w/ Liziana Carter, GROW AI
Today on the show, a deep dive into the ever-evolving world of chatbots. Kurt Elster sits down with Liziana Carter, navigating through the winding lanes of ManyChat's storied past, from its 24-hour glory days to its more restrained present. But beyond the tech, there's a bigger question: In a world racing towards automation, where does the human touch fit? And how do you strike the balance between the convenience of a bot and the nuance of a human? Plus, some sound advice for anyone wading into these digital waters. This is a story of technology, of change, and, most importantly, of finding the human in the machine.
The Unofficial Shopify Podcast
8/22/2023
Kurt Elster: Welcome back, my friends, to The Unofficial Shopify Podcast. I’m your host, Kurt Elster.
Ezra Firestone Sound Board Clip: Tech Nasty!
Kurt Elster: Today, we’re gonna talk chat funnels, and so… The idea here is sound. You can turn… You send a message and ideally that turns into a relationship, and so all these social media followers you have, we can take a percent of those and transform those followers into loyal customers. That sounds attractive and it doesn’t seem crazy, right? Email, a nurture sequence, any email campaign you’ve built out in Klaviyo, similar concept, right? So, why should inboxes that live inside social media be any different than inboxes that live inside Gmail? Similar thing, except maybe with the DMs I think potentially it gets simpler, right? We have constraints to work within there more so than with email.
And so, joining us to talk us through it, because I’ve not done this, is Liz Carter, the brains behind GROW AI. It’s a chatbot agency. They craft chat funnels for D2C brands. My understanding is this happens on Meta, on Facebook and Instagram. A real professional who’s gonna talk us through it. Liz, thanks for joining us.
Liz Carter: Thanks for having me, Kurt.
Kurt Elster: My pleasure. So, chat funnels. How did you end up attracted to chat funnels in a way where you said I want to do this full time?
Liz Carter: That’s a long story. I first started building chatbots in 2016 and I was doing it for my online fitness business at the time, and I did that for a couple of years, and it was great. There weren’t as many let’s say limitations as there are right now with Meta, and policies, and there’s a lot more restrictions now. However, I liked it a lot when I did it then. And so, my online fitness business, I ended up closing it, and then I had about a year working in a Microsoft space, and then I quit that job, and I thought about a business idea. I wanted to have my own business and I came up with this idea of opening a chatbot agency.
And I did that end of 2019 and that’s kind of… It’s a longer story that I started initially with targeting local businesses and building chatbots for them, and then eventually COVID hit, all of those businesses went out of business, and I pivoted fast towards eCommerce because that was the only way that was functioning right then. And that’s how I started. 2020, exactly three years ago, I got my very first eCommerce clients and I started building chat funnels on Facebook Messenger. Instagram wasn’t yet open to the public at that time. And I did that, I got some really good case studies. I think I was lucky enough to get let’s say my hands on some big brands and I built this for them, and then it kind of snowballed. Then Instagram DM automation was opened and made available to the public in late 2021 by Meta and now it’s just… They’re just building more functionality, so they’re pushing conversational marketing a lot.
Kurt Elster: Who are some of these eCom brands that you’re building or have built chatbots for?
Liz Carter: Yeah, so I’ve done it for Obvi, Sugarbear Hair, SNOW Teeth Whitening, Vegamour, and obviously a lot of other smaller ones, but these are the bigger names.
Kurt Elster: For sure. Yeah. I’ve heard of both of those. Yeah, SNOW gets around. They’re everywhere. And when we’re talking about chatbots, what are the tools? Years ago, I signed up for a ManyChat account and then never used it, and that was probably my first and last experience with chatbots.
Liz Carter: Yeah. It’s ManyChat. A big part of it is ManyChat. The limitation with ManyChat is that it does not have an accurate tracking system. So, when you’re building chatbots, unless you know how much revenue those chat funnels are generating, then you can’t make informed decisions. So, a piece of it is ManyChat, and as you know, ManyChat can get complex, or you can also keep it simple. We’ve developed in the last couple of years our own internal tracking system. At the base, it’s something similar using Make and some automation behind it that we’re taking ManyChat, syncing it in real time with Klaviyo, all of the other tools in the client’s tech stack, and we are able to pull revenue numbers in real time, and that allows us to build the sales logic based on that information in real time.
