The "Enthusiastic Intern" and the Impact Effort Matrix for AI Adoption with Andrew Herbert

Welcome to MSP Mastery, the podcast for MSP owners and leaders who want to build a better MSP; one that actually works for them.

I’m Jeni Clift, joined by my husband and long-time business partner, Nick Clift. Together, we’ve spent decades building, scaling, and eventually exiting our own MSP business.

Over the years, we’ve seen firsthand that the MSPs who thrive are the ones willing to evolve. And right now, few shifts are more important than the rise of AI and automation.

In this episode, we sit down with Andrew Herbert, founder of Remap AI, to pull back the curtain on how MSPs can actually implement AI without getting lost in the noise. Andrew shares his "operator first" approach, explaining why treating AI as an "enthusiastic intern" is the best way to maintain quality while scaling output.

Here’s what we covered together:

✅ Why thinking of AI as an enthusiastic intern helps you manage its knowledge and its limitations
✅ How to use the Impact Effort Matrix to identify the high value friction points in your business
✅ The "Annualised Sunk Cost of Time" and how it creates a financial business case for automation
✅ Why MSPs must avoid free AI accounts and the critical security risks of "leaving the keys on the table"
✅ How to transition from being a tech focused MSP to an "AI ready" business partner for your clients

We created this podcast to share the real conversations and lessons we wish we’d had more of while running our own MSP — practical insights from people who understand the challenges, pressures, and opportunities in this industry.

Whether you are looking for quick wins in your operations or a long term strategy to lead your clients through the AI revolution, this episode with Andrew offers a grounded and practical roadmap for what is possible today.

👉 Connect with Andrew on LinkedIn: Andrew Herbert
🌐 Learn more about Remap AI: Remap AI
🎧 Listen to other MSP Mastery Podcast episodes here: mspmastery.blog

Transcript

SPEAKER_03 0:00

Out of the box, AI wants to please you because it wants to help. And so you need to put those constraints and those prompts in to say, actually be cynical. Actually, pull apart this strategy. I want you to view this from a negative perspective. I want you to constrain yourself. Any of your clients are using any artificial intelligence models for free. Your data is being retained to be trained on. Quadrant one is high impact, low effort. They are the ones that we will prioritize first because they make massive financial impact and there's minimal friction for us to be able to automate or solve those with an Agency solution.

SPEAKER_00 0:40

Welcome to MSP Mastery, the podcast for MSP owners and leaders who want to build a better MSP, one that actually works for them. I'm Jenny Clift, and alongside my longtime business and life partner, Nick, we unpack what's really working in thriving MSPs, including insights from the trusted partners who support them. Between us, we've clocked up more than 60 years in the MSP industry, long enough to have tried all the shiny new tools and the latest game-changing SaaS products that promise the world. This is MSP Mastery. Here's Nick, myself, and today's special guest. Today we're joined by Andrew Herbert, founder of remap.ai. He describes himself as an operator first and a tech person second. After starting his career in defense and then spending a decade building businesses in property and consulting, Andrew learned the hard way that the best systems are the ones that actually save you time, not the ones that add more to your plate. Today he helps business owners cut through the AI hype to find practical tools that actually work, whether that's automating the boring stuff or setting up smart AI agents to handle customer chats. Andrew is joining us to break down how to build a simple, effective AI roadmap for your business or for your clients, so you can stop feeling overwhelmed by the tech and start focusing on the growth. Andrew, welcome to MSP Mastery.

SPEAKER_02 2:10

Thank you very much for having me. Hi Andrew, good to see you again, Matt.

SPEAKER_03 2:13

And you?

SPEAKER_00 2:14

A fellow Bali resident.

SPEAKER_03 2:16

Yes, yes. You know, and I wouldn't be here if it wasn't for you two. Absolutely. Oh really? I'm sorry.

SPEAKER_00 2:24

How long we're two and a half years in? What are you one and a half?

SPEAKER_03 2:28

We're just snugging two years now. Yeah. So you led the charge, you got in there and did all the hard work and finding stuff out, and then uh yeah, then I was able to come across leverage on that. That's what it's all about, isn't it?

SPEAKER_02 2:39

That's leverage. Uh R D, rip off and duplicate.

SPEAKER_03 2:41

That's the way.

SPEAKER_00 2:43

Yeah.

SPEAKER_02 2:44

Leverage other people's journeys. And that's what this whole podcast is about, to be honest. It's about getting guests on and speakers that have had a journey, shared their experiences. The typical EO thing. We're all EO members. So we met Andrew through EO, and it's all about sharing experiences and what works, what doesn't work, rather than giving advice. So we stick to those kind of core values. I think the world's a much better place. Absolutely.

