Creating Human with AI
with Jon Tucker, CEO of HelpFlow
In this episode of Ecommerce AI Mastermind by HelpFlow, Jon Tucker explores the integration of AI in creating human-like customer experiences. He discusses the delicate balance between automation and personalized interactions, showing how AI can enhance, rather than replace, the human touch in ecommerce.
Key topics include the use of AI to mimic human empathy, using data to inform personalized service, and the role of AI in supporting rather than overtaking human teams. Jon also covers how businesses can scale with AI while maintaining a personal connection with customers.
Ecommerce AI Mastermind - Creating Human with AI
with Jon Tucker, CEO of HelpFlow
In the fast-evolving world of ecommerce, AI is transforming how businesses interact with their customers. In this episode, Jon Tucker dives into the concept of "Creating Human with AI"—the idea that AI can enable businesses to deliver more personalized and human-like customer experiences at scale.
Jon opens by discussing the need for a balance between automation and empathy. He emphasizes that AI shouldn’t be seen as a replacement for human interaction, but rather as a tool to enhance it. For example, chatbots and automated workflows can handle basic inquiries, freeing up human agents to focus on more complex, emotionally-driven customer needs.
The conversation also touches on how businesses can use AI-driven data insights to offer personalized services that feel distinctly human. By analyzing past customer interactions, AI can predict needs, preferences, and concerns, allowing companies to tailor their messaging and support to each customer on an individual level.
Jon continues by sharing real-world examples of companies leveraging AI to augment their customer service teams, creating a more efficient but still human-centered experience. He explains how AI can assist in scaling these personalized interactions, especially for fast-growing ecommerce brands, without compromising on the quality of customer care.
As the episode wraps up, Jon shares best practices for implementing AI in customer service, such as identifying tasks that benefit most from automation and ensuring human oversight remains a key part of the process. He stresses that AI should empower human teams rather than replace them, driving better outcomes for both businesses and customers alike.
In This Episode, We Discussed:
[00:01] Introduction to the episode and topic
[02:20] The evolution of AI in customer service
[05:12] How AI can create more human-like customer interactions
[09:40] Balancing automation and empathy in ecommerce
[13:05] Using AI-driven insights to personalize customer experiences
[17:15] Examples of ecommerce companies successfully using AI to enhance service
[21:33] Scaling with AI while maintaining personal connections
[26:42] Key strategies for integrating AI without losing the human touch
[30:45] Final thoughts and takeaways from Jon on AI’s role in customer experience
[34:18] Wrap-up and call to action
Hey everybody, John Tucker here from Help Flow. We are super excited to dig in. I've got my team running some things on the back end, so I'll be kind of watching a couple chats on Slack on the other screen, but we're gonna get started. We're gonna give people a couple more minutes to get in.
We'll start at about the five minute mark, but a good initial thing I thought would be good is to kind of start with some questions. I believe everybody should be able to see the comments. Angel on my team, keep me honest in Slack. Let me know if everybody else can see the comments.
for example I see jetty's comment right um but what I want to understand first question is for everybody attending um on a scale of one to ten ten being like the most um how deep are you in the ai space right now like how how uh How much research have you done in the space? How much are you paying attention to what's going on in AI? How savvy do you consider yourself regarding AI? Just give me kind of a 1 to 10 of how deep you are in the AI rabbit hole.
That'll really help me understand how to frame some of the things that we talk about. So if you can put that into the comments, that would be awesome. I'm going to drop this in here as well, so it'll show up in the comments as well. So, Sam, we got you at a four.
Matt's at a five. Jetty's at a five. Cool. This will be a good kind of middle ground, I think.
We'll keep dropping that in. But one of the things I wanted to start with as we're kind of waiting for people to come in and also hearing more about like everybody's experience is just some of the examples of crazy AI stuff that's happening right now. If you haven't seen the GPT-4-0 demo, GPT-4-0 demo from OpenAI a couple of weeks ago, super, super powerful. It really shows how the GPT models can now, or that specific model, it can interpret emotion.
Like if it was watching this video, it would like interpret like how I'm communicating from an emotional level. It's pretty wild. It can... demonstrate emotion. They called it something more fancy than demonstrate, but it can literally like essentially have a conversation with you in audio and change its tone and change its emotion based on like what the topic is, right?