So, for example, if somebody starts a conversation but does not purchase, then we’re able to trigger all sorts of follow ups on them, bring them back, but we have to have that real time insight whether they’re purchasing or not.
Kurt Elster: So, the chatbot, it does… The way you’re describing it, it does sound like a Klaviyo flow. And so, for folks who are starting out, have not messed with this, break down for me when you say chatbot or chat funnel, what is this in simple terms?
Liz Carter: It’s a fully automated back-and-forth conversation between a business and an either Instagram user or a Facebook user. And it’s rule based exactly like you’re saying it. Exactly like you would build the logic in Klaviyo. If this, then that, and that can get of course very complex.
Kurt Elster: But in Klaviyo, we’re not triggering things based on people’s replies. Is this a choose your own adventure? Are they replying and that changes what it does?
Liz Carter: Yeah, so we are guiding them towards a specific path, and if that path is not what they want to do, they can take different paths or be connected with a human team, so that’s a very important piece when we build all sorts of automation like this, because sometimes people get stuck. Maybe they have a specific question or more complex question and that’s when they’re being handed over to a human team. At the same time, it set the expectation to the user how long it’s gonna take for that human team to get back to them. But it’s mainly a guided path from the point where they opt-in and they say, “I want to do this,” towards the point where they achieve that goal or they don’t achieve, pivot to a different conversation, or hand over to a human team.
Kurt Elster: For the end goal here, the primary use case, it’s trying to turn, it’s trying to nurture leads, followers, into customers. Versus say a chatbot on a Shopify store itself, we… Often, it’ll be customer service, like, “Hey, here’s your tracking information. Where’s my order?” Or time of purchase questions, what’s your return policy, that kind of thing. This is more about nurturing?
Liz Carter: Yes, so there are three goals. One is lead generation, so collecting leads off ideally every single piece of content that you put out there. You do a live video, you put up a post, you do a story, all of those could be collecting leads, and by leads I meant at least email addresses. So, put up a story, get people to opt in saying, “Comment the word eBook,” and they opt in to get a freebie, right? They comment, “eBook,” an automation fires, sends them a DM, collects their email, sends this to Klaviyo, delivers the eBook. So, that is the number one thing. You want to move all of these followers from a rented audience, which is Instagram and Facebook, onto your owned audience, which is email.
And number two is actually driving the sales. But we lead with delivering some form of freebie, some form of value up front, so that then we can potentially push them towards a sale. But customer service, as well. Just like you said.
Kurt Elster: In these examples, it’s all… It’s Facebook and Instagram. Is that a technical limitation? That’s just where you’ve seen success? Why are those two platforms where we’re working here?
Liz Carter: First of all, we have already billions of monthly active users on these platforms. Businesses can seem to reach out to and pull them into their ecosystem, which currently they’re not. They’re putting up all sorts of… Everybody’s putting up stories, posts, they’re not leveraging these audiences to the full potential that they could. They’re still a rented audience. For example, if somebody on Instagram, a lot of people look for products, check out products before they purchase on Instagram. They go to a website, if they drop off that website, the only real way to get these people back to purchase is by either triggering email abandonment recoveries if they have their email, or paying for retargeting ads, chase them around the internet, not necessarily bring them back.
Inside of a DM conversation, once you’ve captured that user in the DMs, you can set up the chatbot to follow up on them and try to understand why they’re not purchasing, potentially give them something else, maybe give them a coupon, maybe answer more questions if they have more questions, and so you have a direct line of communication with this audience whereas otherwise you don’t. And by leveraging this audience, you can instantly increase your email list, drive more sales, and at the same time give people value. So, for example, if they have a question and your chatbot can simply answer that question on the spot, you reduce wait times by 99% and instantly increase conversion by doing so.
Kurt Elster: And so, my next question was gonna be why should a Shopify merchant care about a chat funnel system, but it sounds like a big advantage here is you’ve got this social media audience and it is very much impression based. Like, “Hey, you posted a reel, you posted a story, it got X number of views. You should do that again.” That’s always the push notifications I get. It’s like I get how that helps the platform. It doesn’t have a huge benefit to me, right? And I guess it’s exciting, you get that dopamine hit of like, “Oh, so many people saw my video for two and a half seconds before swiping,” versus this, where you are encouraging engagement, getting them onto your email list, and then once they’re on the email list, okay, now that… You’re right. That’s my owned audience.