SPEAKER_00 3:07

And you moved with your family, which Nick and I had a much easier journey, but I'll get you to dig into that in a second. But firstly, as an EOS implementer, I always start meetings this way. Please share your personal and professional bests for the last six months.

SPEAKER_03 3:22

From a personal perspective, family here, gelling, and exploring beyond barley. So I actually didn't read this as a part of the questions. I actually went to the questions part there. So personal and professional best of the question.

SPEAKER_00 3:35

So you didn't follow the instructions. Typical EO.

SPEAKER_03 3:39

I saw the line question. I didn't read the line above it.

SPEAKER_00 3:41

Guest segue, that one.

SPEAKER_03 3:44

Got that part. Okay, so look. Best personal and professional for the last six months. Personally, family, just enjoying the lifestyle here, exploring Bali and beyond. We've got our year planned out for checking out the region, trying new destinations we haven't. And with young kids, that's always a bit of a challenge. But Bali's a great hub for us to kind of, you know, bounce from compared to to back in Australia, which is a considerably longer flight. That was really enjoyable for us to explore locally and now we're looking at regionally. So I think that'll probably be the personal is just spending that family time. And from a professional perspective, look, in the last six months, the number of moments in the AI space where we've just looked at something and gone, what the hell are we going to do? Everything's just changed. And I'm pretty sure those in the emerging tech space have all got whiplash at the moment with the recent updates and changes. But probably professionally, the best from the last six months was a decision we made about nine, twelve months ago to build out a compliancy framework and platform. And all the changes that we're seeing in agentic AI and hyperscalers and things like that, that was probably our best move because that is robust and that is maintained consistently despite all the other things that have gone on. So I think that, yeah, private AI is probably the best, professional best from the last six months.

SPEAKER_00 5:05

Nice.

SPEAKER_03 5:05

Yeah, nice.

SPEAKER_00 5:06

And just to touch on your personal best, we've gone a little bit that way as well, because when we discovered that Indonesia is made up of 17,000 islands, we're on a quest to visit as many as we can. So next Friday we're off to Lombok for a long weekend. So Friday back Monday, which I think will be our 13th island. And then we'll only have 16,987 to go.

SPEAKER_03 5:28

That's achievable, right? Absolutely. We actually haven't done that many islands, to be honest, but very big on the moment in getting a car and just going for a drive and heading three-day window. We're gonna head that way and just discovering the towns and kind of the the local experiences and being a tourist in the place that you live. That's been really enjoyable. And the kids love it as well. Kids are easy when we actually get out of the house.

SPEAKER_00 5:55

They're just add Seriret to your places to visit, North Coast, west of Lavina. There's a little resort, tiny little resort that we found on the beach, and we absolutely love it up there.

SPEAKER_03 6:08

I think Northwest is kind of we're two trips away, heading in that direction.

SPEAKER_00 6:20

Okay, so tell us about yourself. So in your intro, I know you came through the army, but how did you end up with working or founding REMAP?

SPEAKER_03 6:32

Sure. I guess if I go back a little bit just prior to the Army, I did software engineering in university and then got lured away into the defense force. Did nearly 10 years there. And the last couple of which I was in ISSO and uh, you know, information system security and physical security, compliancey side of things for defense. From there, I transitioned into some contracting roles in government and aerospace. And then I found myself really looking at the corporate world, going, I don't think I'm aligned to this right now. Um, I say that because my experience was that it was very much a time enroll rather than your performance. And given my experience in Defense Force, that was performance-based. I just found that that was not something I was aligned to. And I found myself chatting to somebody at a barbecue about their property business, and they offered me a job. And so wasn't a real estate agent as such, so you know, reserve judgment. But we went into the investment space. I worked there for 18 months before going out on my own. So 2013, I started my property business, got an AFR Fast Starter Award by 2016, and the majority of that was around automation and leveraging offshoring and systemization. 2018 started experimenting with artificial intelligence, training our first model. And around 2020, I was having conversations regularly about how to automate with fellow business owners and with clients and for the property business. And I was a bit surprised by the fact that something automation that I've been doing for the last six, seven years at that time, that nobody else seemed to be doing it. And that really started the journey of going to the consultancy space about that automation. And that's where remap grew from. So now we have uh, you know, primarily do artificial intelligence roadmaps, looking at how to bring artificial intelligence into business systems. So remap business operations with AI. And we've also got a couple of apps that we've spun up as a result of what we're seeing in the marketplace. And so that's how I got into AI. That's good. It's an ever-changing thing, isn't it? You wake up in the morning going, oh, that just dropped. It is. I mean, I mentioned before that some days you've got whiplash. And some of the stuff that's happening in the moment in the startup space, we've built out, I think, three, four applications. And every single time I build out an application, I vow never to do it again. And then I talk to clients and there's an opportunity, and I'm like, we can do that, we can fill that gap, and we start building another app. And the issue with that is then a hyperscaler like Anthropic or Google drop an update. They push a product that they've been working on, spending millions of dollars developing, and it wipes out your application's functionality. Solve that problem. Yeah. Claude dropped co-work a bit over a month ago, wiped out$285 billion on the market cap of the SaaS startup space. And they rolled out again with cybersecurity update about two and a half weeks ago, and that wiped out up to 20% on the share market for cybersecurity SaaS. And they've just rolled out the financial analysis. They just dropped another one. Five models have dropped in the last week and a half. Like it is a rapidly changing space and emerging tech. And the challenge that I see is that because the goalposts just shift so rapidly, I'm at the front end in the early, early adopter stage, that front 1%, that businesses they they don't necessarily see what's coming in nine, 12 months. And that adoption curve used to be two, three years, but the speed at which this is happening, it's it's it's ridiculous. It's you see something beginning of the month, end of the month, it's a whole new, a whole new ball game.