Which is wild in some of the demos to see and a whole bunch of other things. There's a ton of other things that can be done based on what's in the demo, but it is, in my opinion, it's like the iPhone moment for the AI industry where it's like, oh my God, like if that's possible, if these things are possible, and there's a bunch of other things in that demo, Then what else is possible? Right. The demo was extremely consumer focused, I would say.
So it was kind of like hokey examples of like, look, you can talk to GPT or, you know, she's going to speak French and I'm going to speak English and it's going to interpret like as we're talking, look how cool this is. Um, so it was very consumer focused, um, for, for a number of reasons I won't go into, but all of it's available on the AI, on the API, on the backend. So you can build tools using all these things and it's, it's wild to see what's possible. So if you haven't seen it, Tom, I know you said you saw it.
Um, you know, Jetty, Matt, Sam, anybody else that's watching this, not commenting, um, make sure that you see that demo GPT 4.0 demo. I would recommend watching the actual demo is like 20 minutes. um it was mind-blowing to watch um and there's a lot of other things happening in the space right now too so um definitely spend some time digging in um part of what we want to do today is help you to understand how the tools work um but I would just encourage everybody to pay attention to what's going on because I think it's um I'm 100 confident that a year or two from now, definitely five years from now, everyone's going to look back at this time period is like, oh my God, like that's when it all changed, right?
And I think we're literally in that window right now. I think 2024 in the beginning of 25 is when we're going to start to see some extremely impactful use cases for AI. For today though, What we're going to talk about, we're going to talk about a lot of things related to AI. But what I want to talk about today is how to create a human with AI.
And I know that sounds kind of crazy, but that's how we think of AI at Help Flow. And that's honestly how I recommend everybody think of AI, which I know is kind of scary. Like, what do you mean it's software? It's not a human, right?
But when you think of it like a human, it opens up your mind into all these different things that are possible. And so that's really what I want to walk you guys through today. Angel or Ray, if you could let me know in Slack at the sound video, everything's good to go. I'm going to jump into the full presentation, but just want to make sure everything's ready to go.
Thank you very much, Angel. So AI is new. AI is a super new space. There is so many different AI tools that are out there and it can get confusing, right?
Like it is just such a new space that so many things are happening and so many things are happening quickly. And there's also tons of tools, right? So you have, you know, ChatGPT, you've got a whole bunch of different customer service tools related to AI. You've got, you know, Zendesk launching their AI that's already out.
Gorgias has announced theirs and it's kind of out, it's coming out soon. They've announced it, so that's not backroom information, but everybody's launching their own stuff. it can be incredibly challenging to actually produce value with these tools. And the reason why is it's moving so fast, it's hard to kind of understand what this stuff means and how it works, and there's always new tools coming out.
So there's also that fatigue or that tool confusion where it's like, okay, I can use AI in my customer service help desk, great. Do I use Zendesk? Do I use Gorgias AI? Do I use Sienna and Digital Genius that sits on top of these tools?
What do I use? What we want to do today is I want to get you to the point so that when you understand the tools that are out there and how they function and how you can connect different tools, then you will start to either create your own powerful tools or you'll start to understand how these different tools work. And so that's what I want to get you through today is like an understanding of how these tools actually function. And one more question in the comments.
And I know we don't have everybody like in the comments because of the way that the StreamYard system works. But for the people that are able to comment in the main system here, let me know what help desk you guys are on. Are you on Gorgias? Are you on Zendesk?
Are you on email inbox? What are you guys on? Because that'll help me kind of adapt some of the tool names that I talk about. But my goal is to get you to the point where you understand things today.
And at Help Flow, where this context comes from, we've been in business for 10 years, which is a long time in internet land, I guess. But we've been in business for a long time. We've always been tech focused. So we've been tech focused from the beginning.
on how to apply tech in the customer service space, how to use different tools to be able to provide good customer service. We did things that kind of looked and felt like AI before AI was a thing. I'm not saying we were like pioneers and like creating AI out of thin air, but we've always been just thinking technical first, right? And in the last three years, we've gone heavily into AI, like very, very deep into AI mode.
