AI Bots Made Simple: How to Automate Real Work

[00:00:00] [SPEAKER_00] Alright, so right now I have AI Bots handling about 20 hours a week of labor for me and the team, collectively probably more than that. Not just tasks, actual work that we used to be doing or that we wanted to do that we now have AI Bots doing. And so today I'm going to show you exactly how starting from a simple piece, a simple skill in Claude and building up to a full automated system that has multiple bots relying on one another to get work done. Welcome to Payrollin.

[00:00:35] [SPEAKER_00] And now, here's your host, Matt Vaadi. Let's go! But before we get started, let's define the word bot because even some of these things I wasn't sure if they classify as a bot, but the definition of a bot is quite simple. It does three things. It takes an input, it applies judgment, and it returns a decision without a person doing that reasoning step by step. So under that definition, a bot doesn't need a face, it doesn't need to be operating independently all the time, doesn't need a memory,

[00:01:04] [SPEAKER_00] it doesn't even need to use the chat window, it just needs to make a call on its own and do some work. So like I said in this episode, I'm going to show four different bots. One looks like a set of instructions, one looks like an automation flow, and one of them looks like the chat bot you might expect from an AI bot.

[00:01:19] [SPEAKER_00] So the first one is Claude skills. So put in the chat if you're somebody who's Claude or if you're a ChatGPT person, I don't know if there's an equivalent in ChatGPT, so we're going to talk about Claude skills today because that is our weapon of choice at our companies. So Claude skill is kind of the smallest working unit here. It's not a chat conversation, it's a fixed set of instructions with five required parts.

[00:01:45] [SPEAKER_00] So a trigger tells it when to run, context layer gives it the background that it needs, a process then defines the steps, and an output format fixes what the results look like, and then a quality check confirms the result before it counts as done. So the example I have here on the screen is what we call our client pre-research skill. And so what this pulls for us is it does a digital presence scan, live web research of a potential prospect before we have the sales meeting.

[00:02:15] [SPEAKER_00] So it pulls up their website, it helps us identify their value proposition, how their website looks, it pulls up different information about their Google business profile, their LinkedIn, their Facebook, their Instagram. It's going to pull up competitive intelligence. So who are two or three of their top competitors? How do they compare in their positioning? What angles are they positioning versus one another? Then it does some framework mapping to fit into our sales process.

[00:02:42] [SPEAKER_00] So what I'm doing there is like, hey, what are our normal initial sales processes look like based on the information that you found? How do those things pair up? So we have a finished brief before we ever set foot on a sales meeting with this prospect that gives us everything we want, the overview of the company, potential problems they have, anything we can find online. It does all that pre-research for us by simply putting a forward slash, putting in the skill, and then the domain of the client, and it goes off and running for us.

[00:03:11] [SPEAKER_00] So that one saves a lot of time every week, and you can use that for, you know, beyond sales as well. Just getting ready to meet with somebody, a potential partner, a potential friend, whatever it might be, just gives you a lot of great context. So next up, the next thing that you want to do inside of Claude is orchestration. So you're going to start to hear this term a lot in AI, an orchestration layer. So effectively what this means is it's a decision maker that sits over top of a set of skills or bots.

[00:03:39] [SPEAKER_00] So in another environment, I'm going to show you later, I have one bot that sits over top of my five to ten bots, and it orchestrates which bot to lean on to do the thing. With Claude's skills, this is the same thing. So inside of, again, our delivery process of how we deliver our service, we have skills built out for each stage of delivery.

[00:03:57] [SPEAKER_00] So now we have a bot that sits over top of it called the orchestration layer that reads and says, hey, which of these things needs to take place based on where the client is in their engagement with us? So it's going to understand, you know, in this example of, all right, hey, this client is missing. The most important first step, which is the client master file. And so it's not going to be able, we're not going to be able to do module D, module E, module A, module B, whatever it might be.

[00:04:25] [SPEAKER_00] It's going to know which steps depend on other steps before we can get to the next thing. So not only is it doing some of that work for us in terms of, okay, great. In this situation, if it didn't have a master file, the first thing I would say is create a master file for me. What are you looking for? And then we can move on to the next step. And so this one's really great. And again, you'll start to hear that term orchestration.

[00:04:46] [SPEAKER_00] One low-hanging fruit example is if you were to be using an AI chat interface that had access to Gemini, Claude, and ChatGPT, it would know which model to pick from to use the lease tokens to solve the problem that you have. And we've built things like that inside of Replit for different tools as well. I talk to founders in the payroll, HR, and accounting industry every single day.

[00:05:13] [SPEAKER_00] They all tell me they need more of one thing. It's not more money. It's not more AI tools. It's more time. How do I get more time? Well, I hired a virtual assistant through OutsourceScale.com that helps me manage my inbox, manage my calendar, handles administrative tasks for me, helps support my marketing efforts, and much, much more.

[00:05:36] [SPEAKER_00] The cool thing about OutsourceScale is that they actually train the virtual assistants on the latest and greatest AI tools to make sure that they can't only do that administrative work for you that you need to get off your plate, but they're doing it in the most efficient and effective ways possible. I've been so happy with my assistant from OutsourceScale that I've actually added multiple other teammates to our other companies through them. Their pricing starts at only $7.99 a month, and they make it super easy to get a new teammate onboarded and part of your team very quickly.

[00:06:06] [SPEAKER_00] Check them out today at OutsourceScale.com. So next up is GrokBot, my favorite new AI tool. I can't say enough about, not bot. So GrokBot adds three things that skill layers don't have. So it has a persistent identity, a memory that carries across sessions, and control of a real browser inside tools.

[00:06:34] [SPEAKER_00] So on this screenshot, you'll see an example of my executive assistant, which yesterday read an email saying that the IPPA owner's retreat is coming up. Would you like me to go and find travel options for you to get there, where to stay, et cetera? And so it went and pulled Google flight information for me. It found the information on the IPPA website about what hotel the group rate is at.

[00:06:57] [SPEAKER_00] Started to put all this together for me into a brief simply by reading my email and understanding and seeing that there was a travel potential there for me, just like your executive assistant would. But to me, this is not even close to the coolest thing that the GrokBot does for me. The GrokBot every week can go through yours or your salesperson's inbox, calendar, and CRM to make sure that the pipeline is completely up to date. So in my circumstance, it goes through all my emails.

[00:07:26] [SPEAKER_00] It goes through all my calendar bookings for the week, pulls the transcripts from the calls, updates the account in the CRM, and updates the stage based on the context that it gets from the transcripts of the call. So it's effectively acting as my sales assistant, moving things through the pipeline for me so that I show up with the rest of the sales team prepared and have things up to date that I'm responsible for. So I absolutely love GrokBot, and there's probably, as you'll see here, we've got a bunch of different bots going.

[00:07:55] [SPEAKER_00] But the most important one I just built yesterday was chief of staff because this is probably only about half the bots I have. And so I wanted to start to have a chief of staff sitting over the rest of the bots to make sure that they're behaving as they should because on occasion they were not sending me the things that they promised. So I needed somebody to manage them, right? So we have a bot manager. And then even beyond that, one project I just had to do today in terms of coordination was it's working with my Claude, with a couple of MCP providers,

[00:08:23] [SPEAKER_00] and helping me build out a dashboard in Claude that reflects all of our scorecard, our measurables that we're tracking on a weekly basis as a business. And it's making sure that those are populated and all the connectors are there, and it's connecting everything I need directly to Claude to make sure that we're able to pull that information. So it's actually doing technical work on my behalf as well.

[00:08:48] [SPEAKER_00] And so the final scenario that we would use, which is a little bit more complicated in terms of the setup, but ultimately helps us to really strap a bunch of things together, this would be more of an automation flow. So we did this in a make.com. So make.com can basically strap a bunch of your systems together, and right in the heartbeat in the middle of this thing is an AI decision-making tree that helps us determine what should happen next with this action.

[00:09:15] [SPEAKER_00] So effectively what happens here is a ticket comes into our inbox. It determines whether or not it already has access to the information to draft a response to that ticket or whether or not there is sensitive data in it. If there is sensitive data, it turns it red and assigns it to an agent. If it has the data to draft or apply, it makes it yellow for agent review.

[00:09:36] [SPEAKER_00] And if it's a very basic or spam or something that can just be archived, it's green and goes ahead and handles the ticket on its own inside of our Zoho desk. So this is a sort of higher level, something we built out over months and obviously have to do a lot in terms of security and protocols and how we handle that. But that is one way that we redirect a lot of what used to take human hours and handle them with AI bots. So those are four examples that we use internally.

[00:10:04] [SPEAKER_00] Looking forward to hearing some different things you're doing with AI, and hopefully this helped you to get some inspiration and ideas for what to do next. If you enjoyed that episode, please share it with someone else you know who might enjoy it and learn from this. And also, please rate us five stars on your favorite podcast player. We really appreciate you taking the time to listen.

[00:10:32] [SPEAKER_00] And also, don't hesitate to reach out with other topics you'd like to hear more about. Thanks so much.