Liz Carter: Correct.
Kurt Elster: And I have potentially much higher sales rate through email campaigns. So, just being able to in an automated fashion bridge that gap between chatbots and social, or chatbots and email, sounds worthwhile. But you tell me if I got it right. Is this why we should care about chatbot funnels?
Liz Carter: Well, that’s one reason. The other reason is what we’ve seen in the data that we’ve seen so far, when you put up let’s say posts, or get people to comment, DM in specific, and you lead with value that they care about, so value that they would be interested in relationship to your brand, you’re gonna collect first of all emails from 95% of those people. They will give you the email. Whereas email collection that high, you don’t get it too often, and that’s just email collection. And then past that point, we’re looking at immediate purchase rate from a specific chat funnel. And there are two ways to look at this. When you’re deploying a chat funnel that is more lead generation oriented, like get a freebie, the purchase rate is anywhere between 7 to 12, 15%. Because we’re delivering the freebie and then we may also say, “By the way, if you’re in a buying mood today, check this out,” and then a lot of these people simply purchase.
Whereas, when we’re looking at a chat funnel that is purely sales oriented, say people opt in with the intent of buying. Let’s say claiming a discount. They can get a recommendation quiz or something along those lines. You can take that purchase rate from 15 to up to 30%. That’s as high as we’ve taken it. The 30% is not immediate purchase rate. We present people with an offer and then we have chatbot follow ups over the next few weeks and even months that start more conversations. Maybe they need more value, more education, a how-to video. Understand exactly how that product or service solves a specific problem. But purchase rate is usually the main reason.
Kurt Elster: 30% purchase rate off a campaign would… Fantastic. Exciting. That would be a win. A 15%, absolutely worthwhile. 5% or better and it’s like, “Well, why wouldn’t I do this?” Do you have examples of these chat funnels? Or the content or topic that’s going into them?
Liz Carter: The reality is that initially, you want to keep it simple. Understand why people are on your Instagram page, for example, and offering that value to them immediately. So, with when we’re looking at examples of chat funnels, some that do very, very well are recommendation quizzes where you get people to, for example, they DM you the word quiz, and that starts a conversation that asks them questions about themselves. People like to talk about themselves, about their interests, and more so when you use that data. So, if they opt in, you ask them let’s say five questions, using that data past that point to make the conversation more personal. Saying something along those lines. You know, “Hey, Kurt. You mentioned you’re looking to do this, this, this, and this. Here’s what I put together for you specifically because you have this.” And all of that is data that speaks to them. It’s very personal.
And past that point, every single follow up that the chatbot starts, let’s say this person doesn’t buy in the next couple of days. In two days, the chatbot would follow up and say, “By the way, Kurt. Here’s what else I put together for you because you mentioned you wanted a product to be fast,” and whatever you said in the conversation. So, all of these make it so much more personal than they’re usually used to. At the same time, all of that data you’re collecting inside the conversation can also be passed over to Klaviyo, and then you can use it in your Klaviyo sequence, as well, and then take that to the next level.
Because in a chat conversation, people share a lot. It’s crazy but they share a lot. And I don’t know if it’s because they feel they’re not being judged, or they know they’re engaging with an automation, but you can get a lot of data off that.
Kurt Elster: And so then, once we have that information, we can save that into a customer profile, and then we can sync that into their Klaviyo customer profile as additional properties, and so we can segment off that. We can trigger things off it. We could use it to insert dynamic variables into an email for personalization. There’s a lot that this unlocks once we have asked the question.
Liz Carter: Correct. And more so it’s this question. Is chatbot marketing better than email marketing? They’re better both, so when they work together. For example, somebody opts in through a chat funnel. You get their email. Then you ask them a few questions, present them with an offer, and also fire an email sequence, maybe 5-15 minutes later, that complements the conversation in the chatbot. Because maybe they drop off, they have something else to do, and now that email complements whatever conversation they had inside the chatbot, but you’re in front of your audience. You have two channels now. DMs and email.
Kurt Elster: And if I’m syncing this into Klaviyo, I can also sync it to a custom audience, and so I can get those retargeting ads going, as well.
Liz Carter: Oh, yeah. Yeah.
Kurt Elster: Really up the total number of impressions we got here.
