Taking Automation to the Next Level with AI-Enhanced Workflows

Hey, welcome to this episode of The Automation Show. I wanted to do this episode to explain how AI works well with automation. So if you're just listening to the audio version of this, in the show notes is a link to the YouTube video of this episode that will show you what I'm talking about, because we're also recording this as a video. And for those that are watching the video, Welcome. This is the automation that we have built for a client using GoHighLevel. We have our version called Client Connector. That's what we're going to be looking at today. So let's take a look at the use case. So our client has a wellbeing assessment on their website. People that go there, it's a free lead generator. They can answer different questions about different areas of their life, and it really comes down to scoring in 5 different categories of wellness. There's one on social, community, health, financial, um, and I don't remember the other one off the top of my head, but regardless, there's 5 different areas. Now, prior to what we have built here, they were sending everybody that filled this out through the same nurture sequence. Everybody got painted with the same brush. They all got the same email first, the same email second, the same email third, and it went on just like that. Now you may be saying to yourself, well, that's common, right? And that may be something that you're built out. But in this use case, they are actually giving the person that fills out this assessment a result, and it says, here's the area that you have done really well in, here's the area you're not doing so great in, and then the areas in between. So it would make sense that you would want to start to deliver content to the areas in which they are doing the worst in, which is great. You can do a lot of that. But where it came into making us have to think about this automation was the client asked, well, then could they next go to the sequence where they are the second worst in, and then the third and the fourth, and then finally finishing off on the one they're the best in? So I, I started to think about this and went into their HighLevel account, uh, and I'm going to walk you through what it is that we do. So their assessment is built in Gravity Forms. And the idea is that every one of the questions has a score. So if it's strongly disagree, it's 1. If it's strongly agree, it's 5. Okay, and there's 5 different options in between, as you can see here on the screen. Each question is then, um, given that score, and then it's totaled up in the end. Now, in their CRM record, here's just looking at my test account in here. We have custom fields set up for each of those. So each of the different categories— we have purpose, social, physical, financial, and community— there's a total score based off of what they answer. This is great, perfect. But how do we make the automation do what we want it to do? So what we ended up doing was creating a workflow, and we could actually several workflows with this. So we started first with what we call the AI engine. So when somebody completes this, the contact record is populated with the scores for each of the categories. They're going to run through the initial workflow, and we're going to be using the ChatGPT element because when we do this, we are able to actually program it through just written language, just like you would with ChatGPT, to output something. In this case, what we ended up doing was we told it a whole bunch of information about it that you've just completed, you know, the wellness assessment. Here's how we score. Here's the ranges for the different categories that are considered high, average, or low. And then we went and merged in those values. We put in, in this case, all the questions and all the answers because we did put the scores for every question into the contact record so we could utilize those later on if we need to. But in the end, in this first step, we've gone ahead and we said, okay, now take all the information that you learned above and compose an email to go out to insert first name, right? Have this email talk about their highest and lowest areas, like a better summary of how they are doing overall. Let them know that they are going to receive a series of emails over the next few weeks that will help them get better, get a better understanding of these 5 areas of wellness. Write the email to be conversational versus transactional and have it written like it's coming from an actual person. The call to action is to book an appointment with us, and then I have the URL in there. And then I said make sure it's spaced properly and doesn't include emojis. Do not include a subject line or your name at the end of the email with— and end it, sorry, with warm regards because we're using the signature field from the contact in there. And then in the advanced options, we went ahead and just told it a little more information. Say make recommendations on each of the categories based off of what you already know, write in longer sections, write longer sections, shorter paragraphs. And then the other key thing was there, output it with plain text without any markdown or formatting. That's the key with ChatGPT outputs is If you don't tell it to not use markdown, what it will do is it'll put those little asterisks next to the bold things, and it doesn't look good when it comes in an email. So once we've done that, we tested it a few times to get the output of the email proper of what we wanted. In the email itself that goes out, it basically— we merged in the ChatGPT response in here, which was fantastic because now the email came out, had the email that was written by the GPT, and had the signature of the client. Pretty, pretty interesting. And for most people, just being able to have a custom response email is very much a step ahead using AI. But then we went to the next step. Now the client has asked that everybody that goes through goes through an initial sequence of indoctrination emails. So everybody's going to get the next sequence automatically, which is the indoctrination. We don't need to go into that because there is no AI in that. Basically, they're going to enter and then they're going to go through. Now there is one element of this that is key in any automation is when you send somebody through an indoctrination sequence or any other sequence where they could potentially be put back into, not by design but accidentally, meaning if I came back and did this, this evaluation again, I don't need that indoctrination sequence. So when we go through that, I'll actually show you this. What we have here is a branch logic at the top that says, okay, go into the indoctrination sequence if they don't have the tag saying that they completed it. So at the very end of this lovely sequence, they are going to get a tag. So they're applying the completed QLA indoctrination tag. So they came in again, it's actually just going to go and hard stop them, and they're not going to get it a second time. So once they've gone through this sequence that lasts, I think, 6 or 7 days, they're going to go into what we call rank sorting. And this is where we've pushed this automation to a different level. So if we look at one sequence which branches things out, we need to know that we could get them to a certain point and then pick which one of those sequences they would go into. But the problem lied when they finished one of those streams. Let's say it was social. They got to the end. How do we get them to go back through and have the system understand that they now need to go to their second lowest score? In order to do that, you need to have a control workflow. So that's where our rank sorting workflow— and you can call it anything, we just called it ranking sorting. So the idea is they finish that sequence and they go in here, and what it's going to do is this is controlling which additional workflow are they going into. So we made a master workflow with all the different streams in there of the 5 different scenarios. And then we duplicated it 5 times, and then we called it 1, 2, 3, 4, 5, because in those— and you'll see here in a moment— we have the information that will determine where they're going. So let's look at this one here. So the purpose of rank sorting is we go in here and we figure out what is the sequence they need to go into. So the condition on this is to go into rank 1 They don't have to have, or they, they shouldn't have any of the outcome tags. So we called it Wellness Outcome 1 through 5, and there should be nothing on the record. If there's nothing on the record, they're going to go into Sorting 1. When they finish that and they come back to this, then they're going to go into Sorting 2 by having the Outcome 1 tag. Okay, now if, if you can't visualize this and you're just listening to the audio, you you should probably go and check out the video that's in the show notes because you can see it in there as well. Once you get wrapped your brain around this concept, it's gonna unlock a whole lot of automation ideas for you. So let's just say we're gonna go in here, a new contact, we're always gonna go into Rank 1. So let's go down and let's go into the Ranking 1 workflow. So within the Ranking 1 workflow, They come in here and then we have another GPT action says calculate category. And this is essentially, aside from applying a tag, the only variation in each of these. All the emails down below are the same because they are under each of the 5 categories. So we come through here and we just tell the GPT module. The information below contains the scores of how the contact ranked in the 5 categories. Please output the lowest scored category. Output just the name of the category. And then below I have purpose, and then I merge in the score, social score, and all that kind of stuff. So though— so it's outputting this word right here if that's the one. Now I did that and then I just threw in a note tag while testing to make sure that it outputted that. So I put the response in here and just check the record to see if it's outputting properly. And it is. So now it comes down. So it's going to output that and the condition logic as we go down here is for social. It's going to say is that output containing social and or does the tag not include that tag that's there? Okay, so that's what I want to have in place for that. So it has that in there, and then it doesn't contain that one there. So that's the key in here. And for each of them, it says, does it have the purpose? And so forth throughout the sequence. So as they go through, the step here is what the tag is— it's putting on that Wellness Outcome 1 when they're in here, meaning when they get to the very end of the sequence and it throws it back to the rank sorting because their contact has that outcome 1 tag, it's going to now put them into the outcome 2. And then they're going to go into that, and essentially in that one looks pretty much the same except now when it calculates the category, I say output the second lowest scored category. And then we're gonna apply the wellness outcome 2 tag. Now one other thing I mentioned or didn't mention in the last one was that when they go through and come back through, we need to remove the previous tag that's on there so they're only remaining with the one that they're going into. So that's how the flow works. When you've got it built this way, you have a completely Customized journey based off of where people are coming from, their outcomes, their scores, their ranks. They're going to get the sequence that makes the most sense for them. Now, one other element that we haven't built into this that could be done at any point is we could branch logic the top by generation. So the client has been asking about, you know, well, I've got a wide range of clients., in different generations. So maybe, you know, people under 30 and people over 30. And if we had that question on the, the assessment, we could essentially then copy everything. And when they hit each of these, make a complete duplicate of everything from the top down, but change the wording in the emails to be more in line with the, the younger generation or the older generation. And then the logic's just gonna say, does the contact record say under 30 or over 30? And then they're gonna go into the right stream. So as you see here, you can start to build this journey out to really speak to your audience very particularly. This is fantastic. Like the ability to be able to tell the GPT AI module to output something based off of what you're feeding it and then run branch logic off of that output opens up unlimited possibilities to you in automation. So hopefully you followed along with this. I know it's probably a little harder to understand in the audio version of what we built. That's why I'd recommend go check out the YouTube video. Again, the link is in the show description for this. And if you have any questions, please leave a comment on the video. I'd be happy to answer those for you as well. Well, until next time, keep automating, and thanks for listening to The Automation Show. If you like what you hear, please like it, subscribe, rate us on whatever platform you're listening to us on. Thank you and, and have a great day.