AI for Nonprofits: Building a Scalable Thinking System

Transcript

Mohamed Hamad: Good morning, everybody. How’s everybody doing today?

Mitch Schwartz: Good morning.

Mohamed: Yes, got everybody in the room here. Happy to see everybody today. So, we’ve got session number two in our workshops in collaboration with the CHSSN. Today we’re going to be talking about. AI for nonprofits, and in this session, we’re going to be talking about how to build a scalable thinking system. And what we mean by that is how to set a foundation for your AIs so that it can learn about who your organization is, what you guys do, who you do it for. And to have a set a set of rules and guidelines for consistency, and basically to make sure it represents you without creating hallucinations or going off book or anything like that. Before we get started, my name is Mohamed Hamad. I am the founder of Third Wunder. Been working with the CHSSN for a few years now and. We’re a digital marketing agency that does web branding, design, and AI consulting. And with me today, we’ve got Mitch. Mitch, introductions.

Mitch: Yeah, hey, I’m Mitch. I’m the founder of OpsMachine, and we help teams work well with tech and figure out what can an AI workspace look like, how can we really enjoy the work that we do, automate away the mundane, and just really get. In that flow, in in a way that feels like aligned, ethical, and that people get excited about. I like to say that if you have to do change management, then something’s already gone wrong. And I’ll stop right away because that upsets some people. There’s a lot of great practice in change management, so I’m I’m not throwing the whole thing out. But what I am saying is, if people aren’t going, yeah, let’s do this thing. Then something has gone wrong. So rather than driving the change, how do we get people pulling it and and really wanting it? And that’s a lot of the focus of the work, up until the building part.

Mohamed: Yeah, amazing. So, this is the second session in a three-part series in collaboration with the CHSSN. In our last session, in September, we talked about marketing for nonprofits and how to set up. A system to create consistent voice tone per channel for your goals and how to establish marketing goals and all of that. Piggybacking on that, we’re going to talk today about how to take some of that information, setting that ground rules for your organization, and how to leverage it with AI so that, one, your team all has the foundation to create consistency and to create a a cohesive voice. But then how do you use that that information and those ground rules with AI and to also establish a foundation for your AI? Our next session in the three-part series is going to cover the CHSSN map and it’s a more advanced workshop which is in November slated for November 19th around the same time at 11. So without further ado So where do nonprofits sit in the AI world right now? From research that we’ve seen from Imagine Canada and the say the Canadian Center for Nonprofits Digital Resilience, we’ve we you know we they found that 80 percent of Canadian nonprofits use AI, which is astounding. And about 10 percent of them have a formal policy. Right, and 64% of them have none and aren’t building one, which is an alarming number. Mind you, these are from these are stats from last year and earlier this year. Now, when we what we what do we mean by 80% are using AI? This is a broad range, so this counts things things like Grammarly, for instance, for spell checking, which because that uses an AI, all the way up to Advanced instances of Claude and ChatGPT in the workforce. I see some people are having sound issues in the chat here. I okay. So people are trying to figure that out.

Mitch: Looks like just one person so far.

Mohamed: Yeah. Okay. Let me know if you’re having any issues with sound or. The stream itself in the chat. Yeah, as as we’re figuring that one out. Yeah, so what does that mean? That just means that a lot of people are trying to figure out how to use it, and what we’re trying to establish today is a foundation for everybody to create a a baseline for how to best use AI. Today we’re going to be talking about how to use an AI not necessarily just as a tool or as a technology, but to really think of it as a team member. Because you know Mitch loves to say this: nobody wants more technology, and I kind of agree with him. We love the idea of more technology, but we don’t really want more technology.

Mitch: And this I love to define that because like something that works. All the time isn’t something we call technology. So like a light bulb or a power outlet or like an old school TV, that’s not technology. Technology is the name we have for things that work like some of the time, but not always. And so dealing with all of that is not the goal at all.

Mohamed: Yeah, and in this respect, we have a technology that that almost feels like a team member if you set it up right, you know. We talk to it, we set it up, we chat with it, and it can feel like you’re talking to someone else. So, if you are going to be thinking about it as a team member, with like with any team member, you need to onboard a new employee, a new coworker, a new team member, and how does that work? Right? You usually have a handbook of some sort. You define their Job role, you give them the history of the organization, the mission, the values, all of that, and you give them access to certain tools depending on what they need to do. So, you know, if they’re marketing, you give them some of the marketing tools. If they’re in HR, you give them the HR tools. So you give them the tools they need, the description of what they should be doing, rules and regulations on what to do and what not to do. And and then you have a feedback process. So in their first three months of onboarding, and then you kind of do a a one-on-one. And it’s almost the same thing with AI in a certain sense.