Liz Carter: We didn’t get into that, but you can do a lot with custom audiences, as well. So, for example, let’s say somebody starts a quiz in the chat funnel and finishes it, but they don’t purchase. So, then you have one audience of people who started that quiz, one of people who completed, and one of people who purchased, and based on that you can play around in ads manager and put in front of them potentially retargeting as that speaks exactly to them, saying, “Hey, Kurt. We noticed you took our quiz last week. Come back and purchase.”
Kurt Elster: So, we’ve got this very interactive, proactive, automated, dynamic relationship happening that is also then making enabling other interactions, like email, and in remarketing, but also making those that much more relevant to the person. So, I totally get how this is a fabulous form of relationship building that could be very effective. Is there any other tools and tricks in there that we’re doing to build those relationships?
Liz Carter: Not really. It’s as simple as that. It’s ManyChat. It’s having a good tracking system in the backend that allows you to get real time insights into what these people are doing. And then connecting ManyChat with not just Klaviyo, like your SMS marketing software, ads manager, like we said. Any other tool that makes sense to have this type of information synced to.
Kurt Elster: All right, I’m excited. I’m interested. But I know how these things go. I get excited, I sign up, and then I don’t do anything with it. How challenging is it to set up a chat funnel system? Assuming you’ve got moderate skills, like you’ve made it this far, you’re listening to this show, you have some idea on what you’re doing. How tough is this?
Liz Carter: Well, think of it like this. There are three ways to do one thing. Either do it yourself, do it with somebody who knows what they’re doing, or you just have it done for you. Now, do it yourself is just going around the… Going on YouTube, trying out different things, you get stuck, you don’t know what to do. That’s one thing. Done for you might sound as if it’s the best way, but if somebody else does it for you, you don’t know what’s happening. You can’t control. Whereas chatbot marketing will become just like email marketing, so businesses will have in-house teams just like they have email marketing who have to know how to manage the technology.
So then, the easiest path forward, and the fastest, is getting somebody who’s done it, teach you how to do it and how to manage it. So, a hybrid done with you, let’s say version of you learning how to do it, getting somebody to help you. All right, we’re having all of these tech snags. I don’t know how to do this and how does this work? And then learning all of that and now you have a skillset that allows you to… You know, Black Friday is coming. How do I prepare my chatbot marketing department or strategy to hit Black Friday?
Kurt Elster: This is doable but obviously it’s technical and with experience it gets easier. Well, my advice to people often is try it and fail. Give yourself permission to just screw this up. Because once you’ve tried it and failed at it, you’re gonna be a lot more comfortable talking about it and hiring someone for it versus if you haven’t attempted it at all and you’re going straight to hiring someone for it, that’s a difficult conversation to have with a vendor for both of you. Because you really-
Liz Carter: Don’t know.
Kurt Elster: You don’t know. And you don’t know what you don’t know. So, are there any tools or resources you recommend to get people up to speed? To play with it?
Liz Carter: ManyChat has a few very good free courses, so on their website, and that’s the best way and the fastest way to get started. You can get started with ManyChat for free. You don’t need to pay although the Pro plan is $15 a month. And that’s the best way to start. And a lot of help inside of the ManyChat, as well, so you have templates you can get started with and just model them, and that’s the best way and fastest way to start.
Kurt Elster: Okay. So, again, using that… We compare it a lot to Klaviyo because people probably have experience with that similar to Klaviyo, where I have… They’ve got their flow library, like a series of templates. ManyChat has that, as well, so I can get started, and they’ll have some eCommerce workflows to work through?
Liz Carter: Correct. Yes.
Kurt Elster: When you were running through your examples, it sounded like it’s a series of flows and they trigger each other. How many flows should I be starting with? How many do you set up for a client?
Liz Carter: You’ll want to start with one because if you get into too many, then the logic can get complex really fast, so one. When you start with a chat funnel and you deploy one chat funnel, you will see results overnight. It’s as easy as building an automation behind let’s say a post. You go put up that post, tell people to DM you or to comment a specific keyword on that post, and then capture all of those inside of the DM conversation. So, that automation would look as simple as this. DM us the word 20 for 20% off today. People DM you, that triggers, take their email, sync it to Klaviyo, and ManyChat has an integration with Klaviyo, so it’s very easy to set up. And then at the end, simply give them the coupon. That’s it.
