPTs stack up against other approaches for doing in, in AI, the, in the AI realm, people call this RAG, , which stands for retrieval augmented generation, which simply means that you're augmenting the language models knowledge with specialized content.

And this is a really important area for [00:16:00] associations to pay attention to. There's a bunch of tools and a bunch of ways to do RAG. This is one, right, where you can upload a PDF or provide web links to a custom GPT. There's some pros and cons to that model. There's also tools that are out there like inside Microsoft Azure.

One of our keynotes at the Digital Now conference. Was it last week or was it two weeks ago?

Mallory Mejias: I think it was last week.

Amith Nagarajan: Yeah, I honestly can't remember I went to Starbucks today and I got a cup of coffee And I noticed it was the holiday theme and I didn't realize we were close to thanksgiving until that happened. So it's all moving along fast.

But in any event one of the keynotes, from Microsoft that spoke at digital now She illustrated an example very similar to this using Azure's framework for building an agent where she was able to upload a PDF and then build an agent that had a conversation with people using that PDF. There are lots of third party tools you can use to do very, very similar things.

Some are very simple and some are more enterprise scale. And so actually one of our sponsors of this [00:17:00] podcast, BettyBot.ai is a good example of the latter. Betty is a very sophisticated framework for ensuring what's called grounding. Which is this concept of essentially guaranteeing that the answers are coming from your content.

Whereas, you know, and you can kind of look, think of this as a sliding scale where something like Betty is going to be really, really, really rigorous in terms of guaranteeing that the answers actually are coming from vetted sources with citations from your content and only from your content. And for a lot of associations that’s really important.

But if you're doing something a lot more casually where that's less mission critical than something like a custom GPT would be great because the custom GPT still uses the broader general knowledge of ChatGPT to form an answer. So if your email in your case, Mallory of the sidecar system email, it's not only using your emails, it's using all of the emails it's ever seen to formulate its response.

And maybe that's good in your case, actually. In my case, for the Amith's [00:18:00] Insights, I'm actually kind of nervous about that as an author because I don't want people to interact with that custom GPT and get, really receive information that is outside the scope of my two PDFs that I uploaded for my books.

Now, I did attempt to, and I put this in air quotes, “ground” my custom GPT by giving all caps instructions multiple times in my instructions text box to the custom GPT saying, you will only use my PDFs. You're not allowed to use your general knowledge. And I put that in all caps, just like I'm kind of, you know, forgot to undo my caps lock or I'm yelling at them.

And actually, by the way, that affects the model. The model understands what all caps is. The model understands when you're repeating yourself. The first and the last thing you say to the model are often what it remembers the best in the instructions. So I tried to mess with it a little bit and see if it would give me like some generic answer outside of the scope of content I know that's not in my books.

And it was okay, but it's still kind of generalized. Whereas a tool like Betty is on the other end of the spectrum being really buttoned [00:19:00] down in terms of that type of capability and is able to ground it and actually force itself to have citations from your body of knowledge. In order to consider an answer valid, but that's not necessary in all cases.

So I think there's the beauty of this and what's so exciting about custom GPTs from OpenAI is this is the first consumer grade tool that's really easy. That's, you know, any of us in few minutes can go set it up. So the broader opportunity is that the knowledge base you have can be activated. That's the simple way to put it.

You can activate these incredible assets you have and increase the accessibility of them.

Mallory Mejias: This concept of grounding is important. I know for me, I was wondering when the Sidecar email assistant was generating emails, exactly how much of these emails were coming from what I provided. Or what the ChatGPT, large language model has access to, I can imagine association leaders or staff right now from scientific organizations are listening, thinking we could never build a custom GPT, right?

Based [00:20:00] on our knowledge base, because we would have to guarantee that the answers it's generating are 100 percent or as close to 100 percent as possible in terms of accuracy. Do you think we'll have that ability with custom GPTs at some point to say only use this information, or is it better to look elsewhere?

Amith Nagarajan: I don't know. You know, it's a great question. I imagine that over time, the capabilities of custom GPTs will grow really quickly. So I wouldn't be surprised if they added capabilities that were along those lines. You know, things that you might want to think of as being similar to the grounding concept we've talked about here.

But I also think that you have to look at what the vendors trying to go after. So open AI is going for the broad swath. They're trying to enable well literally millions of people to create these things, which is super exciting. And that may not be a high priority to them. It may be a higher priority for them to make it even easier to create at a basic level.