Mitch: Yeah. And if you had a new teammate, let’s say you brought in a junior bookkeeper, you would probably give them maybe read-only access to the bookkeeping software. You wouldn’t. Probably give them the ability to send money from your bank account and then just the instruction, you know, don’t don’t do that, and then hope for the best. And it’s the same with AI. So really considering what actual keys would we give a person, and a lot of this maps over. So people come in and talk to us and say, “Hey, you know, I’m a little lost. How do I think about this?” And so this one mental model, Don’t anthropomorphize the AI, Yeah, but use the mental model of how would I treat a teammate. And of course, it’s a really important line.

Mohamed: Yeah, and once once you get someone onboarded, you know, you work with them slowly. You test things out, you get their ideas, you go back and forth with them, and that’s what we’re trying to establish today. So, how does this work, right? We like to think that context plus skills plus tools plus policy equals a scalable thinking system. So, there’s a lot of pluses there, and we’re going to go through one by one in this this idea of like if you’re building a building, you’ve got the ground floor, the next level, the next level, and the next level. The foundational piece here is the context. Right, and this applies to pretty much any AI agent, or AI platform. It doesn’t matter if it’s ChatGPT, if it’s Gemini, Copilot, Claude. You need to let that platform know who you are and what you do and and how you conduct your business. But before we get started, I want to throw a poll out there. I want to get a sense of you know which platforms people are using, and I’m just going to throw out this poll. So which AI platform do you have or use?

Mitch: Now while you’re all answering that, what we will say is there’s a lot of information in this presentation. This is Maybe 30 to 40 percent of a training that we give, and so there’s a lot you may not catch it all, and I don’t want people to be getting overwhelmed. So what I’d say is focus on the headlines and the high level stuff. We’re going to send out a copy of this to you later, and you can even pop that into your AI system as well and and ask it questions about and and interact with the document later on. So, if you find you’re struggling with something. You can always ask about it and we’ll be happy to answer. Otherwise, and and I know this is maybe easy for my personality type to say and harder for some people to follow, but really just try and follow the big picture and the headlines and and the logic of it rather than the granular stuff at this point.

Mohamed: Absolutely, yeah, we’re going to be sending out in the recap email a summary of this session. Along with the recorded video and this presentation as well, so yeah, have a look at the the the high level picture, get under get an understanding of the concept in itself. All right, so we’ve got about 19 votes in here with 74% of you on ChatGPT, with some of you on Claude and a smaller set on Copilot. Excellent. That is good to hear. Now, so talking about context and what do we mean by context? So when we’re talking about context, it’s basically the foundational documentation about your organization. This really is about who you are as an organization. What are your values? What is your mission? What is it that makes you tick? Who is it that you’re serving? The community that you’re serving, the services that you provide, and who is it that you’re specifically targeting with with your with your services? You know, information about the languages you provide services in, for instance, if it’s English only, French and English, French only, you maybe you provide services in multiple languages. That also helps an AI allow, you know, help you translate and localize some of the content or some of the output that it has. In the last session we talked about creating a brand voice and tone document. And we sent out a guideline or a template there. That in itself also helps an AI be able to. Speak in your voice. Make sure that it represents the organization in the way you want, in the specific channels, right? So if you’re asking it to, let’s say, write a Facebook post versus a formal report, it knows how to adjust its language based on your rules and guidelines.

Mitch: Yeah. Now, I think it’s it’s really important to understand that having. This critical context in the AI is going to be the difference between getting generic AI slop and getting company specific AI slop. And I say that because it is only the first floor, and so there’s more work to do to get really great results, but specificity and relevance is going to be the first step.

Mohamed: Usually, these types of documents are already available because if you have an onboarding handbook. If you have a company policy or guidelines, you can extract those and create those documentation. So it’s not something that you would just create from scratch and it’s you know it’s a it’s a head scratcher there. A lot of this information is already available within the organization, and in a lot of cases, you can actually take all of the the baseline company information that you have, and give it to an AI and say, “Help me create these foundational documents that will help you.” Understand who our organization is. Now, all of this information usually gets stored somewhere, and we want to store this information in a knowledge base. A lot of organizations either using Microsoft Suite or Google Workspace. You can store that information in, SharePoint or OneDrive or Google Drive or whatever it is that you guys are using, and that creates a shared space with the entire organization. All right. So, if you have a space where you can just create a folder where it’s just all of these these text files or these documents that define all of this information. You can point any of your AIs, whether it’s ChatGPT, through its connector, and we’ll talk about connectors and all of that later. And it will one when you’re talking to it, it’ll go and source that information, read it before it creates any output for you, right? There are two things when it comes to AI and what we call memory. Right, and you’ll you’ll see that you know sometimes in ChatGPT or Claude or any of these that will remember things about you specifically. The difference between memory and a knowledge base is that the memory in a session or a an account remembers things about you specifically as an individual engaging with it. Right, so it’s not organizational memory. So let’s say person A has a memory with the That stuff, that that memory and the that personalization doesn’t go into someone else’s or person B’s memory. Each one is individual, so it’ll it’ll start to remember things about you, but information or instructions about your organization, that would be something that would be in a knowledge base.

Mitch: Yeah, now the memory can often be automated, and that means that the AI is going to remember things that seem relevant to it. And so, you can go in and clean that up. It’s recommended to check it out every couple of months. Sometimes I’ve caught it just storing things that are completely irrelevant or so niche to that situation that they don’t matter. And at the same time, I could be getting great performance based on something that’s in my memory, but isn’t actually shared to the rest of the team. So it’s worth being cognizant of that. Now. There is a question from Anne. (Question from Anne): If our organization doesn’t have these documents formalized yet, should we be implementing them with humans before giving them to a bot?

Mohamed: I think that’s a great question. I think it is something that internally the team should get together and really discuss the type of information that you want to have exposed to an AI. One of the things is. Also around privacy, and this we’ll get to in the policy section of this conversation, which is what to share with an AI and what not to share. But for now, this is sort of like creating a foundational document. And usually, this you already have that information. It’s now just putting pen to paper and putting it down. Yeah. One thing I just mentioned.

Mitch: Oh, do you want to go ahead?

Mohamed: I was just going to say that we’re going to be please put all your comments and your questions in the chat. We’re going to have a Q and A session at the end of at the end of the presentation to field all the questions. But yeah keep them rolling. We’ll be we’ll be addressing them all at the end.

Mitch: Yeah okay so I was going to add in now hold on do I still remember what I was going to say yes so if you have these documents and They’re newer to your org. Having people read them can be great. Putting them into an AI and then seeing what comes out can be a great way to test it too. And that gives you a different lens on it and a nice way to iterate quickly. So you can sense misalignment between what your words are intending to communicate and what at least is being understood by the machines. So it’s worth doing. But in terms of Setting up formally those documents for everyone’s AI to run on, I would work in phases on that, and and essentially, you know, the framework is to set up gates where you say, “Okay, we trust this enough to use it for you know a bigger scope or or larger blast radius.”

Mohamed: Absolutely, and there’s also documentation that is specific to a role. Right. So, for instance, there are there’s information about an organization that applies to everybody, and then there is information that is specific to a role within the organization. So, for instance, there’s you know things related to finance that are very specific to finance that you wouldn’t give like a marketing or communications person, or a administrator wouldn’t have the same information or handbook or guides for let’s say community outreach or fundraising. For instance, so part of the conversation about what documentation to have is, you know, there’s stuff that’s generic and general to the entire org, but then each team would say, okay, what are the things that are relevant to us that maybe is irrelevant to everybody else? And we don’t really need to, they don’t, they don’t need that. And let’s create that document specifically, but we’re not going to put it in the general pool of information for the the bigger AI. So. And that will also come into the whole policy side of things and how how to define those in general. So that’s the ground floor. That is the foundation there, the knowledge base, the information about your organization, who you are, what you do, who are you doing it for, the the roles themselves, and where to store that. Moving on to the next floor is skills. Now I’m going to throw in a poll again, and I want to find out from you guys how many of you know what skills are or have created skills for your AI. Let’s throw that out there. Okay, quite a few people have created some skills. This is this is amazing.

Mitch: I was gonna say, I guess ChatGPT must have implemented them more broadly now because we were like 80% ChatGPT users in here.

Mohamed: Okay, so we’ve got we’ve got almost a 50-50 split on people creating skills and those who don’t know what skills are or have not created them. Yeah.

Mitch: So I mean, I’ll jump in and say that skills were probably the biggest, like, exponential gain for me in the way I was working with AI in the last year. And you know, as the slide says, we’ll see in a second, they’re essentially SOPs. And what they allowed us to do, especially in Claude, was say, “Hey, this is what I want you to do.” And then while in the chat say, “Oh, that part sucked. Can we fix that? Can you do it differently next time?” And it didn’t require going to a different screen and manually typing it, but Claude could update it on its own. And so that feedback loop became all of a sudden so much more effective. Whereas I was always having this cognitive resistance to, “I don’t want to stop my task to work on the.” The meta of it, now it just became inherent and really easy.

Mohamed: Yeah, skills are basically SOPs. You know, in an organization, we create an SOP, a a set of instructions when to use it, what to do, how to know that you’ve done it right, and you know it’s it’s a set of procedures, right? You know, you would create a set of procedures for any of your team members when it comes to. Figuring out something or addressing something or escalations or anything like that, it’s the same thing, but it is for an AI, right? And it really is in a specific format. It’s it’s a text-based document. It has a title. It’s got a description of when to use it and where, and then a series of instructions of this is how you’re supposed to do something. This is how you format it. This is the language you use. And maybe sometimes there’s even reference files. So for instance, you know, before you do this, read this guide or you know, set yourself up. So it’s fair. It’s almost like training another person in something very particular, right?