So, now you have their email, you give them the coupon, bam.
Kurt Elster: So, initially just keep it simple. And having just that, way better than not having it at all, because now we’re bridging. That very first thing is like, “Hey, let’s bridge that gap between social and email.”
Liz Carter: Correct. And when you’re looking… So, that’s how you start. So, that is one very simple, basic chat funnel. And then once you build more, a number of chat funnels make up a chat funnel system, which is what normally most businesses end up having. Multiple chat funnels that sit behind their DMs and fire depending on what the user is looking for. So, for example, one may be a quiz, one may be leading them to a master class, or a webinar, or booking a consult, and all of those sit behind their DMs. Maybe they have 10 chat funnels, and they rotate them. Like over the course of a week, they tell people, “Hey, DM me the word quiz,” or, “DM me the word master class,” or, “DM me this word.” One word on Monday, another one on Wednesday, another one on Friday, and that’s how they rotate and give people more value than just telling them to do one thing.
Kurt Elster: The starting conversations with chatbots, they’ll proactively DM the person. That’s where I get tripped up. It’s not like email. There’s not a subject line, right? They’re not expecting it. When I’m trying to cold DM somebody, essentially, what does that look like? How am I gonna get a response versus getting deleted?
Liz Carter: You don’t. You don’t cold DM. That goes… So, the chatbot will never cold DM anyone. That goes against Facebook’s policies. The user has to initiate that conversation. And the way that you initiate it, let’s say Instagram for example, is by telling people, one, comment on my post or reels. That is going to fire an automation. Comment on my live. That is going to fire an automation. Reply to my stories, that is going to fire an automation. Mention. When people mention you in their stories, that could trigger an automation, as well.
And then for Messenger, you have a little bit more functionality. For example, you can use a Messenger link, add it to the website. You can also run chatbot ads, whether they are Messenger ads, or Instagram DM ads, but the user has to initially initiate that conversation. You can’t send out cold DMs to people with the automation.
Kurt Elster: So, no matter what, it’s reacting to the first step taken by the customer, the user.
Liz Carter: Correct.
Kurt Elster: And this content usually is you’re giving them something, whether that’s a coupon code, an eBook, a download, a quiz, so a recommendation. I’m always trying to provide some value to them, it sounds like.
Liz Carter: Ideally. Because otherwise they don’t have a good reason to start a conversation. And there’s gotta be something in it for them. So, at the point where you offer them something, it makes sense for them to take this step, and then it’s a back and forth. As long as they find that initial reason appealing enough for them to DM you in the first place, then the conversation kicks off, and then from there, get their email. And it can do a lot more, like it can get a lot more complex. For example, right now Meta is pushing the concept of DM lists, of getting people to subscribe to your DM list just like they do on your email list, and then you can broadcast at scale to all of that list.
Kurt Elster: What are the typical kinds of content that work best to engage people within these chat funnels?
Liz Carter: It’s usually some form of free value that… You know, if you have something on your website where you say, “Enter email address and get a lead magnet.” Whatever lead magnet you have right now that’s working for you, that is the best way to start inside of DMs. Just deliver that PDF, or a how-to video, or if they’re opting into… Maybe they’re opting in to register for a webinar. Get them registered via DMs. So, whatever makes sense for the brand, and whatever the brand is using in their marketing efforts right now, that is gonna be the very first thing to use inside of the DMs. Take what’s working and translate it into a conversation inside of DMs. And then past that point, you can come up with something new. But ideally, whatever’s working for you right now on other channels, take that and move it into Messenger or Instagram conversation, and that is going to at the minimum 2 to 3X the performance that you’re seeing on the other channels.
So, if the opt-in rate, collection rate, that is going to at least 2 to 3X inside of the DMs.
Kurt Elster: 2 to 3X. I mean, that’s gonna pay dividends.
Sound Board:
Kurt Elster: Chatbots have been around a while. Really, ManyChat has been around a while. You’ve been doing this, playing with ManyChat seven years. Is there a risk of saturation? Of copycats? Really, do I have to do something to worry about standing out? Is it enough to just take your advice and have the chatbot?