SPEAKER_02 10:09

And that's why I thought it'd be a good good idea to get you on and talk to our listeners because most of them are in a technology space, managed service providers, and we are trusted advisors to our clients, and our clients are asking us, they're seeing all the press and all the reports saying 85% of businesses are investing in AI. And they the question always is how can I help my client? How can I how can I even use it myself so I understand it enough to advise the client on it? And and it's and like you just said, it's not about building an individual app to solve an individual problem, it's more of a a framework and an understanding of a of a roadmap and being preparing your as an MSP. How do we prepare our clients to be ready to take advantage of some new Wangfangled thing that drops and it's a game changer? If you haven't got your stuff in the background ready, if you haven't looked at your business processes and you know what the where the opportunities sit, you're never going to be able to take advantage of them. Therefore, you're going to be behind your competitor. So that's where I hope we hopefully we get to touch on a little bit of that today.

SPEAKER_03 11:12

That's exactly right. How to do AI and how to keep in touch, it's it's a hard ask. That's what we have we do on a daily basis for clients. So looking at how to bring AI, answering that first question, where do I even start? Bringing AI into my business or my or my clients' business. The way that I would recommend people look at AI is there's three pillars. So the first pillar is an off-the-shelf solution. So a meeting note-taker or a one-hit tool that does a very specific thing for you. Off the shelf is relatively low cost. It's self-serve, so setups relatively quick, but it's minimal configuration and it doesn't really give you much beyond what it's meant to be doing. On the other end of the spectrum, you've got artificial intelligence applications. So these are custom applications that you build out. In our experience, they can take more than you budget financially, they'll take longer than you forecast, and they'll be more complex than you anticipate. But the upside of that is it does create underlying marketable differentiation or of value that can add on to your company valuation in the long term. And in between these two, you've got this middle column, which I used to call automation, but is now really very rapidly becoming a dentic AI. And I just want to touch on this. And this is the part where we take off-the-shelf solutions or custom and we chain and daisy chain a number of those together to create a workflow or a solution that kind of fixes a problem or solves a gap. And so that would be the go-to place for our clients, and I'd recommend anyone listening to to look at automation or agentic AI. And just looking at that though, I should say that there's probably a couple of different levels around what there is in the agentic space. There's there's a lot of news at the moment about platforms like OpenClaw and like Claude Co-work and things like that. That is that is probably at the one end of the technology spectrum. Businesses can get a lot of advantage from the other end of the spectrum, which is just simple automation. If this, then do that. It's connecting things up. If an email comes in and it's got an invoice or some form of bill in it, you can load that directly into Xero. But that doesn't need any AI, that is just a connector and a basic workflow, and that will save time. You've then got AI augmented automation, which is where you've got that same workflow, but then you put an AI model, an LLM, in the middle of that and allow it to make a choice or assess something and make a decision about pathways forward. And so it might be email comes in, there's an attached PDF, get the AI to look at the PDF and say, is this a bill? Extract the information. And if it is a bill, load it into zero. So you've got that quality control and you've got that AI to kind of, you know, reduce stuff ups essentially. But then you've got agentic, and that is where we're building a completely different framework, and that's where it's AI agents first. So you've got a couple of different tiers there in that middle column that I'd say people should look at. But that's that's just something you need to be aware of. And then realistically, of how you look at bringing artificial intelligence to your business. I'd point at James Clear, good old 1% incremental change. So 1% 20 times rather than 20% once. Look at these incremental change, these rapid wins, these small wins because you're getting that flywheel effect. You also break down the barriers inside your organization around the resistance to change. A lot of people are really concerned about this artificial intelligence of the threat to jobs. But when you're looking at these incrementally changes, incremental changes inside the organization, which is reading the email and putting the file or the bill into zero, nobody really wants to do that, which means that you can gain that traction. But if you do bring in that 20% shift in and bring an entire new, arguably an autonomous agent like OpenClaw in, that's a massive threat to your existing team. And you may get a lot more organizational pushback. But that'd be the first thing is looking at uh we call them friction points. So those little 1% that you can look at doing. Uh, the second thing would look at is innovation is not something you schedule once a month. It's an underlying attitude inside your organization of we want to improve. So encourage staff to identify friction points or issues or frustrations or manual tasks that they're just they're just sick of. And that can be done, you know, 30 minutes or an hour a week, getting everyone to write it all down, capture it, complete a web form, submit it into a database, and then once a week, pulling these things up and kind of are discussing and workshopping which one are going to take priority and then start implementing those changes. The third one there as well is that making sure that staff, when they are testing these solutions, they have funded options. So there's a massive difference between free Chat GPT and paid ChatGBT. So I'd make sure that staff have the subscription fees that they need. And for any artificial intelligence deployment or just automation deployment, there needs to be that business case, that return on investment for the organization. Not just because it's shiny and so you can say you've got AI in your business, but to make sure you've got that ROI. And so that will be the kind of attitudes to have to look at.