Mitch: Yeah. Now you can actually have the AI write a skill for you, which at this point I recommend. Claude comes with a skill writing skill, so that it just knows how to do that well, and. Writing one or two becomes really easy and straightforward, and iterating on them is as simple as saying, you know, I want more of this, less of that. This part didn’t work. What starts to become maybe a little bit more of a challenge is once there are many skills managed across multiple people. So it’s a really easy thing to start playing with, and I do encourage everyone here to experiment with it. It can be a really great replacement for projects, where often we’ve used projects or custom GPTs in, you know, to capture certain tasks and procedures. But then when let’s say we’re doing some writing and half of what we’ve taught that project applies in a different context, now what do we do? Skills became the answer to that. We can just pull them from anywhere, use them in any context. Yeah.

Mohamed: So skills could be anything from anything that ends up with a structured output or a a format that is repeatable in a sense. So for instance, if you need your board minutes to be in a very particular format to capture very particular pieces of information and in a specific way, or for instance, if you’re translating something and you want to make sure that the AI goes to the vitrine linguistique, the OQLFs. what you call it guidelines on Quebec translation before it translates anything so it makes sure that it’s in Quebec French. If you want to make sure that your social posts if you’re creating them or your blog posts are in a particular format and use a specific style you skills are there to make sure that it has the checklist it needs and you know it’s it’s guidelines of how to to create any of its output. Right. So skills where we’ve got the context, which is the information about the organization. Skills are the the the guidelines and the instructions of how to do a specific thing. Next up we’re going to be talking about tools. And tools are one of my favorite parts of. AI and this is a one of the more recent developments. You know it used to be you just chat, you know you would chat to a ChatGPT or Claude and you know maybe it’ll search on the internet. And then they created connectors. I think ChatGPT calls them plugins. But basically it’s how do you get the AI to look at some of your tools. Right. So for instance You know, if you’re running a campaign on Mailchimp, you can connect Claude or ChatGPT to Mailchimp, and you can ask a questions about the latest campaign without having to go into, let’s say, log into Mailchimp, go to the right campaign, go to the report section, copy paste what you see on the screen and put it in there. You can just say, “Hey.” Connect to Mailchimp and tell me how well the last email campaign that we sent out for this subject did, and it’ll go in and it’ll just do that for you. So it it shaves off friction from having to do all of the little tiny steps to get to the information, because now you got an AI that knows your organization because of the context that we created in the ground floor, and in In a sense, we can also create a skill of how to report on, let’s say, campaigns. So it’ll take the skill, it’ll connect to the system, and it’ll pull out the data, and it’ll create a report in the way that you want it to because you created a skill around it. So those things interconnected are game changers in a sense where it just removes a lot of the friction because now you’re not copy pasting or exporting a CSV or. Exporting a document of some sort and then putting it back into it. Yeah.

Mitch: Yeah, I feel like this is where I start to sound like an infomercial, but this is that thing that makes you work 10 times faster. And so, I mean, Mo, you’re giving the example of looking at one campaign, but then, hey, let me look at the past 10 campaigns and which ones actually had the best open rates or click-through rates. And doing analytics crunching, put that into a spreadsheet for me so that I can have a conversation with my team lead on it or whichever other person. So being able to compile that data is is so meaningful and so available to us now. So a lot of what is a little bit mundane and and kind of the grind of our job can really be automated now.

Mohamed: Yeah, I just threw out a poll here just to see how many of you have actually created a connector or connected your AI to any of your systems. Like Mitch said, it could also look at, let’s say, historical data. Now, the magic happens when you have multiple systems that are related to, for instance, let’s say, marketing and communication. So you’ve got your social media, you’ve got your Google Analytics, you’ve got Mailchimp for email. And you want to find out how well all of these communications that you’ve been doing for a campaign. And if you have all of these things connected to your your your platform, it’ll and you just ask it, hey, connect to Mailchimp, Instagram, what do you call it, Meta with Instagram and Facebook, for instance, and our Google Analytics. And look at all of the things that we published within this space, and tell me how well the engagement is or was by channel, right? And it’ll go in and I’ve done these reports. I love doing these reports because it’ll go off and it’ll look at the historicity of a particular campaign in a in a in a time frame, let’s say last quarter or last month, or from the beginning of the campaign, and it’ll pull out the information. From each one and it’ll analyze it and synthesize it based on a skill that you go you’ve you’ve created. So, you can reuse that over and over again to say, every quarter give me this report about everything that we’ve done in in specific marketing campaigns. You can do that for financial reports, for instance, or any types of Campaigns.

Mitch: Yeah, so I just looked it up and historicity is a word. I learned something new just now.

Mohamed: Oh, that’s my favorite word.

Mitch: So I I was working with with a client and you know he’s a solo founder and so pretty much everything falls on him. He has a a website that he was working on and and not you know he’s not the most technical, but. Who does it fall to, right? So he’s working for about four days on trying to make some important updates, and he’s using ChatGPT. He’s copy pasting some stuff from the website into ChatGPT, copy pasting what ChatGPT gives him back to the website, and I said, “That’s great, but it at this point it’s the slow way to work.” So, I whipped something up for him, and he pretty much wrapped up the project in a half an hour with. Claude having a direct connection to the website, lots of safety rails in place, but with the ability for it to see in and then make updates locally and test them and get them to where they needed to be and remove him as the, you know, essentially the gopher, for all of all of the things that needed to be tested and checked, the acceleration can be massive. So. You know, we’re talking about connecting to systems, and then now, I’ll show you a little demo of how I work. So, I have, you know, I have my CRM. It’s running in Notion, but it could be Salesforce. It could be any other CRM. And what I’ve done is connected to Claude with it. And not only that, I’ve created a skill so that it knows, you know, exactly where the database is. How to go about filling it out, each field, what the default is if I don’t say anything, what needs to be filled out in case I forget to mention it, it’ll ask me. So it is effectively like a perfect admin assistant. Now, we’ve had all these tools for so long, CRMs, task management tools, and the biggest failure point is that they’re tedious and they suck to be updating all day and. And so we either don’t or we spend a lot of time in it. But now there’s really a way to eliminate a ton of that. So let me jump into the demo here, and we’re gonna take a look at the demo CRM. We’re actually live in the call right now. So can you tell me what does my week look like? Who should I be following up with? Give me a summary. All right, so you can see that it’s got the demo CRM skill. It’s pulling things up. You know, those of you familiar with Claude have seen this sort of interface. If I wanted to see what it’s thinking and doing, I could open that. Sometimes it’s fun to see its thoughts, but okay, so here are the three deals. The stage for each one, the value, temperature, last contact, next action, and when it’s due. Follow-ups in order. So this is the only action due this week. It’s the hottest deal, furthest along. The proposal is sent. Here’s my second deal. It has the stalest contact, eight days. The follow-up isn’t scheduled. And then here, you know, was contacted two days ago. So I’ve got everything really neatly. Stacked up for me. So now, if I want to say, actually, there’s a couple of updates. So for Harbor Foundation, we won the deal. It’s closed, and then Northern Lights actually was really upset about one of our posts, and they told me where to go. Now I have been a little bit intentionally colloquial with the way I said that last one to make the point here, which is so here are both updates into the CRM. One is won, the other is lost. There’s a good understanding of nuance and ability to translate different ways of of saying the thing, which is really nifty. And so there we go. That’s it. My CRM is updated. Now that’s. You know, one of the ways where I’m doing this directly, what I have running every day is, let me just get rid of that. So what I have running every day is actually, Claude will check my inbox to see what emails have come in. It’ll check my calendar to see what calls I’ve had, and it’ll check the transcripts of my calls. And so what that does each day is it proposes updates to the CRM at the beginning of my day. Based on emails that came in overnight, based on calls I had yesterday, so it’ll remind me actions that I’ve agreed to take and any of my commitments. It updates my task management in the exact same way, and so everything is tracked in a way that’s effectively automatic. I review it first, because of course, if it got something wrong, I want to see that, and I want to make sure that we don’t get misaligned. But this is the kind of stuff that can eat up. Well, easily like a third of our day, all of the admin work, but here it is, just running so beautifully smoothly.

Mohamed: Yeah, and in this example, it’s one of those things where it saves you time for these micro interactions. In a in in in the general way, you would have to log into your CRM, you would have to find the record, you have to go in and edit it, and then go to the right fields and update those fields and write your notes. But in this situation, Mitch just just had a very quick casual conversation and it updated a few records here and there with with the right information. So these types of things are small quality of life improvements, but it’s a small example of how you know you can really shave time and you can really improve the the the the The work experience because you don’t have to do the small mundane stuff at all. You can just trust the system. Now, all of this gets very tricky and hairy unless you have a proper policy of what to do, what to connect, when to connect it, and. How you go about it, right? And this is where the policy comes along, right? Internally, you would have these guidelines anyway for your employees or for your team in general. What not to do and what to do, how to go around a specific situation. So in terms of privacy concerns, obviously we are, we have a few. regulatory concerns there with let’s say Quebec’s law 25 we have CASL compliance for the Canadian anti-spam legislation we have general PIPEDA laws around the country and just generally making sure that you don’t expose personal information or personal personal identifiable information things like social Insurance numbers if you’re working in HR or date of births or health information, sensitive client records for instance, never put passwords or serial numbers or API keys in there. Those things you wouldn’t want to have just publicly anyway, or exposed anyway. So you would have those let’s say you know let’s say serial numbers would be in a password manager of some sort. So What is a policy or a a a foundational policy is the do’s and don’ts, that you would apply to the team. But then also you can also put in it’ll it’ll help you develop the the guardrails for what the AI can do and cannot do, right?

Mitch: Yeah, and it’s really important as always to make sure that you know a guardrail A weak guardrail is an instruction: “Don’t do that.” A strong guardrail is what I’m going to say: the mechanical inability to do it, which means a read-only account doesn’t allow modifying things that are sensitive, right? Having no access doesn’t allow reading. That’s a mechanical stop, and it is so important because, especially in the non the not-for-profit sector. Like a data leak or a privacy issue can really be like a betrayal of the community and and the mission. So, so it’s just so serious and and important to make sure that the systems are built around privacy first. And this is part of the design spec. It’s not just an afterthought and like, oh, hey, do we have privacy? Let’s let’s get someone to look at it. It’s it’s the foundation.

Mohamed: well I guess it’s the roof here but you know when you’re designing things it’s it’s part of that from the beginning absolutely so obviously with any of the AI systems when you do connect your your your platform to any external system there is a list of things that it’ll tell you these are the things that I can do because when an when an AI platform talks to another platform the The third party or the external platform tells it these are the things that I’m allowing you to do, and for each of these things here, you have as as the administrator or the person connecting it, you can say never allow it to do this. Yes, you can do this, and you know ask me if I should run this. So there’s an an allow everything. There’s a do not do this, and there is a Ask me before you ever run this, right? So, for instance, if you were to connect it to QuickBooks as your finance system and your yeah your your finance system, you don’t want it to ever create invoices or send out messages, for instance. But if you wanted to read your profit loss statement or your financial report. You can say, yeah, you always have that, or ask me before you ever touch this, right? This can also apply to segments of your CRM, or anywhere where there is personal information or anything like that. So you can actually define and say, never have access to, let’s say, your contacts, for instance, but you can have access to a company record where the company information in a CRM is generic. It’s just like their website and. And all of that, but the people in your CRM, you have zero access, right? So you have fine-grained control, but that only comes in when you actually internally have the conversation of what you want your AI to do for you, and when to do it, and who has access to this, right? It’s the same thing for an employee, for instance, or a team member. You know, you’re never going to give a junior team member. The ability to go off and send out invoices and make payments from your bank account, for instance, you know, they can have a look at some reports, sure, but you wouldn’t want them, you know, unless they get to a certain level. Right. So, if you have any questions about this, please throw it into the chat. We’re coming towards the end of the presentation. And the webinar, so we’re going to be fielding some of these questions fairly soon. So, how does this all stack up? We’ve got the context; it personalizes the the platform to your organization. You’ve got the skills that gives it instructions on how you want any of its work to be done. You’ve got the tools, which is the systems it can pull in information from, and it avoids too many copy pastes. It reduces friction. It helps you streamline your process, and then the policy itself of how, and when, and why, and who is allowed to do anything, and why. Right, so this this should be generally a company-wide document that everybody should read, but also it defines how you set up your system. I want to throw out another poll just to get a feel for how many people actually have a formal AI policy. Okay, that’s a resounding no so far, but I’ve got a few here that do have one. Excellent, it’s good to see. Okay, so we’ve got a few people here. We’ve got a small minority of you that do have an AI policy, and it’s it’s quite a few of you that don’t. It is really important to have a think about how that works. One of the things that Mitch and I work on and help out with is creating these policies, standardizing the documentation within your context, helping you write your first skill, helping you work on connecting your tools and how to connect them and what to do with them. Sometimes people connect a tool, but they’re not really sure. On the day-to-day, how they can get the information from these connected tools and operationalize them, and you know the the most important part is drafting an AI policy, making sure that internally everybody’s clear on what that means, the do’s and don’ts, and and and primarily why, right? It’s it’s all fun and games to say don’t do this, but you know why are you not doing this and what are the repercussions of that you know from a regulatory standpoint but also from an internal risk standpoint of the organization.

Mitch: Yeah, data and AI literacy are are just critical to good decision making and you know one of the things that also often happens is you know this is a topic where people have lots of concerns sometimes some big feelings and Just being able to vocalize what are the things that feel like risks, what are the concerns, what feel like boundaries to to some of the people on the team, and how can we craft a policy that really respects all of that and is you know inclusive in in that way and and a few others, and really make sure that you know coming back to one of the opening things I said that this is something where people go, yeah, we wanna we wanna move forward with this, not. We’re afraid of it and we’re we’re gonna kind of drag it along because we feel we have to. So how do we really get all that in the out in the same room? And then build all these things.

Mohamed: All right, so we are getting into we’re getting into the Q and A session. So I’m just gonna throw in a If you want to connect with me and set a meeting, you’ve got a book a meeting link right there, and that will help you set up a time to discuss if you need an AI policy. We run a program called the Jumpstart program, which runs every organization through all of these these floors, setting up these guardrails, setting up these. Foundational documents connecting everything together, making sure that all of everybody’s instance in an organization has the right tools and connections, and get an understanding of like how all of this works, and defining a specific workflow per you know for the organization to streamline that process so that we’re showing you tangibly how to shave or save time. Or improve a process, or really unlock the potential of AI in general.

Mitch: All right, so there’s a couple of easy questions we can bang out here. So I’m just going to jump on that. Is one one of the oldest ones here is (Question from the audience): can you ask AI to forget content if the content you work on is sensitive? So most platforms are going to offer A private chat to begin with, so it’s ephemeral, kind of like a private browser window. If there’s a memory that’s stored from a conversation that isn’t private, you can have it deleted, or you can go and access that and delete it yourself. If you delete a chat and there is no data retention in your, you know, agreement and policy, then that would no longer be there unless it was documented in some other external document. Which you would see happen in the chat. It wouldn’t happen as a sort of secret or surprise.

Mohamed: Absolutely. Another question here is, this is a good one here from Anne. (Question from Anne): Wouldn’t giving your AI access to your CRM and email expose all sorts of sensitive slash personal information? This is a good one. You want to take this one, Mitch?

Mitch: Yeah, well, I I I feel like we’ll both take it, but I I want to jump in and say, so, there’s not a lot of personal sensitive information that I actually send to people through email because I have other systems to handle that information. So some people send passwords through email or credit cards through email, for example, and. Don’t do that at all, because you’re relying on the other person to delete the message. You’re you’re better off using WhatsApp where you can delete the message for both people, and it’s fully encrypted, for example. So, yes, you are inherently exposing data, and if you have a lot of private sensitive information in your email to begin with, that’s a real problem because it gets copied onto the email server of. Every person on the chain, without your ability to delete it, so I would actually pose that as the upstream major issue to be concerned about.

Mohamed: Yeah, and if you do recognize that there is personal information or sensitive information in your emails, then block it from connecting to your email.

Mitch: Yeah, don’t give it.

Mohamed: Your email.

Mitch: This is an org by org thing, right? Like I’m not saying that’s the right move for every organization either. By by any means, I’m not suggesting that.

Mohamed: Yeah, but as a baseline, if you’re sharing any passwords, credit cards, serial numbers, any any security based information, do not do it.

Mitch: No, just stop it. Don’t do that. Tell tell us and we’ll give you some alternatives. And as for the CRM, it also depends what you keep in the CRM. So, it really is an evaluation. Like, can we give it access to that, or can we update what we’re storing in the CRM? Is that really sensitive information critical to be there? Can we make a trade-off to allow ourselves to work with this new tool? By not putting that information in there and keeping it in some other secure place, which is untouched.

Mohamed: I’ll add to that that in a CRM, usually you have like a company object or company record versus a personal or information a a identifiable personal information like a contact of a person. You can choose to say you can have access to the company record where company information is available to the AI. But not the individuals in that organization, so their emails, their phone numbers are not exposed. Your conversation history, if they have like like you have you have HubSpot or Salesforce that keeps track of all of that, you can exclude it. So you have granular controls to make sure that you don’t overshare with an AI, which is the mechanical thing we were talking about before. Yeah, someone here, Sarah, (Question from Sarah): is create a skilled equivalent of a copilot agent? Yeah, they’re fairly similar, but they’re not.

Mitch: Yeah, and agents, if I recall, I mean Mo, feel free to jump in, but agents are essentially the equivalent of projects or custom GPTs where you have a sort of less modular context.

Mohamed: Yeah, so with a copilot agent, I believe it is you create a persona as let’s say a marketing coordinator agent that has access to certain baseline information and has access to skills. And if you say, “Hey, marketing coordinator agent, create a comms campaign for a new service that we’re building creating.” And it’ll go off and it’ll write a blog post and it’ll create a social post and it’ll do all of these things within guidelines because it’s got that persona with it. Agents are are a tricky thing to to create. They are super fun to do, but they’re pretty high level on top of skills. But we’re seeing more and more like let’s say Google and Claude have been moving away from. People creating agents more towards just creating a series of skills that you as a person you’ve got one agent which is the the chat itself to say create me a social post with this skill in mind and it’ll go off and do that as opposed to having these multiple agents with these granularities.

Mitch: Yeah, and one of the reasons why people would want It set up this way is, let’s say you have one agent for writing blog posts and one agent for writing social media posts. Well, you might say, why don’t I make that the same one? And the answer is, you may get worse quality. So now I have my brand writing rules, which exist in both places, and I updated one, so now I have to update the other. Can’t I just share them? And the answer to that essentially became skills. The skill is I said I’m writing a blog post, so look up the blog writing skill and the blog writing skill says look up the brand voice. And so now we have these Lego blocks that we can easily mix and match and combine rather than having this knowledge management burden that really was getting created by the sort of siloed agent model. So it has some uses, but. This is why skills are really a sort of next level way to work.

Mohamed: Okay, we’ve got two questions here, which are a pretty big set of them. And oh I just lost the question here. Okay so From Sophie, (Question from Sophie): what AI platform do you think is worth paying for, or do the free versions allow you to create skills?

Mitch: So don’t use the free versions.

Mohamed: I’ll start off with saying, as an organization, do not use the free versions of any of these platforms. One primarily because they will use your information for training data. Some of them now allow you to turn that off. But that’s not always guaranteed. When you’re using the company version or the team version of it, they now have a responsibility to keep your information safe because these are you’re entering into a contract with them that basically there’s going to be personal information or sensitive information that is company related, right? So you wouldn’t expect, let’s say, Bombardier to that’s a terrible example, but like a large organization that has trade secrets, for instance. To trust, a organizational version of a platform without knowing that it’s just not going to leak out their trade secrets, even if they have the most insane policies. So in saying that, all of them are pretty good, and they’re getting better by the day. So, you know, it depends on what you do and what. You would like to do. So, for instance, if you’re using the Microsoft stack, you’ve got Copilot. It’s embedded everywhere, and it’s relatively good for a lot of things. We’re pretty biased here towards Claude. We find it to be the best all-rounder in a lot of cases, but ChatGPT is also a very strong platform. Mitch, you got thoughts?

Mitch: Yeah we have 2 minutes left. I could talk about Claude for a while, but we better move on to the next question. We got 1 minute left while I was saying that.

Mohamed: The last question is a pretty big one.

Mitch: The formal policy. Oh we have 2 last questions. So I mean yes yeah this is meant to be implemented across the team or else it doesn’t really make sense. Yeah.

Mohamed: A formal policy is something that’s a conversation within the organization itself. So every team member should be involved in terms of, or at least the leadership, in terms of what access should be given to which platform and why. And there is this one is a very big philosophical one. It’s going to be very hard to answer this question. I I have an answer.

Mitch: Yes. Yes. I mean it it’s it’s a personal decision. So we could talk about this for a long time. But ultimately sure it’s worth saying.

Mohamed: Yeah it’s it’s something that’s already been done. You know a lot of these platforms to get to where they are now. Have been trained on lots of public data, lots of bought purchase data, lots of books. Some parts of it are ethical, some parts are not. But now that the cat’s out of the bag, there is a lot of conversations about regulating it, putting in these guardrails. You’ll see a lot of these AI platform CEOs up in Congress every other day talking about how we need to slow down or we need to be. Regulated and all of that, and there’s bigger conversations about the ethics around it and what kind of training data. So these conversations are being had as as we go along. It’s really hard for as the individual to be involved in this, but at the same time, there are privacy advocates, intellectual property advocates, and people pushing back. So this this is a some a story that’s being unfold. Unfold, you know, it’s it’s coming along as we’re we’re we’re writing this journal.

Mitch: Yeah, and there there are models that are more ethical leaning as well. So there is maybe like a pick your poison kind of thing, but there there’s like shades of you know trading like maybe quality and performance against you know having less of some of that, but also a bit more work to set it up yourself. You know when it comes to the environmental concerns. Being effective at using AI in terms of being efficient with prompting can be at least one way to reduce the impact as well, and understanding at least how that works and how to work efficiently.

Mohamed: Okay, so we are at time. Like to thank everybody for joining today. Thank you, Mitch, for Copiloting this session with me, this is a it’s a big conversation. It’s a big topic. As I said, if you need any help or need to implement AI within your organization, you can book a meeting with me. There’s a little bit little button at the bottom there. We can discuss what you what you need and how we can help you with that journey. But In the recap email, we’re going to be sharing this video, the presentation itself, and and a summary of everything. So you can share that with your team or rewatch it later. But yeah, for session two of three of this this is it. Yes, Mitch.

Mitch: We missed one question. (Question from the audience): Is Claude a better AI tool than ChatGPT? In my opinion, absolutely yes, 100%. I just wanted that to go on the transcript before we ended. In my professional opinion.

Mohamed: Thanks, everyone. Thank you, everybody, and we’ll see you in the next session in November.