Liz Carter: If you have a business that sells a product, or a service, or a course that people want, you’re already standing out because you’re unique in the sense that you have something that people already want. It’s just a matter of changing or adjusting the delivery of that service or product and meeting people on the platforms that they’re at. Because when we look at ManyChat, yes, ManyChat has been around since 2016. The technology is there. And so, we have the technology. We know it works. It’s just a matter of using it. And everybody’s gonna use it differently. Every business is different. Obviously, if you have a business that doesn’t work and your product is not selling, ManyChat’s not gonna sell it. So, the way that you become unique in using this technology is by having your own unique product, which you already do, or course, or service.
Kurt Elster: All right. Common challenges and mistakes. Are there a few mistakes you see Shopify merchants, D2C brands just making over and over that you want to save them from now?
Liz Carter: Well, these are mistakes that Meta has already been tackling in the last few years. So, you said you’re familiar with ManyChat, right? So, ManyChat has still a 24-hour window in which you can engage a user for free, past the point where they’ve engaged with you last. Now, years ago you could also broadcast to them at least once outside of that 24-hour window. And so, businesses abused that. They looked at this channel as just another promotional channel and they blasted all of them, and of course that didn’t work out. Meta took that feature away in 2020. And then after 2020, you could not possibly reengage those people at all. It became a nightmare.
And then they started playing around with what’s the functionality that we can allow businesses to use to still reengage these users but be mindful of their inboxes? And so, in late 2020, they brought in the functionality of a one-time authentication. You can collect people’s permission once about a specific topic and then broadcast to them about that topic. But also, you have to broadcast to them only about the topic that they gave you permission. Otherwise, your page got restricted from sending messages, blocked, and so on.
And that was a good start. However, you could only broadcast to them once. So, that again made it very difficult, because after you send that broadcast then you have to go back and collect another permission, broadcast again, so it was a bit of a… It didn’t work out. And so, later they introduced an NTN notification. N stood for three types of broadcast permission. You could either collect permission to broadcast to them daily for six months, weekly for nine months, or monthly for 12 months, and that’s where it all started. That was a bit of a beta test to see what works best. And the daily for six months worked best, and now that is what we have.
So, for Messenger, and this is already open to the public. Instagram is in closed beta. We have been using it ever since it was in closed beta, since last year, but you have to apply via ManyChat to Meta. It’s a bit of an approval process. But now you can get people’s permission once to broadcast to them daily for six months. After six months, Meta sends an automated message. Hey, do you want to opt out or do you want to stay opted in? And now that makes businesses and whoever uses the feature be a lot more mindful of people’s inboxes, what they want to see, and what we normally say is that ideally lead with as much value as possible. 80% ideally give people value, and then 20% sell the product or the service.
Kurt Elster: And I think if someone were listening, they were to call into the show and have push back, it would be something like, “Oh, well, I don’t like talking to bots. I want to talk to a real person and my customers are just like me always, and so they only want to talk to a real person.” How do you handle a situation where your customers prefer human interaction over a chatbot? I assume you’ll get a few who just don’t use the chatbot? What’s the thinking there?
Liz Carter: I hear this a lot and I think that results that we’ve been getting say the contrary. So, people do… People want to solve problems fast, not wait, and get value fast, and make the most of their time when they engage with their favorite brands online. And at the same time, we’re not talking about 100% automation, so when we build a chat funnel like this, we always keep in mind the scenario where people want to speak to a human, and that should be very easy for them to achieve. If they want to speak to a human, the chatbot will immediately hand them off to a human team, set the expectation to the user, “Hey, I’m handing you over to my human team. It’s gonna take them about an hour to get back to you. In the meantime, you can continue browsing through our chatbot or leave the chat if you want.” That bot-to-human handover needs to be very well put together whether the brand is handling ManyChat, or they have Gorgias, or Zendesk, and then tickets will be opened in there. It has to be very seamless for both the user and the team who handles all of this.
So, again, we’re not talking about 100% automation. We’re talking about about 80 to 90% chatbot automation and then the rest of 10 to 15% is gonna be a human stepping in to solve the more complex situations.
Kurt Elster: Okay. That seems reasonable. Looking toward the future, this isn’t AI. It’s not generative AI. It’s not ChatGPT. You’re giving it copy, scripts and templates to work from. Is there an option for generative AI, like ChatGPT? Can I make this thing just start talking to people?