I also think custom GPTs are very likely to go multimodal very soon, meaning that you'll be able to talk to them the way you talk to regular ChatGPT. And possibly the creator can infuse their own voice, which is both [00:21:00] exciting and scary, right? To be able to do a voice clone and you're actually like as the author talking to people when they talk to your, your custom GPT.

I suspect that they might go down that path before they go down the enterprise. content path. But to be determined, I think the space is moving so quickly. You can't really, you can't really know. It's the short version of my answer on that.

Mallory Mejias: How do you envision associations using custom GPTs? Do you have any practical examples?

Amith Nagarajan: Sure. Well, I mean, I think custom GPT is like what you did. Mallory is a fantastic way to get started. It's low risk and it's a quick way to learn. So building an email assistant GPT or you know, if you're an association that deals with a lot of abstract submissions, an abstract review GPT that might be trained on your particular standards of what you require for a submission. For, let's say, a call for or a call for papers. You can provide all of the technical requirements for each of the submissions, and you can ask the custom [00:22:00] GPT to grade each of these pieces that are coming in. And it'd still be, at the moment, it would still be manual where you'd paste it in and say, here's the submission, give me the report.

But that would be a very useful tool that's what I'd call semi-automated, where, you know, there's still a user involved in passing the data in and pulling it back out. That could be interesting. I think there are some member facing opportunities out there. Maybe a custom GPT to help with events, to help with, you know, answer member questions about common topics.

But custom GPTs probably aren't going to have great gateways in a secure fashion to your enterprise data. So you're not going to likely have like a customer service bot that's able to renew memberships or register someone for an event, because I don't think you're going to, first of all, I don't believe that the actions capability as it is now, or probably what's planned are intended to do that kind of enterprise connector capability. But I mean, I could be wrong about that. That may be the direction OpenAI goes. The good news is you can build stuff like this in a lot of other flavors. There's ways to create those experiences.

But I would suggest that associations listening in to this. [00:23:00] Do exactly what Mallory did and create a custom GPT that's intended for staff use. That's just kind of like a smarter version of ChatGPT. If you like ChatGPT a lot, but you're like, I wish I could just have it remember every time I'm creating a new campaign to remember these rules.

Well, now you can. You can create a custom GPT that has that set of rules in it. And it's like having another person on your staff. It's pretty powerful.

Mallory Mejias: If this is someone's first time listening to the Sidecar Sync and maybe they haven't played around with AI a ton and they are hearing about these custom GPTs and they're ready to go. Are there any potential risks, Amith, that you want to highlight or warnings that you want to share with listeners about just diving in with custom GPTs?

Amith Nagarajan: I think the number one thing is, is the content you choose to upload is something that you have to be thoughtful about. OpenAI does have some privacy safeguards in place. So contrary to popular belief, what you pass in isn't automatically ingested by a future model training process. It is possible that it could be and you can read the terms of service, but if you're either an enterprise customer or if [00:24:00] you're using the API to pass in data, we know definitively that that data is not used to train a future AI Model by OpenAI model far as what you do with your paid 20 a month Chat GPT plus account. There's a setting you can turn on to not retain history. And if you do that, it's not gonna retain history. Now what I don't know, because I haven't evaluated the terms of service for custom GPTs is whether or not like PDFs like what I uploaded are things that also are protected. I believe they are from what I remember hearing in the keynote, but I'm not, I'm not sure about that. So that actually a great thing for us to follow up on and put in the show notes if we have a clear answer on it.

But either way, I think you want to think about the content you put in there as being kind of pseudo public where you wouldn't be too stressed if it got out. Like in Mallory's example, she took eight prior sidecar emails. Well, those are emails that have been sent out to people, so they're not You know, super sensitive data.

Mallory didn't take, like, financial statements from the company or something like that and put them in the GPT. So I think that if you're operating in this kind of, you know, [00:25:00] semi public, quasi public space, you should, you should feel totally comfortable. If you're going beyond that, where the data, like, if it got out, if it became public, if a model was trained on it, if that freaks you out a little bit, then that's probably not the first thing to experiment with.

And then the other thing I'd say is a safeguard is just think about how you want this thing to interact with the world. And that's why I suggest internal examples first because nobody's quite sure how these custom GPTs will behave and will they represent your brand the way you want them to. So, be thoughtful about that.

I wouldn't rule them out but for experiments externally and for internal, and consider the whole thing an experiment. But, I would absolutely encourage associations to jump on this and try it out. It's a fantastic capability. It really is a force multiplier. And I think it also will bring more, Creativity to your mind in terms of what's possible with AI if you go and do this experiment.

Mallory Mejias: If you do create a custom GPT, please let us know how it went, what you think on the Sidecar Community. We have a sidecar sync space [00:26:00] there just for comments like that. This next topic we are diving into today, I can sum up as legacy projects versus AI projects, and I'll explain more on that.

Amith, last week, you briefly highlighted a critical decision facing many association leaders. The choice between continuing with costly IT upgrades like AMS systems or pivoting to explore the rapidly advancing field of artificial intelligence. While traditional system upgrades demand significant investment and time, often yielding only incremental improvements, AI is doubling in capability every six months.

This rapid growth suggests that by the time a typical IT project is completed, AI could offer far more transformative solutions at a fraction of the cost. You argue for a strategic pause on legacy upgrades, advocating for a shift toward AI experimentation.

Amith, can you share your initial thoughts about legacy projects versus AI projects?

Amith Nagarajan: Sure. [00:27:00] Well, you know, a digitalNow, and for literally months preceding digitalNow, when I've spoken with a large number of association executives, particularly CEOs, one of the objections I've heard in terms of why don't you get out there and start experimenting now, which is always what I'm, you know, talking to people about doing, is the answer- I totally get why this is important, but we're super busy. We have a bunch of existing projects and certainly within technology, which AI is partly a technology conversation, of course, but it's, it's much bigger than that. But people often refer to, well, my team is fully consumed by this big AMS upgrade that we're planning or we're fully consumed by updating our website or whatever it is, there's always some big project going on at an association.

Typically, that's at least partly a technology endeavor. And so my thought process behind this is simply to say not that those projects are not important or that AI essentially undermines the need for those projects, although that's possible. That isn't what I'm [00:28:00] saying. What I'm saying is that many people are telling me they can't pursue AI right now because they have to finish these legacy style projects, so they have to complete their AMS Upgrade and then they'll think about AI. They have to complete their website and then they'll think about AI. And on and on. And I think that's a strategic mistake. I think that if you have to choose between them, I would recommend that you pause these other projects, at least temporarily, temporarily.

And you explore a I really deeply and do it now, because to what Mallory pointed out earlier, AI is on a doubling process, meaning it's every six months. The capabilities of AI are literally doubling relative to cost. And so an AMS upgrade takes at least a year. Usually it's a year and a half to two years.

And in that time, you've spent a lot of money. More importantly, you spent a lot of your energy and a lot of time has elapsed. And so there will be three, possibly four doublings of AI capability while you're working on your AMS upgrade. And if you have not devoted [00:29:00] any resources of significance to AI in that timeframe, I mean, that's a massive, massive disadvantage to try to overcome compared to everyone else that's out there in the world, not about associations, but in the world that's thinking about this stuff.

So, I think that associations owe it to themselves to make sure they have the resources available to deeply explore AI. As we go into the new year, and if that means pausing legacy projects, that's what should happen. There's only so many resources, and I certainly get that, and I think that's where the most difficult strategic choice, Mallory, in my experience, isn't what projects you're gonna go do, but it's what you're going to say no to, and what, which projects you're going to remove from the list that seem critically important, but you can't do them all.

You know, if I tell you, you have five critical projects, you can only get three done. Do you just kind of keep going with all five on the list? And just let fate determine which of those five maybe get done? Or do you make the hard choice up front and eliminate two of [00:30:00] them, defer two of them, pause two of them, and focus aggressively on the three that you actually determine are the most strategically important projects?

So to my point an AMS upgrade, a website upgrade, an LMS upgrade. None of those are going to transform your organization. They're going to incrementally improve your organization. In comparison, AI could either be the tidal wave that catapults you forward in lots of cool ways, or it could be the tidal wave that washes over you.

In not a gentle way. So that's why I'm being so aggressive about the way I'm pursuing this is I think we need to have a pause on projects that are preventing us from pursuing AI experimentation and certainly AI learning like if nothing else, an association has to devote the time and a few dollars to train everyone aggressively on AI capabilities so that those gears start spinning in all of their team members minds.

Mallory Mejias: Legacy systems might not transform your association like AI potentially could, but they're often [00:31:00] critical for day to day operations are you suggesting that association leaders step back from this idea, leave it all together, pursue both, or are you saying leave the legacy systems to the side for now and put all your energy and resources into AI, even though there's no guaranteed benefit?

Amith Nagarajan: What I'm saying is you're contemplating going from a car to a slightly faster car or a more reliable car. Or a car with more seats and more capacity, but it's a step change as opposed to an order of magnitude change. And so, rather than upgrading from the same mode of transportation, essentially, to continue that analogy you, instead of upgrading to the same mode, but a slightly better version, you might have something completely different that is radically better, like flying in a jet, or flying on a rocket.

That's capable of doing something completely different than what you're currently doing. And you might not even see it because you're so focused on what you're doing day to day. So let me be very clear about what I'm [00:32:00] suggesting. If you are in the midst of a legacy system upgrade and you are dealing with a mission critical problem where, let's say for example, you have a corrupt database and you cannot serve your members because your legacy systems are literally leaking data or they're incorrectly processing data and you have this like, you know, code red mission critical problem, by all means, finish that project and upgrade or patch that old system and get it to the point where it can operate.

What I'm assuming in my statement is that the upgrade or the lack of the upgrade isn't going to kill you. It might not be ideal to not upgrade your AMS or not upgrade your LMS, meaning the old system's creaking along, it's not great. It's like driving a 15 year old car rather than a brand new car.

It's not ideal, but it still kind of works. I'm arguing for you to change the oil inflate the tires, you know, make sure that the brakes have been checked and keep driving that thing for a little bit longer. So that you can then see if cars are even needed in the future we're entering into. That's the point.

So I'm not suggesting that you drive [00:33:00] around unsafe. I'm suggesting that you, you do get to safety if you have an unsafe ride that you're currently in with whatever these systems are. But to consider the future as a higher priority than most do because most associations, there's so much inertia around the idea that these projects control your destiny.

You have to do this AMS upgrade or this other upgrade that people just kind of, they, they assume they can't stop these things or pause them.

Mallory Mejias: Are you suggesting that some of these legacy systems might not be important in the future or necessary?

Amith Nagarajan: I am suggesting that’s possible, but even if they are important in the future, I don't think that they will be more important than what could change. In the world of AI. Meaning that the like, think about it from an external lens versus an internal lens. So the people outside of your organization, will they care if you have a new AMS.

A really shiny one, right? That's faster and better. And your staff likes it, which is uncommon. But let's say that actually happens. The short answer to that question is probably not because it [00:34:00] won't really affect them and you might even say, but our new website is going to be way easier to use.

They'll be able to register for meetings so much more fluidly. They'll be able to renew their membership so much more easily. They'll be able to see their member profile so much more easily. Yeah. And even if I grant you that and assume that's actually going to happen the way you want it to, which I'd love for it to happen that way that doesn't mean that your business model is any better.

And what AI affords the world is a better way to engage. It's a different way to consume content. It's a different way to educate. It's a different way to learn. And it's a different way to get everything done. And so... In a time where change is happening this rapidly it's not so much that these systems won't be better if you upgrade them.

It's a question of what systems do you need in order to support the business processes that are aligned with the new strategy, whatever that strategy is. And I don't know if those new strategies will be aligned with current AMSs and LMSs or not. I don't have a crystal ball to look through to say this is what all associations or even some associations are gonna do.

I have suspicions of what's going to [00:35:00] happen. But my point essentially is, is that in a doubling scenario like we have, things are changing so rapidly that these massive investments of energy and dollars into systems that are incremental improvements. On legacy processes are worth pausing in order to take a deep breath and look around the world and see what's happening, try to predict that future a little bit and then determine what your priorities are with that insight.

That's all I'm arguing for. And you might very well come back to and say, yeah, we're really glad we paused this because we have a much better understanding of what might happen in 12 months. We still do need that AMS upgrade. Cool. Let's knock that out. Let's go get that done. Or you might find out that actually, you know what, maybe I don't need an AMS at all.

Maybe I need a CRM and I need an e-commerce system or something else, right? Because we just don't know until you think about what the external demand is going to be on you. People outside of your organization, how do they want to engage with you and your content, your services? What types of services will they demand from you going forward?

We know [00:36:00] this, they're not going to be the same as what you've been doing. So let's figure out at least a little bit about what's going to be demanded and where the opportunities are and make sure our systems are capable of handling that. And the last quick point I'll make, super fast, is that software development is a digital service.

And when you're on a doubling curve like we're at, and when reasoning is becoming something that, you know, which is what AI is able to do, is to do reasoning, when you're able to do that at scale for lower and lower costs at higher and higher speeds, It means that software development is essentially trending towards zero marginal cost, which basically means building software is becoming easier and easier.

Therefore, building solutions for your association's needs will be more accessible to you if you do need to build something custom you need to hire a vendor to do it. And I believe that there will be a bunch of products coming to this market that are specific to associations that are aligned with these new business models I also think I’m optimistic that the AMS and LMS and other vendor communities [00:37:00] will embrace these technologies and bring new capabilities to market.

And so if you've just upgraded to the latest version, but there's a new AI native version that comes out, you know, wouldn't you rather have waited for that and put your effort into getting to that version? And those things aren't there yet, but I expect that they will be in the next year or two.

Mallory Mejias: I imagine what's especially challenging is leaving the certain for the uncertain. So leaving the website upgrade where you have this certainty that you're going to make improvements and your members should like that, to the uncertainty of this world of AI, which I think for a lot of people could feel like a gamble.

Like you're stopping all these processes that you're working on and you're taking a gamble on something that you're not sure about, but that you think could be really beneficial for your organization in the future. What do you think about that? About AI being a gamble?

Amith Nagarajan: I think it's a fair question to ask yourself. And certainly there is a degree of certainty that comes from An incremental improvement to that which you already know. So, you know, going back to the automobile example, if I know that I have, you know, a [00:38:00] two seat vehicle and I'm upgrading to a four seat vehicle, I pretty much know what I'm gonna get.

There's a pretty good idea, like I go to a higher mileage vehicle, I know what I'm gonna get, I understand that. So I totally understand that, I empathize with it, and I think associations should make some investments in those incremental improvements. The thing that I am fighting against and pushing for with this argument is the generalized idea that we can't do AI stuff in any meaningful way because we're so busy.

So when I hear an association exec tell me I can't deal with this in 2024 or in the next six months because I have these other projects, that's when I come back to, okay, if that's true, then let's look at your portfolio and see if there's even one or two things we can pause. Maybe not pause everything, but selectively pause something or some things for, say, six months and use that time to go and deeply explore AI to determine what makes sense. So I actually think the downside risk from this is you're six months behind. And let's just say let's just say is a complete bust [00:39:00] and that it's all garbage. And none of this is ever going to materialize in any business value.

I think we're way past the point where that's a possibility. But let's just hypothesize that's a possible outcome. And the six month delay that I'm recommending to some of these folks Is a total waste of time. The worst case scenario is that outcome, and you've spent a few dollars and spent some time exploring this, right?

But six month delays happen all the time in these projects. And it doesn't kill the association. The downside risk is manageable. I'm not saying wait two years to upgrade stuff. I'm just saying take a brief pause and take a deep breath and look at the world around you a little bit more thoughtfully than when you're hurried and when you're overwhelmed because you don't have enough bandwidth to do anything.

That's all I'm arguing for.

Mallory Mejias: Can you talk more about the concept of future proofing and association and what that means?

Amith Nagarajan: Sure, as a generally statement it’s a bit of a pie in the sky thing, right, where everyone wants to future proof when it comes to software, people are thinking about how to make systems more resilient to [00:40:00] change, how to make them more upgradable when it comes to business processes, lot of times people are talking about, you know, what can we anticipate will likely change? And how can I make the business process a little more adaptable? You know, a lot of what I talk about when I'm referring to future proofing and associations is data centric. So I'm a big advocate of associations owning their own destiny when it comes to data.

And that's an AI conversation for sure, because data is necessary to make AI work at any significant level. But it's also just more of like an ownership of your future type question. I've seen for so long associations struggle in deep pain from not having control over their data and so not having their data in their own hands.

It's not being able to manipulate it, how they want to use whichever tools they choose. And now, of course, being able to use AI against their full data set, it's a real problem. And so when you are in an ecosystem where [00:41:00] your data is highly distributed in lots of different apps, you haven't connected it together.

And also if you're in an environment where You really beholden to the rules of those vendors, you know, yes, technically, as per the legal contract, the data belongs to you, but as a practical matter, it's largely inaccessible. It's in these different ecosystems that you don't have direct control over. I view that as a problem.

That's definitely not future proof. To me, that's the opposite of it. It's super brittle. If anything breaks, you know, it breaks everything. So those are environments that I urge people to find ways away from. I think that one of the smartest things people can do right now is, first of all, inventory all of the data that they have.

So, system by system, you don't need to do every one of them because there's probably dozens of them, but go through your systems in terms of mission criticality. And inventory them, figure out what you have. What types of data do you have in each system? Just do that. And don't do this as like a six month, you know, painfully accurate process.

It's not about the accuracy. It's more about the order of magnitude. Just [00:42:00] get a sense of what's in each of these systems. And then think about this question. Which of these systems, if it went away, would kill your business? And which of these systems, if it went away, would not? The ones that would kill your business, make sure you get your data out into another place.

And there's a whole conversation we can have about common data platforms and data repositories. But the point about it is, you have no resiliency and you have no future proofing if those truly mission critical data sources are things that you don't have control over ultimately.

Mallory Mejias: Can you give a brief elevator pitch of a common data platform?

Amith Nagarajan: Sure. Well, a common data platform is essentially a database that you control. And the idea is to pull in your data from all your other data sources into that common data platform. On a continual basis, not just a once and done kind of thing, but on a continuous basis and then to build your analytics, your inquiry, your exploration. Build all of your functionality for that type of business [00:43:00] activity on top of the common data platform rather than intertwining those other solutions on top of your line of business apps.

And then, of course, your AI strategy gets built on top of the CDP as well. So it provides you a mechanism for resilience. But it also really is the launch pad for all of your AI initiatives. There's a lot of ways to do this. If you're a bigger association with a lot of resources and a lot of  IT.

Talent, you can certainly build one of these yourself. And there are a number of open source solutions for building common data platforms. One that we helped launch recently is called MemberJunction, which is a free open source Common data platform that was designed with the Association of Nonprofit Market in mind.

But that's just one of many options you can use for building a common data platform. It's more of a concept than anything else.

Mallory Mejias: Amith, you posted this maybe controversial argument about legacy projects versus AI projects on LinkedIn, and we got some really good comments there. I encourage all of you to check out that post as well. We'll link it in the show notes. We received a comment from an [00:44:00] association leader who found that many AI educational sessions they've attended are pretty basic, focusing on things like creating new images or using chat GPT for simple tasks. These sessions haven't really demonstrated how AI can be a practical alternative to investing in traditional systems like AMS or LMS. What do you think of that Amith?

Amith Nagarajan: I think I think that association leader was spot on. I mean, that's what most of the conversation is. So on the one hand, I'm really happy that the conversation has gone mainstream and that most people I talked to in the association market are thinking about this stuff at that level. But that particular individual's point was really solid in that.

You know, how do you take that beyond the seemingly trivial examples that are cool? But like, how do you actually apply it is what I think he was really asking. And so from an educational perspective there's a number of resources that we'll link to in the show notes, but I'll just rattle off a few of them here.

One is that we made available a few months ago, this book called Ascend, Unlocking the Power of AI for [00:45:00] Associations. And this book walks you through a complete transformative process for an association. With artificial intelligence, and it hypothesizes a number of transformative changes that can occur in really embracing AI in a in a much more meaningful way than simple basic examples like the association exact was referring to.

So the Ascend book is available at sidecarglobal.com/ai. Again, the book is free to download, and we will continue to update that book roughly every six months because that's when AI doublings tend to happen. There's lots of new material that will be adding to the book towards the end of the year for the next release in early Q1.

Also sidecar has been conducting boot camps on AI since the April time frame. So for the last 6-7 months, we've put hundreds of association people through these AI boot camps. And they are very much hands on practical applications of AI. And what's been really exciting about these boot camps is they've been interactive.

Lots of Q. A. We have expert AI [00:46:00] instructors who are also deeply knowledgeable in the association domain. That's the key combination. It's not just a I knowledge, but it's also understand the domain. And in these boot camps, People have learned all the fundamentals of these different tools, but then learned how to apply them to business problems they actually have.