SPEAKER_02 16:40

We're building a training LMS platform. And to generate the content for that, we've got our own IP from running our business for 30 years, and we're working with another consulting firm to build out the training content. And then take that content, then we put that into an AI tool to say, make some slides that we can put an audio overlay onto to deliver the content in a you know in a in a platform.

SPEAKER_00 17:05

Using avatars, yeah.

SPEAKER_02 17:07

Yeah, then you take that and you put it into something like gamma and say make them sexy slides that are on my theme. And then you take the output of that and you put it back into something like the script that can then generate the we're talking avatar over the top of the slides and the script, and it ultimately comes out with a video that you can post on YouTube, then you can take that and put it back into the LMS. Like, and I'm thinking, I've done 87,000 mouse clicks. This has got, and I've got to do 50 of these. So that to me is like the kind of thing how thinking of everything you do, if it's repeatable and it's not creativity and it's not inventing something, it's just processing something, like the human creativity. We've got the actual training, the IP, the the kind of our model of that, but all the processing of all those four or five different steps, that to me is where I'm thinking, can I get an agent to work with all those different platforms? And the weird thing is most of those platforms already have some form of agent inside them. So am I having my own lunch? And that's where I get confused. And I I mean I I'm into this stuff myself, and I go, man, I just don't want to be doing this all.

SPEAKER_03 18:12

It's got to be an easier way to this. And look, there is going out and using a tool, something like Scribe, if you go out at scribehow.com, I believe it is, if you go out and grab that, that's a workflow capture tool. And so if you've got something in mind about what you want to automate or this friction point you're frustrated by, is just go through and just record your screen. It'll map all the steps. And then from there, map that out as a workflow, as step by step as dot points, and then work out how to automate in between the points. Don't do the whole thing in one hit unless you've got a development team to support you. But look at how to connect those two applications up. Like, for instance, taking your training content and putting that into an LLM or an AI model and generating a transcript and then sending that over to the avatar in a platform like HeyGen to generate the talking voice, the talking head. Okay. That is a flow that you can just automate fairly easily using APIs and a platform like Zapia or Make. And so plug that together. And so you can do that first leg work and then look further down about taking that same transcript and then pushing that into gamma or Google's got their new slide generator as well, pushing it out into gamma, get the slide deck generated, and then put that all into an online cloud storage where you've got the components compiled for you. Um, that's that's the the way to look at automation. And then you've got maybe two workflows that have got a human in the loop review step. And that, for to be honest, that is probably the optimal way for you to do deploy artificial intelligence in your business. The human in the loop is crucial. We always say think of the artificial intelligence models as enthusiastic interns, potentially hugely knowledgeable, but they need that tempering of experience and wisdom and context of what it's meant to look like.

SPEAKER_00 19:59

That is like a very enthusiastic intern, isn't it?

SPEAKER_02 20:02

It is, right? Yeah, absolutely. Especially my my my mate Bob, who's harassing me on Telegram all the time now, and I've told him to do a few things. Very I'm treating as a like a two-year-old. He's I said, Go find me some flights to Europe. And it goes, Oh, okay. Do you want me to ring them up and book them for you? No, I don't want you to book my flights for me. Oh, I'll check them every four hours and tell you if there's any changes. Fantastic.

SPEAKER_03 20:24

Oh, absolutely. That's one thing about AI, is well remember, is that out of the box, AI wants to please you because it wants to help. And so you need to put those guardrails, you need to put those constraints and those prompts in to say, actually be cynical. Actually pull apart this strategy. I want you to view this from a negative perspective. I want you to constrain yourself. Or it's just going to tell you that's a great idea. This is Oh, yeah, yeah. Self self-affirmation is it's just that absolutely right. That's a great idea. Love that. Love that, Andrew. Absolutely. Groundbreaking. Nobody's ever done that before. And so one of the things I'd say, coming back to that, where do I start? Looking at those three pillars of different types of tech, and then looking at the incremental change side of things, where you start is asking your team, asking your staff, because they know the manual steps that they hate doing. They know where they've they've got that friction or that frustration and get them to put that in a list for you. And whether you do that on post-it notes in a strategy session and you put them all up on the board, or whether you get them to complete a web form or just write it out in an email, whatever that looks like, you get those friction points first. And then from our perspective, we then go and take all those friction points we gathered from a business and we plot them on an impact effort matrix. And so the impact, thinking about that business case further down the line, the most basic metric you can use is how much is this task costing the company weekly, monthly, yearly? And so working that out from a time perspective and work out your annualized Sunk cost of time into that activity. And that's really a a facet of impact, amongst other things, amongst friction points and slowdown and opportunity cost and so on. But from a basic perspective, look at how much is this costing us as an organization? And then the impact, sorry, that's the impact in the effort side of things, a real question of like, how hard would it be for me to train a junior to do this? And if that's like on a scale of one to 10, it's really freaking hard, then that's a high effort score. And then what we do is we then take that impact effort and we plot it on a matrix, just a basic chart, and we create the quadrants. And quadrant one is high impact, low effort. They're our go-to's, there are quick wins. They are the ones that we will prioritize first because they make massive financial impact and there's minimal friction for us to be able to automate or solve those with an agentix solution. The second one is low impact and low effort. And we still go after those because they are arguably still quick wins. They are still something that we can do that will have arguably minimal impact in the company, but they'll help getting that flywheel effect, getting that momentum, getting that confidence by the team to coming on. So maybe they're a secondary project in a schedule. From there, we look at the high effort, low impact. We're going to defer those. That's not really of interest right now because if we've got to prioritize all of our projects, we want something that makes a financial sense to the business. And so if it's high effort, it's going to cost more to solve. And if it's low impact, there's not going to be that return on investment that we need there. And then the last one is the quadrant three, which is high impact, high effort. These are the longer-term projects. We may defer them or we may schedule them out, but they also might require some of the earlier stage projects to be done first because they need to prepare the space and prepare the infrastructure and the framework that's needed. And so we survey or we engage with a team, we get an understanding of all the friction points and opportunities. We plot them out. And then from there, we just go, we're going to do this six, this 12, this 18, depending on what the capacity of the company is. And then from there, you've got your starting point based on your organizational friction points. And the friction points could be something really easy. They could be something as simple as plugging that email account into zero so that you get your billing. And it could be something in my takes 10 minutes. And you could say it's only a five-minute job, but they do it 10 times a day. And that's that's where the incremental change, you know, the compounding interest you get out of utilizing automation and artificial intelligence. And so that would be what I would be saying to business owners, to to look at where do I even start going down that pathway?

SPEAKER_00 24:46

As our audience is mainly MSP owners, they have a client come and they say, I want to do this one thing. I don't want to look at anything else. You say I too overwhelming, too much. So just I want you to help do this one thing. So how would you take that selling of a tool or a solution to one problem into a conversation around let's build a strategy?

SPEAKER_03 25:13

The one thing I I want to have AI in the business and I want to fix this problem. Great. Absolutely got it. That could be identified by the business leaders of inside that impact effort matrix. This is the highest impact or the, you know, the highest friction point in the organization. And you've you've simply got to listen, but you've also got to look at what is the longer-term strategy around it, as you said. And to build that out, there's a conversation and a dialogue there around what's the 12-month and two-year plan, three-year plan, and what kind of infrastructure or tech stack they're running. And I would say that if there isn't other projects or opportunities, the thing you have to do right now is prepare your clients for agenc solutions. And so that means making sure that the landscape from a tech perspective is agentic friendly. Um, and I mean, you know, looking at data structuring, looking at how your client's information is being stored, basic things, running meta tags, have you got a naming convention? Have you got a folder hierarchy? All of this is preparing the data architecture for artificial intelligence to interact with it as best case. Um, you know, in in private AI, when we've got our ETL running, we call it creating structure out of chaos, creating, you know, sense from it all. Because the majority of businesses just go into Google Drive or OneDrive and just go file save, whatever, because they know they can just search for it. And they don't necessarily file things the way there needs to be. And it doesn't necessarily have meta tags. Or all of which helps AI understand what is this file about and how do I engage with it.

SPEAKER_02 26:47

I am just thinking from an MSP perspective, what we're really, really good at is if we have some form of assessment tool. Like back in the old days, we would build a disaster recovery plan. So we we knew we had a checklist of things that we needed to make or sure in place to be have a successful disaster recovery plan. Everyone has backups, but who tests the backups, for example? So I I think what would really help MSP listeners out there is is there somewhere or something where it's like an AI readiness checklist survey, some some kind of tool? Because that's what MSPs are really good. They're really good at taking a standard, auditing to that standard, and then working with potentially an external vendor like yourselves to come up with a solution to solve the problem for their customer. They're not going to want to build the skill set to be uh an agentic AI development house themselves, but they need to know, have a platform and a framework that they can get up to speed on, understand it, have the conversation with the client, run through some kind of audit or test or framework and then come out at the end with a bunch of questions and they can go away, work on that, and go back to the client, say, hey, that problem we talked about, yeah, there is a solution out there. This is the pathway forward. And we we all did this with SharePoint back 10 years ago, but now it's AI's out there and it's just really confusing. Like you can you can go into Claude and ask it to do the framework for you, but that's not what MSPs want. MSPs need to be able to do repeatable quality stuff and be as similar as possible across all their clients. If you have 100 clients, you don't want to have 99 different solutions out there. That's the big challenge, and I was hoping we'd be able to get talking about that today.

SPEAKER_03 28:29

You'd probably find some third-party platforms out there that may have a free online tool to be able to do some form of AI readiness assessment. But I'd I'd say that you actually need to dig into the tech stack and into the business systems to understand how ready they are. And anything else is a little bit arbitrary. From our perspective, we do have an AI roadmap tool that we work with clients to set up and define their six-month strategy and then support them in implementation. And our go-to is using existing systems. We don't want to migrate somebody away from an existing tech stack because they've already got that business traction and the business adoption side of things. And organizational change is really hard. So the majority of what we find comes out of the AI tool that we use, our road mapping tool, is that agentic solution, is connecting things up and making sure things are streamlined and reducing friction inside their existing architecture. I would say, not necessarily out there, there's a third-party tool that I can point at and say, hey, look, go use that one. That's really great. I would say go onto our website under AI Roadmap. There's some widgets in there that can get conversation starters going. There's also a return on investment calculator that we use to kind of talk to clients about whether or not this is a worthwhile activity for them to pursue. Please go and go and use and abuse.

SPEAKER_02 29:49

Yeah, and I I mean there was one, I listened to a guy from Singapore AI at a conference I was at, and that's a government-funded entity designed to increase the AI skill set of Singapore, right? So if you're a Singaporean citizen, you can basically get free, I think it's up to$100,000 Singaporean dollars to do a joint project with Singapore AI. And they have this roadmap tool. And I I downloaded it and I thought, holy crap, I need to be in three PhDs to understand this 4,000 page document. This is this is written for their engineers, not for an end-user business to do a self-assessment tool. It's technical.

SPEAKER_03 30:25

Whoa, it's way complicated. And that's the massive divide at the moment, like in the emerging tech space and all the things that we're seeing, like even with open core and things like that. If you are not a developer, if you are not a software engineer, if you are not technically minded, you're going to have a really hard time in comprehending and understanding what's going on. And so, you know, that's why in my intro, I'm an operator, I'm a business owner before I'm a tech person. So everything that we do is about communicating that from a business context. So that's why I say I don't think that there's any reliable tool out there that you could run on a system. Realistically, if if you want to kind of jerry rig something together for your users, I would be going into something like Claude or ChatGBT, your large language model of choice. I'd be building out a series of questions or a survey that you can then put in there and just basically get an understanding of API connectivity, whether or not there are other solutions to be able to connect up to and looking at arguably agentic frameworks and the existing connection packages that would work with them. So I I would just ask the survey questions of clients or complete it on your clients' behalf and then submit that. And I would think you would find that a majority of platforms that are cloud-based at the moment have got existing APIs or even MCP servers. MCP servers are essentially an API connection for AI models. And so just standardize things and makes things easier. So I would say do your tech stack, submit it in, and say, give it a grade of out of a hundred. How ready are each of these based on connectivity to agentic frameworks, data availability, speed? For instance, we we just looked at working with a client out of the UK recently. And unfortunately, their third party, their MSP, was using a platform that was, it took us six weeks to be able to get some credentials and things up and running because the massive resistance they had internally with working with third parties. And so we were not actually taking the business away, we were just interacting with them to enable our client to have a better user experience. But massive, massive pushback and resistance. So I'd make sure that you ask that in any assessment you're doing around what's the corporate appetite for this provider as to whether or not they will play well with others. And I think you'll find things like Google and Microsoft, the hyperscalers, there are some tools and ability there from an API perspective to just plug in and to work with them in able to adopt AI.

SPEAKER_02 33:05

Got all that, Jen? You understand all those terminologies?

SPEAKER_00 33:07

Oh, look, I'm all over it. Yeah, I've just been uh in claw, open claw now, checking it out.

SPEAKER_03 33:12

Okay. If we're gonna talk OpenClaw, if we're gonna touch on this at all, I just want to say to everyone listening, OpenClaw is the next level of agents. So Google it, go and read the background of it. Basically, it's just been acquired by OpenAI. It's gonna maintain an open source project. The major thing that you need to be aware of with OpenClaw is it's essentially MVP. It's minimum via product, it's the version one, it's the first one out there. So there are problems with it. There are security concerns, there are challenges.

SPEAKER_00 33:40

So you go and put it on your network and just let it go. Is that right?

SPEAKER_03 33:44

You just you just go out, and every single time you leave the house, you leave the keys in there, you leave the car parked in the driveway, you leave your jewelry out on the table, and you just walk out and leave for a week. That's exactly right.

SPEAKER_00 33:54

So And all your credit cards with the pictures.

SPEAKER_03 33:56

All your credit cards with your cash, open up the safe and just leave it there. So OpenClaw installs at root access, which is essentially super admin, and can do whatever it needs to do to your system. So don't for a second install it on your laptop unless you can start hardening it and locking down points and limiting connection points, sorry, and can limiting what it can do. There was a survey done about four weeks ago when OpenClaw was really starting to take off, and they found 42,000, over over 42,000 publicly available IP addresses that had root access for OpenClaw. So these are people that have just gone out and installed it on their laptop and just played with it and hadn't thought to harden it, hadn't thought to do anything around it, and it was just leaving a hole, a gateway for anybody to be able to come in and just take the keys to the kingdom. So just be conscious. The second thing I'd say is because it's version one, just wait three months. There are so many things happening in this space at the moment. It is moving quickly, and you'll see a better solution, more production-ready solution coming. There are already ones out there like Ironclaw that have looked at it from a perspective of a business user and remove some of those security risks. So just wait a couple of months. Definitely keep your finger on the pulse and have an understanding of it. And the understanding that I would I would have you have is that AI agents and AI augmented automation is prompt driven. So what that means is that you need to give it the structure of what you want it to do. And you need to give it an understanding of this is the pathway that I want you to take. And if you get to this decision here, if this, then do that, right? You've got to give it this prompt in this series of instructions. Open core is a context or environmentally aware. So what you then do is you give it the, I'm going to steal terminology here, the North Star, the purpose of what it's meant to be doing. And then you give it the context of what it's got access to and granular, granular information about what's going on around it, and then also your role and how it's to interact with you. And then you leave it alone. And it then is triggered by the environmental, the contextual information and says, I just received an email. I know what I'm going to do. I'm going to check that email and see if it's, oh, the email is a zero invoice. I know what I'm going to do. I'm going to load that into zero. Rather than if it was prompt driven, you would have to write out the prompt. If I receive an email, do this. Scan the email. If it is this, then do that. Open claw and the new, I'm arguing that they're autonomous agents. The new autonomous agents are contextually aware, and you don't need to give them if you see an email, go check it. You need to say, you are this for me. This is who you are. And then you load it with skills and you load it with the ability to do things and you leave it.

SPEAKER_02 36:47

My first attempt, I decided I I've always liked the TV series suits, and I always wanted my own personal Donna. Donna's the PA, the ultimate PA. So I actually used, I can't remember what I used one of the tools to go and research Donna and come back and write me a persona of Donna, and I loaded that into Claw. And straight away she goes, I'm not here to muck around. Tell me something to do. No, you can't do that. Stop wasting my time. Hang on, calm down, no, no, I'm not calming down. We're here. Yeah, oh yeah. If the attitude was crazy, it's Bob the new one. I've said, Bob, you're the Bob the Builder. Yeah, as in the cartoon character Bob the Builder. So you're here to help me build things and do things. It's much more relaxed in its style. But it is amazing when you give that context and play around with it. But you also have to be very careful with what you what you tell it you want to do. Yes.

SPEAKER_00 37:38

Yeah, so you got your donor and you decided you didn't want a donor very quickly.

SPEAKER_02 37:42

No, no, Donna's been retired now. Bob's the new guy.

SPEAKER_00 37:49

So I'm conscious of time, Andrew, because I know you've got a hard finish. But you mentioned security. I also wanted to talk about trust. There's a huge fear around data privacy, open claw, and all of the the I was gonna say limitations, the opposite of that. How do you build safety into, I guess, into that AI roadmap to give people that sense of we can go forward with this and we're not putting my business at risk?

SPEAKER_03 38:19

Security and compliance should be the first thing you're looking at as a part of any AI roadmap. And I would say that from a perspective of where is the data going? And um, you know, we've got a short term. We've got private AI. So private AI is a secure container, and inside of that container, the secure and compliant, inside of that container is a database and we can load information in there. We've got agentic frameworks that means we can run AI agents securely inside this environment. And then we've also got an AI model that we're hosting. So it means that the data doesn't go anywhere. And from a security and trust perspective, I would look at those questions. Where is the data being stored? Where is the data being processed? So the inference. So when you work with an AI model, you send a chunk of information over to it, it processes it and sends the response back. So the question is, where is that model being hosted? Who owns and controls that model? And what constraints have they got on themselves and what's the term of use that they've got? So if you look at somebody like OpenAI, they had a period last year where, due to US federal court ruling that all of their chats over a certain window of time had to be retained for legal review. So if you use, despite your user agreement saying we're not going to touch your chats, we're not going to hold them and retain them, they were still being retained for legal review. So it's an external party that's influencing the models. So from a security and trust perspective, I would really look at how much data is being sent across. So that's the context window and how much information is being sent each time. And then who is it going to? From a technical perspective, you can start hosting your own models. There are platforms out there. I mean, GoTo on Azure, and you can host your own model, which means that it doesn't leave that environment. You're sending the data to that model, it's being processed and then coming back. That's if you trust Microsoft. That's if you trust Microsoft. And look, I used to argue that if you trust them to put your data in the cloud environment, you should probably trust them to process your information. That that used to be my line. But then having seen what happened with OpenAI and others around third-party influence, they can't say no to the US court to say retain the data for legal review. And so I would say if you've got confidentiality concerns, you really should look at hosting your own model or looking at secure environments where you can send the data without any concern. The other thing, as well, is that just be aware that when you're sending data via API to one of these hyperscalers, Google and so on, and OpenAI, please look at your compliance frameworks. Usually ISO compliance and SOC2, things like that, GPR, they may be native in some ways, but they won't actually pull a fush compliant full compliance set down to the average business unit unless you're enterprise. Which means that you could be sending data to an endpoint that isn't compliant and you're breaching your responsibilities as an MSP or a platform provider.

SPEAKER_02 41:17

Definitely, guys, and especially if you're doing this on behalf of a client.

SPEAKER_03 41:21

Finance, legal, medical, government, fintech, just be wary. And the other thing as well is data sovereignty is a big play. So if you're looking in Australia, data sovereignty, onshore processing, other countries as well have that same revision. So just be very conscious about who's getting it and what they're doing with it and coming how it's coming back. So yeah, that's that's what I'd say around just being aware, communicating that to your client as well. We're going to send your data in chunks over to this third party. They will process it, they have a retention. The majority of the hyperscalers have a retention policy of 30 days as a part of logs, and they'll delete it. And they're doing that for appropriate use management and AI model reviewing, not training on your data. And that's something I'd like to stress as well. There is a big difference between logging and chat logs and API logs versus training on data. So the difference being that they're not on the logging side of things, they're not retaining your information to improve their model. They're just doing it from a compliancy perspective, their own side of things, keeping the 30-day log to make sure that their model is not being inappropriately used. Whereas from the training side of things, they send the information into that environment, they retain it, and then they use your information to improve their model. That's the that's the worst case scenario. And if any of your clients are using any artificial intelligence models for free, that's what's happening. Your data is being retained to be trained on. And so that would be the first thing that I would shut down.

SPEAKER_02 42:57

That's the right of use, isn't it? Your data's getting trained on. So yeah. If you're not paying for it, you're the commodity, right? Yep, 100%. Awesome, Andrew. Well, that was a great intro. I'm sure there's a lot more we can go into and a lot deeper. And maybe next time I schedule it and come, we'll put it out to our listeners. And if they've got some specific questions, we can come back and do another session and ask for that. So yeah, Jen, over to you to close up.

SPEAKER_00 43:19

Yeah. Yeah. Thanks, Andrew. Really appreciate your time. I'm sure there'll be a lot, as Nick said, a lot of questions that our audience, because this is something that we keep getting asked about, we want more about AI. How the hell do we do this? We're as curious as our our customers are about the It changes every month.

SPEAKER_03 43:36

So you're more than happy to come back. But yeah, it's a it's a constant battle to to maintain an understanding of what's going on in the landscape.

SPEAKER_00 43:43

And as an MSP, your customers are looking to you to be the expert. And like I said, we're as curious and confused as everybody else is in a lot of cases, or just don't want to know about it. And we've seen a bit of that as well. But thanks again, Andrew. And Nick, thank you.

unknown 43:59

All right.

SPEAKER_01 44:00

Well, that's good. One of my topics I spend about 15, 20 hours a week on.

SPEAKER_00 44:04

Absolutely. Yeah. If this conversation hit home for you or got you thinking, head to mspmastery.blog and keep the conversation going. You'll find all our episodes there and more wisdom from the peers and partners who are shaping the future of our industry. And make sure you subscribe so you don't miss future episodes. We've got plenty more great guests and stories coming your way. Until next time, this is MSP Mastery.