Liz Carter: Absolutely. And we’ve been doing AI, as well, for the last three years, but this was just more appealing to the brands that we work with. So, for some of the brands, we have implemented AI on top of Facebook Messenger, for example, and we’ve done it with the Dataflow, which is Google’s AI, and it works integrated with Facebook Messenger, and the way that it works is that we deploy a chat funnel and then from that point on we start collecting inside of a Google Sheet all of missed user inputs. Whatever people are asking the bot, we collect all of that data, and we collect it for about let’s say three months, just so we have enough data, and then past that point we take that data and model it into topics. Let’s say of that data, we’re gonna take out 10 to maybe 20 different topics these people are inquiring about, and then inside of Dataflow we build all of those topics, which we call intents, and then add all of those training phrases to each topic. And the more training phrases we give it, the better it gets at understanding in context what people are asking.
And that’s how we’ve been doing it with Dataflow. Now, ChatGPT is already in beta testing with ManyChat, and we’ve tested it. It’s not accurate enough and it’s also the matter of that you have to figure out the right prompt to stop it from firing outside of the client’s proprietary information. So-
Kurt Elster: Yeah. It’s tough-
Liz Carter: It needs to stick to a specific data.
Kurt Elster: It’s inconsistent in how it behaves I’ve found with ChatGPT. The idea is-
Liz Carter: For now.
Kurt Elster: 6 months ago, it was real exciting. Today, playing with it, you start to see the limitations, and one of them is just like if you wanted to use it as part of an integration in something else, trying to get it to act in a way that’s consistent enough that you’re comfortable just deploying it to the general public, I don’t… I think there’s many instances where a brand or company may say, “We’re just not comfortable with this thing.”
Liz Carter: Yeah. I agree.
Kurt Elster: As much hype as they’ve gotten, it’s not perfect for everything. And in this case, the idea of just like letting it talk to customers willy nilly, maybe not where I’d start.
Liz Carter: Yeah, so we are… I feel the same. However, I also feel like it’s going to evolve fast.
Kurt Elster: For sure.
Liz Carter: And you know, when it does, we’ll be there to take it over.
Kurt Elster: So, currently Facebook and Meta are who play nice with ManyChat. Are we gonna see other platforms on here? Like when’s a TikTok integration gonna happen?
Liz Carter: Oh, I don’t know. I don’t know about TikTok. Meta has WhatsApp automation, as well. I wouldn’t be surprised that they look at this for Threads, as well, in the future. In terms of other platforms, I don’t know what to say about TikTok, about anything else. We’ve been mainly focused on Meta platforms so far. And so far, they have Facebook Messenger, Instagram DM, and WhatsApp automation.
Kurt Elster: Understandable. All right, lastly, any parting advice for folks who are about to get started with chatbots or trying to stay ahead with chatbots?
Liz Carter: I guess the advice would be to keep it simple. Start simple. And get the wins early on. Just like I said, focus on building a very simple use case and determine what you want that chat funnel to bring you. Is it emails? Is it sales? Is it automating customer support? Focus on one, get that first win under your belt, and then build on top of that. I mean, I know we spoke about all sorts of asking people questions, and understanding interests, and then building. If you start and go straight to that, and start doing that, it’s probably gonna be overwhelming. It’s not gonna work out. So, ideally start-
Kurt Elster: Yeah. Walk before you can run.
Liz Carter: Correct.
Kurt Elster: Where can people go to learn more about you?
Liz Carter: They can look me up on LinkedIn or Twitter, Liziana Carter, or even Instagram. Everywhere. I’ve started producing content just last week, organic content, so I’m hoping to be everywhere now. Or go to our website, GetGrowAI dot com.
Kurt Elster: Get Grow AI dot com. I’ll throw that in the show notes along with your LinkedIn link and I’ll put the ManyChat link in there, as well. Liz Carter, thank you so much. This has been fantastic. I gotta play with ManyChat again. I really do. It’s been years.
Liz Carter: You have to.
Kurt Elster: I love automation. I love marketing automation. This seems like a thing that I would really geek out over.
Liz Carter: I think so.
Kurt Elster: To our listeners, thank you for listening. Don’t forget to subscribe. Leave one of those five star reviews on iTunes. We live and die by that. And join our Facebook group, The Unofficial Shopify Podcast Insiders if you haven’t, and thanks again for listening. I appreciate you.
Sound Board: