BE_Ep10
===

​[00:00:00]

Blessing Richardson: There are a lot of benefits to talking about AI on your team, like a person that fills a role on the team, and there are a lot of benefits to looking at AI as one-off automations, that complete work and complete tasks one at a time. We're gonna talk about both and why I think you need a little bit of both, but ultimately, we're gonna start with one task at a time. So it might sound a little bit undecided. I'm really not start with the task, but we'll come back to that. So let's talk about, AI from the perspective of filling a role. Most of us, as we are scaling our teams and our businesses, we hit this bucket of [00:01:00] admin things that need to be done. And that's why I think we're hearing the language of, you know, you want an AI assistant, you want an AI executive assistant, because it's kind of like the catchall.

For all the things that we're doing early on that are not a part of our main revenue driving activities or said another way, the things that people pay us to deliver. So that could be things like managing our inbox, our calendar responding to inquiries, whether those come from, you know, our inbox or slack.

It could look like cataloging invoices and receipts. It could look like for a while owning social media. As part of an assistant role, right? All these other things that might traditionally sit under different departments and quotes, you know, that just are outside of revenue generating activity. And so we have a list of those things and we say, you know what?

I just want an assistant. I just want an AI thing that does all the things for me, and if I had to write my laundry list, this is what I would do. Now, notice the things that I came up with are pretty high level, right? We're gonna have to sit down, [00:02:00] have a conversation, and figure out what does it mean to do those things for you.

I like to start the conversation though from a task first approach. The reason is because when we start talking about tasks, I reframe that as pain points. Where are the points of friction? Where are the points where in your processes, in your systems where things aren't working really well, that also helps us to get to priority rather quickly and get really, really well defined on what it is we're trying to automate.

So for instance, if I'm gonna talk about my AI team member being less of a team member, more of just an automation in the business, then I would probably say I need an automation that manages my inbox. And by that I mean I want it to wake up in the morning. Check my inbox for any emails that are new and that came in.

If there are newsletters, I want it to flag it as something to read later. If it's something that requires an urgent response, it marks as urgent and it stars it for me to see that. If it's something that needs to be done within the next, you know, seven days, again, that has like a deadline [00:03:00] associated with it.

Mark Market urgent. You see how like I'm going and painting this picture now because now that I'm starting to think of it as a, as a task then my brain walks into gear about, okay, so what exactly needs to be done? 'cause in order for somebody to do a task, you gotta sit down and have a set of steps you gotta complete to then say the task is done.

Where I think roles will make us think about responsibilities. And ergo, we still are kind of fuzzy on the task when we start there. If I were to build that agent that manages my inbox in the morning and simply helps me label it, that's one. That's one skill, one task at a time. Great. And then if I add the ability to know whether or not I need to draft a response to the agent, that one's a bit trickier because how does the agent know whether or not I need to generate a response?

It can make a best guess, and I've built some of these systems and I found that sometimes if I get an email that's telling me, Hey, this domain name is gonna get renewed, so I have like, you know, something.com. And it's coming up for renewal. Then sometimes the AI [00:04:00] will realize and think, oh, okay, well let me just draft an email, responding and letting them know.

Thanks for letting me know. The domain will get renewed, and it's just like, that wasn't necessary. One better prompting, of course we'll fix that, but two, how does the agent know? How does the agent know? We have to come up with rules to that point, we have to come up with rules around what we deem as needs a response.

If I have a rule that says, Hey, if it's any inquiry from a client, then we should flag that as a client email and then just roll it up and let me know, okay, what's a client email? What system should I go to to figure out who's a client? That's naturally what we should imagine the AI talking to us. If it could, if the AI agent could speak, like we should anticipate that conversation, to happen where it would ask us, how do I know who's a client?

Hmm. You know, it might be obvious for some emails, but what's the difference between a returning client or a new client or just a lead? Like it's just somebody who's asking about general services they don't even provide, and we wouldn't even consider a potential client. How does the AI know? [00:05:00] This is why I like tasks.

Tasks begin to service these questions a lot sooner in the process and conversation and figuring out what it means to actually delegate something to, if not a person in AI. And we ask a lot of the similar questions upfront when it comes to delegating to an AI versus a person. But then we go a little bit further because we have to get super, super concrete about the tools and the systems and the hubs of information and knowledge.

And so I like to call this hubs within the business where we have a communications hub, which is where we send all our notifications. We have our project management hub, which is where we track what's getting done. And then we have our kind of data hub, which is more like our databases.

And where we create what I love to call sources of truth. This is where we say, yes, we might use ConvertKit for our email marketing software, and in that software we have clients and audience members. But a source of truth says in order for somebody to actually be a client that the business is aware about, they [00:06:00] have to be in this Airtable.

That's a rule that tells us no matter what other software we have in the business, it could be Dubsado, it could be HoneyBook, it could be some other tool. You know, some other CRM, if you say authoritatively, any client that we serve has to be in the Airtable. 'cause that's our source of truth of who we've served over time, that we know that all roads point there.

What that also means is that for an agent to look at an email and say, is this a client? We now have a single source of truth for the agent to check to know if someone is or is not a client or has been a client.

And so talking about things from a task first perspective, make it really clear to figure out what are our hubs, what are the tools in those hubs, and does our agent actually need access to those tools and platforms and systems? 'cause if so, then we take that information into our audit and they can quickly figure out, well, can this even be built?

Can the things be integrated and automated in a [00:07:00] way to do it? Is it worth the effort to build the agent this way? Right. Or is there another way to get at what we want, given how our tools intersect? It's not that we wouldn't have gotten to this point in the conversation from a roles first approach is to find that sometimes when we talk about the roles, we're going down this vision of the way we want the world to be and we wish it was.

Because a task often lives within a process, it implies that we are improving an existing process, and that's why we can be so concrete with what we're building and what we're doing when a task can be clearly articulated. And oftentimes the best person to articulate a task is the person who does the task.

That's the right person to have to, in that conversation will start. Building automations around particular tasks. When we start with roles, oftentimes we lose that type of granularity and focus, and we start to design the world as we think it should be. That's easily detached from the world as it is when it comes to [00:08:00] identifying what our hubs are, particularly our communications hub, our project management hubs, and our data hubs. Right?

We start to build something that might not be anchored in the reality of what we have. And that points us back to, well, maybe we need a systems overhaul to make it work. Maybe so. But if we're gonna start to build AI team members, then we need a bit of a solid foundation to start on. Because building AI team members and building more sophisticated and complex automations mean that we're gonna park here for a while.

This isn't a three month lease anymore with our tools. It probably looks like you are considering if you don't already have annual subscriptions for your software that you use in your business. You might even have one to three year contracts or certain vendors for some of these tools, right?

You're moved in, like you're in it and now you're trying to double and triple and quadruple down on where you already reside. When we start building these more complex automations and systems. Because the cost of a tech overhaul at that point would [00:09:00] be substantial, right? It's not necessarily that you would pay somebody else to build it.

You could have a team member who does this, or you do it yourself. It's the cost to your business to build the plane while you're already flying it to stop. And I am speaking from experience. To stop what you're doing with your clients and your other projects and your speaking and all this other stuff to go, okay, wait, we have to like, we have to fix what's already going on.

Versus build something that you can transition into, oh, that's painful. And that often happens when our systems and our tools, we haven't come up with a tool set that we want to quote, marry, you wanna move in with, right? You want to become more attached to, if we're uncomfortable with them,

the role first approach makes it really easy to gloss over those kind of things, whereas a task first approach is more rooted in an existing process and those kind of things come up very quickly in the conversation. So if you're looking for where to start with designing your AI team, I think it [00:10:00] might be helpful to do a little bit of both.

Yes. Start with the kind of responsibilities, you don't need to personify your agents or anything, but I would rather think about responsibilities under like, departments if you want to really take that approach. If your team isn't that big to where you actually have departments and that's a little hard, then I would just look at them as responsibilities.

What are the things that need to get done? The buckets of things, and what are the things inside those buckets? They have to get done and start breaking those down into tasks. So let's keep the hierarchy very simple. Responsibilities, social media management. What needs to happen? Well, chances are you're probably gonna be on your own for creating the long form content as you should be.

But you want to create original content. But then you might have an assistant say, okay, I'd want an assistant to be responsible for breaking up content into small parts. Okay, we're getting, we're getting good definition for a task here. Let's break it down more. Let's say my brand is only on Instagram and Blue Sky.

Okay, so breaking up content into smaller parts looks like taking a long form piece of content, which might be a [00:11:00] podcast like this, and breaking it down into small audio clips and to carousel posts for Instagram. Into posts for Blue Sky, knowing that both of these platforms have different rules and cultures around hashtags, content length, about what types of media does well on the platform.

Do I just need words, right, or do I need images? Knowing that these things are different, and now I might have to come together with an SOP that says, okay, well, we generally have like a set of hashtags we tend to use for certain types of posts. Right? Because we want it to be brand aligned and we want, you know, people to be able to find that.

So this is what we prefer to use on this platform. We tend to, whenever we say, oh, we're promoting a product on Instagram, it might make more sense to say, link in bio. And so you might put in your SOP for this agent that when we're writing Instagram content, if there is a promotion or a call to action that has a link involved.

Always end with link in bio. You see how this is becoming much more well-defined. [00:12:00] The more we get down to the task and say, okay, let's get more definition around what we're doing and what we're trying to delegate and how, right? And that's the kind of granularity you need when you start really putting pen to paper on defining your tasks.

Now, here's the thing. You don't need your task document or your SOPs here to account for all 100 possible scenarios and use cases and all this other stuff, but getting comfortable with saying, when I build this social media agent, these are the things that it has to do on the first bill.

These are the one to two tasks. So maybe to start we say, you know, I don't have a lot of preference between Blue Sky and Instagram, but I do know what Instagram things that we need to be in place to make that work. And so, okay, we're going to write down that we're gonna do Blue Sky, we're gonna pause.

I'm gonna go figure out my Instagram SOP 'cause that's an easy lift and we're just gonna make a simple automation that will listen to my podcast, RSS feed because every podcast has an RSS feed. So we're [00:13:00] gonna make a simple automation to listens to my podcast RSS feed. It will take the link to the media file so that MP3 or audio file, that's every episode that we stream, and then it will generate a transcript, and then it will pull all of the suggested posts that it can make out of it.

Very simple. Just to start to make sure that we can build something that works. Once you have that in place, then you can go back and make some tweaks. We might link it to an SOP and I and I keep saying that because Relay app makes it really easy to do this with how they set up knowledge for your AI steps.

do I have dreams for more? Yes, but I'm also not unhappy with where we are with it. So you can go and create AI step and you can link it to a specific file. You can upload file and say, this is your knowledge. AI agent refer to this whenever you're doing this task, awesome. But you can also link it to a Google Drive file.

This is not the same as having a step where you download a Google Drive file. It's not the same because Relay now has a knowledge base feature, which means that any Google file I link to an AI as a [00:14:00] knowledge file. It will sync on its own. You see where I'm going with this? See where I'm going? So let's say we make some changes as to how we write for Instagram. We don't wanna use hashtags anymore because, hashtags apparently aren't a thing for Instagram. That's what the interwebs told me. And so we're gonna go update our SOP to say, you know what, we don't use hashtags anymore.

We do use certain phrasing in our posts though. And so we're gonna go into wherever we store our, SOPs, which could be a Google Doc, we're gonna update that. And then the next time our automation runs. It's gonna pick up that file update, and then it's gonna just have our updated SOPs. This is awesome because now when we build this way, your team can do what it does best.

If you're a solo team, you can do what you do best. You get to tell AI what to do. You just get to go to the Google Doc and update the Google Doc without having to actually go and edit and change the whole automation. The automation will pick it up and that SOP will be downloaded, and now that agent will know what to do whenever you run it, to write [00:15:00] that post for you.

In this era of AI, I am so much more attuned with picking tools in my tech stack where I do have an opportunity to make a choice, that have AI native features. ' cause it makes it so much easier for me. Now, like I mentioned, I personally manage my stuff in Notion, which means that I actually have an automation that it's job to export my Wiki database into files for Google Drive.

Because what I found is that most AI tools actually integrate really well with Google Drive. There are a number of reasons why. Probably not a conversation for right now, but they integrate extremely well with Google Drive, if not Notion directly. And so because of that, I make sure to say anything that I know that could be considered part of my AI team's brain is stored in a particular folder in my Google Drive.

As I build my systems, I can just link all my knowledge together and then my AIs kinda get smarter. I don't have to change my process of updating my SOPs and Notion. My AI processes and agents, they [00:16:00] just inherit those changes, and so it creates a really simple flow for managing things. So hopefully I'm giving you a number of things to think about.

One, what approach are you gonna use to design your AI team? I always recommend task first if we're not sure, but even if you do the org chart, don't get so into building the hierarchy of things that you don't touch down. And actually get to the point where we talk about the tasks. How will this AI do the things that it's supposed to do?

And then we wanna talk about does our tool do our hubs support this? Does our project management, the communications hub, and our data hubs. Do these things actually support each other in making this an automated process? And then how do we manage the SOPs around the work? Which part of our data hub is gonna play well together and making it so that our agents can always get an up to date information or instructions on how to run.

If you have started thinking about your AI team and you really wanna make some [00:17:00] time to design it. I invite you to join me for your AI team blueprint, literally at aiteamblueprint.com. This is a small cohort that I'm running with business leaders and builders. Let me be clear, this is for leaders and builders.

What I mean is if you are a team of two, of one or 10, there's gotta be somebody on your team who ultimately. Within the team has the knowledge of how these systems are supposed to work. Even if they're not the person who builds them for you, you can still hire out to build them, right?

What I'm talking about here is the person on the team who says, Hmm, maybe Sam had a little bit of a brain fart today and isn't working. Let me see if I can go. Figure this out and make minor edits and minor changes. Or the person who simply says, look, I might not make changes to Sam, but I know how Sam's supposed to work.

I'm aware of how it's supposed to work. Right? And I work with the consultants to tell them what it was supposed to do, right? We want that builder there. We want of course, the business leader there. And then maybe [00:18:00] without necessary explanation, but for the fun of it, the leader is the person who is helping to authorize and inform what the design of this team is. You have to make decisions about what AI will and will not do in your business. What things it will be responsible for. If you have sensitive information, there's legalities involved.

There are just personal beliefs and ethics around the things that we want AI to do, right? Do you want assistive intelligence where AI is more of an assistant, or do you want artificial intelligence where it's actually making a lot more decisions and has more autonomy? That is your choice to make, which means that this is a really important conversation for you to be part of.

So this cohort is designed for both people to be present and to go through three days of workshops and trainings around how to actually create this AI org chart and how to audit your tools and figure out what makes sense for you. To identify do you actually have systems that you are happy with and that you're ready to build and automate more into?

Or you're like, actually the system's never really felt quite right, and now we have a [00:19:00] way to figure out how and what needs to change before we double down and triple and quadruple down on building custom automations in our business. So that's what this cohort is all about. Go to aiteamblueprint.com and get signed up for the wait list.

I do think the group container is gonna be amazing because hearing how other people are solving problems is gonna be really, really helpful when you're trying to automate systems that you haven't really automated before, or you have and haven't landed on that you like.

And so that is part of the reason, the main reason why actually, I thought about this in a group setting. As opposed to my normal one-on-one setting is because these ideas help to shortcut the learnings that you might have to do on your own. So again, go to aiteamblueprint.com and let me know that you're interested.