So we've learned a lot in that 67 months since we started putting hundreds of association folks through these boot camps. We've gotten great reviews from them, but there's been a couple challenges actually with the boot camp format that we've run. It's episodic, meaning we have people go through a series of four boot camp sessions of 90 minute each.

They're great, but then they're over. And AI is changing really rapidly. So even if you were the ultimate student and you were super tuned into those four 90 minute sessions, you know, back in April or back in June or back in September, you've forgotten a lot of what you've learned. And AI has changed a lot since then. And so we've decided to experiment with a different format of the boot camp, which I'm really pumped about. This new boot camp format is going to [00:47:00] blend together the best of these lessons, these instructor, expert instructor led lessons on different AI topics ranging from prompt design to, you know, detailed applications of how to actually put this stuff to use in your association. And they're gonna be available as recordings on the Sidecar community in the boot camp course.

So you can take them at any time, but they're gonna be updated constantly. We're gonna be adding new lessons all the time as new topics arise and updating existing ones to make sure that they stay fresh. And then the people that are joining these boot camps, I should say the boot camp will be able to take the lessons on demand.

And then once per week, we will have one of our experts online at a specific time each week in an office hours to answer any questions that come up along with a continuous loop with an online forum where people can ask questions and get answers not only from the experts who are hosting the course, but from the community of hundreds of people who are going through the experience together, and it's a 12 [00:48:00] month subscription. So what's super cool about this is that it's a journey rather than just one particular piece of it. And it's a resource that folks can use to both rapidly get up to speed, but then also stay ahead of the curve in terms of what's happening.

And have a community of people to work with. Not just the AI experts, but the entire community. And that's where I get really excited. You know, is, is the idea that people can learn together on this journey and share what they're learning.

Mallory Mejias: It's so exciting. I posted on my own LinkedIn page, AI is changing every day. Not just that, AI is changing every hour. And if you don't have a way to stay up to date with these changes, you'll fall behind really quickly. And not only that, but a way to stay up to date with these changes in the context of the organizations that you work for.

Associations, the greater non profit market. So I think this hybrid bootcamp is really exciting. We're planning to launch this bootcamp next week and we'll be offering a special promotion for the first few members to join.

So be on the lookout for that. Podcast [00:49:00] listeners will be the first to know. Also Amith, we mentioned the book Ascend. We mentioned the hybrid bootcamps. Can you touch briefly too on the CEO mastermind group?

Amith Nagarajan: Sure, and that's another opportunity that's available to help association leaders in this journey. I formed a group along with a co facilitator, Mary Byers who's also been in the association community for a long time. And she and I put together this CEO centric group a couple of months ago. And the reason we did this is we felt this is such a transformative period for associations that we need to get buy in from the top staff leaders in these organizations.

And so we've got about 30 people in the CEO boot camp right now. We're enrolling for 2024 and the boot camp essentially consists of a monthly meeting. With the CEO. The CEO is able to bring up to two of their key staff with them.

We meet 12 times a year. In addition to the actual mastermind meeting, which is a mix of some directed education where somebody will speak on a [00:50:00] topic that's really relevant on. Then there will also be pure sharing. So members of the mastermind will share projects they're working on and get feedback from the rest of the group.

In addition to that, though, we'll have in between each of those monthly meetings an office hours with me where people can bring any questions and I'll be online for an hour and answering any of those questions. And we'll also have an online community available just for the participants in the CEO mastermind.

So we'll link to that in the show notes as well, I'm sure. But really excited about getting CEO level engagement. The reason that's so important here is because the CEO does not need to be the expert in AI, but the CEO needs to be an expert in what AI is capable of. And that's an important distinction.

You're not the technician that needs to know how to operate the thing, but you need to know it exists and you need to know what it's capable of so that you can devise the best path forward for your organization. And that's what this mastermind is all about that. We call it the CEO mastermind on AI. But AI is really more about understanding what the tool can do.

Mallory Mejias: You heard it [00:51:00] here. We've got the Ascend book on AI for associations. We've got the hybrid bootcamps coming up really soon. We've got the CEO mastermind group. Sidecar is excited to provide AI education for association professionals at every level of their career. Amith, thank you so much for sharing your thoughts today.

Amith Nagarajan: Thanks Mallory.

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 ascendbook.org. It's packed with insights to power your association's journey with AI. And remember, Sidecar is here with more resources for 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.