And an example of this, just so you guys understand where Help Flow sits in the marketplace, the BPO space, the customer service agency space, the model of staffing customer service agents is already massively changing regarding AI. And about three-ish years from now, a lot of customer service agencies are not going to be around because they're in the business of selling humans. That's essentially what they do. I know that sounds a little blunt, but they are in the business of staffing humans.
We have been extremely focused over the last three years on understanding how does AI change that? Where does it go? What's it going to turn into? And frankly, I was having a conversation with a buddy the other day, and I said these words, which I think applies here.
I told him, I said, once I got over that weird feeling in the last couple of years of like, this is literally going to change my entire business. Like my business in its current form is not going to be around in three years. And I do firmly believe that. Once I got over like the I guess you'd call it fear.
It wasn't fear, but it was this sense of like, Oh my gosh, like the whole thing's going to change. Once I got over that, there's some warped thing in my mind as an entrepreneur, as a CEO, or maybe just the person I am where I was like, this is wild to be able to like be a leader in this type of environment and be at the forefront of it. And like, yes, my industry is being completely disrupted, but it's also like strangely exciting. And so that's where like a lot of the energy is going to come from today of seeing like how we think about this.
Like that is literally how I think about it as my industry is completely being changed. And we are at the forefront of that, and that's super exciting to me. Other agencies are not going to be around in a number of years precisely because of that, because they're still in scarcity mode and fear mode. But that's just how massive technology shifts work.
So that's how I want you to understand the frame that I'm coming into this as and how I think of AI. But again, like we're very tech focused, very focused. We work with a ton of stores, hundreds of stores. We produce almost a billion or half a billion dollars in CX revenue for our clients.
Like very, very involved in the space, I guess I would say. And getting to this, what I want you to think of is I don't want you to think of AI like a software. I don't want you to think of it like technology. I don't want you to think of it as like this crazy innovative thing that's happening in the world.
Just think of it like a human. Like think of AI as basically a human. And this should technically say think of AI as a human, not open AI as a human. There's a typo there.
But think of AI as a human. And here's what I mean. You have a role that needs to be done or a goal that needs to be done. Like you have a role such as, you know, customer service agent or a QA reviewer.
That's actually the role I'm going to use today is like a human that analyzes customer service tickets to understand if it was done well. But basically you have a role that needs to be done. you define what's actually needed, right, for that role. And some of that's going to be objective, right, of rate the chat according to this, you know, criteria.
But some of it's going to be subjective, right? Like you are a QA specialist, evaluate if the customer service agent did a good job on this ticket. Good job is like extremely subjective, but somebody that's been in customer service for 10 years and analyzed thousands of tickets knows exactly what that means, right? So AI knows things at that level.
And it has the experience for the role. That is the most important part that like will break your mind open with AI is it already knows how to do. I don't want to say everything because I'm sure there's things it doesn't know how to do, but any role that you can name and hire for. at a fairly narrow level, it knows how to do.
And I'm also going to touch on later, I might touch on this in the Q&A, there are tools now to connect different AI. So I'm going to be talking about like, think of this human specialist. All they do is review tickets, right? And evaluate how good they are.
That's all they do. I'm going to talk about that today. AI is already at the point where it can handle that. Once you get deep in the weeds on AI and realize, oh, I can connect this one to this one, to this one, to this one, and then have an AI that manages all of them, that point you get to the point where like you literally have a team of of ai running the whole business right so it I do think it will get to that point at a certain level um but I'm gonna focus just on that one human so I want you to think of ai as a human it already has experience for this role and then like I mentioned you can connect multiple ai together but I'm gonna I'm gonna pull this back from going too deep into that unless it's in um the q a if you guys want to dig into it so What is most important for this particular session is think of AI as a human.
That's it. And what we're trying to do is we're trying to create a human. So let me explain the pieces that are involved. And now this is supposed to say OpenAI.
So OpenAI Assistant API is basically the human. It is the specialist role or the human that you are building. And so let me explain what this means. And I'll go a little bit slow on some of these parts.
So I'm sure that everybody here has used chat GPT, right? Like a chat GPT is AI to many people, but chat GPT is powerful. It can do a lot of things. Some people get a little bit discouraged when they use it because they're like, oh, it's not that good.
That's usually because they don't know how to prompt it or how to have discussions with it or like what it's capable of. But there's chat GPT. That's one thing. Okay.
I'm going to assume that everybody knows what that is. So I'm not going to spend time talking about it. Then there is a custom GPT. So a custom GPT is a tool within ChatGPT that enables you to create a custom version of ChatGPT that you can chat with.
For example, you can define specific instructions that it will always use. So every time you view that chat, I guess we'll call it, every time you go to that chat, it is pre-programmed in whatever way you want it to be programmed. For example, it could be pre-programmed as a customer service manager to help on escalated tickets and manage things according to these rules. And then also you can add files to it to say, use these as references, right?
So you could use like FAQs, you could upload a list of FAQs to the custom GPT. So then essentially what you're able to do is take the power of chat GPT give it custom instructions and custom files, and now it essentially is your own chat GPT that is trained to do exactly what you want it to do. That is a custom GPT. That is accessed in the same interface as chat GPT.
You just chat with it, that's it. Technically, you can do some slightly other things, but I won't go into that today. So a custom GPT is a custom version of chat GPT. An OpenAI Assistant is the same exact thing, but you can access it from other tools programmatically.
So you can hit it with a API request is essentially what it is. I don't think API, the term really matters today, but essentially what I'm saying is it's exactly what I just described, but you can access it programmatically. What that means, and I don't think I go into this in the next one, but what that means is you could have it as part of a workflow in your business. So for example, this is a different example, but I'll use it anyway.
Every time an invoice comes into the inbox, When there's an invoice in the inbox, it goes to the API, reads the invoice, interprets the invoice like a bookkeeping specialist, and then comes back to the CRM or the email system, we'll call it a CRM in this case, with all the invoice information, okay? And then the CRM can use that to create a task in the system for finance approval or whatever needs to be done, right? The human in that case is like the accounts payable specialist that's reviewing the invoice and coding it in, right? You've now replaced that person.
So what I want you to understand is that the OpenAI Assistant API is basically your own custom version of ChatGPT, but it can be accessed with an API. And so basically what you do to create the human in this particular case is you configure the open ai assist and you configure it the same way you do a custom gpt get deeper in the weeds during um during the q a but you configure that with the instructions that I mentioned so for example in our case act as a qa specialist this would literally be what's in the instruction this is how you program chat gbt it's all plain text you would say in the instruction set act as a qa specialist for a customer service team and you are in charge of ensuring that the customer service agent handled the ticket well you will be given a customer service ticket and you need to rate it according to your experience as a customer service manager and according to these rating guidelines that are in your files you've been given a file like a reading rubric that's also how we want it to rate the output that I want is xyz I want a rating of the ticket according to the rating guidelines I want um a summary of how the agent could have done better and whatever else you want.
You tell it what you want. Then you can also give it examples of output to say, these are actual examples of what I want. And then from there, you test and tweak it and see how it works. So in the interface, and we're going to give you a resource at the end of this section or session that will show you literally like how to build this and how it works.
But basically, then you just test and tweak it. So you give it an actual ticket and see how it performs, tweak the instructions, tweak the files, do it again. Do that over and over until you get to the point where it works really well. And then you have now created a human in the OpenAI Assistant API that does QA specialist work, right?
So you've now created the human. This is probably the most important part of the session of basically how to use the OpenAI Assistant API to create a human. So if there's questions on this, just drop them into the comments and we can go through that. But the key thing I want you to understand is essentially you're using the OpenAI Assistant API to create, in our case, a QA specialist who is really good at reviewing a ticket and rating it according to all these guidelines, right?
And the output is exactly what you want. So now you've created a human. do you onboard that human into your team right like just like when you hire somebody that's a qa specialist like you gotta bring them into the team you gotta give them accounts you gotta give them tickets you gotta introduce them to everybody and you know do the company meeting or whatever um how do you do that with ai right so I'm gonna walk you through that now so an open ai assistant can be sent data do the task and send back the result so slow down on this again For example, when there's a help desk ticket that you want to do a QA round on or a review on, it needs to be sent up to the OpenAI Assistant API.
That's going to do all the magic, and then it's going to respond back with the output, the ratings, or whatever you want it. So it needs to be sent the data, and then it'll send back the results. The easiest way to do this is doing it in Zapier. Zapier is essentially a tool that enables you to move data between different software via APIs.
Again, the word API doesn't matter. That's technically how it's done, without needing to know how to work with APIs. APIs can be pretty complex to work with. But Zapier enables you to do it in a way where you say, OK, when a ticket is done in Zendesk, send that entire ticket content over to the OpenAI Assistant API.
It'll do all this stuff. When OpenAI Assistant API is done doing all this stuff, send it back to that same ticket in Zendesk and add it as a note. That's all technically done programmatically with an API. but you don't actually need to know how to use an API if you do it within Zapier.
Zapier is the tool I'm going to recommend today. But let me give you a very specific example so you can see the entire thing. In this particular case, what I'm talking about doing is creating a workflow that essentially does QA as I described. It rates the ticket, it gives commentary on what the agent could have done better, does all those different things, and it does it for all tickets processed in Zendesk.
This is exactly how it would work. In Zapier, what you would configure is, and again, remember we have this resource at the end we're going to share to you that will like walk you through an actual Zapier workflow I built. And then it goes into an open AI system. Like it shows you the backend of all these things.
We wanted to be careful not to get too nerdy on the actual webinar here because like we could go on forever. But I want to help you understand how it works at a high level. So what would happen is that when a ticket ends in Zendesk or whatever help desk you're in, that would kick off the Zapier workflow. So that's like the first step in the Zapier workflow.
The ticket data is then sent to Zapier as a Zap. So technically Zapier grabs all that stuff from the ticket, pulls it into the Zapier workflow. Then in that Zapier workflow, what it's doing is it's taking the ticket, it's sending it up to OpenAI, and then it's receiving the results back. And then the last step is you're given the QA insights about the ticket or the agent, and that can be sent somewhere.
via Zapier. The last step in the Zap is, you know, put it into a spreadsheet or send it as a Slack message or, you know, drop it back in as a note on that Zendesk ticket where it came from, right? Or do all of those things. Or you can technically get crazier and say, if the rating is below a five, for example, five out of 10 or however you're going to rate it, send it to the CEO also to fire the agent, right?
Like you could do a whole bunch of different things. Um, but basically the Zapier workflow manages the whole thing. Um, and so again, like technically this is done, uh, through API systems, but it enables you, uh, Zapier enables you to do it without needing to bother with an API. Um, let me be crystal clear on this part too.
Um, so Sam, you mentioned that, uh, you know, it's cheaper to use make a hundred percent like make is a make.com is for the sake of this discussion, a competitor Zapier, right? Zapier has been around forever. There's now make.com.
There is, I forget all the names we use a couple of different ones. Um, this is how I would think about this in terms of like, it's cheaper to use make.com. The right way to build high-scale AI tools like this is to build it on Zapier or build it on Make, like build it on whatever platform you're comfortable with. But build it on that platform, prove that it works, and then hire a developer to say, please do this without Zapier, right?
Because the stuff in this case that Zapier is doing is very, very simple to do for a developer that knows how to run APIs and move data around systems. The reason why Zapier has a gigantic business is because they make it very easy to do that without a developer, which is great for a proof of concept and frankly, like, you know, scaled up tools that a business uses. But one of the things we found in the last, probably the last four years, five years, I don't remember, somewhere into the business, we realized like, oh my God, we're spending so much on Zapier. And technically what changed things for us is they changed their pricing model and it kind of pissed me off.
So I started paying attention to the cost a little bit more. And I said, we could just replicate a bunch of these things with our team. So we actually started replicating the top 80% of the volume in Zapier. We did an 80-20 on that volume and said, hey, we can just replace all these things.
And we've gradually started to do that. So we're on a much smaller Zapier plan now. But I actually don't recommend thinking like that yet. Just focus on getting it to work.
See the power of it. And then once you say, OK, this would be really powerful if we did it on all tickets or these other use cases, then you hire a developer to replicate it. The first step of the ticket ends in help desk is gonna be the same. The last step of the insight ends up in a spreadsheet or Slack or wherever you send it, that's gonna be the same too.
Everything in the middle is gonna be different. The developer is gonna figure out how to run all that, but your cost will be, 70% lower by building it yourself in terms of API costs. So let me zoom out here for a second. So essentially what I'm saying, and then I want to kind of open it up to Q&A.
We're not going to go super long today. But essentially what I'm saying here is in this use case, we have built an actual human, the QA specialist, to review all tickets that happen and to provide insights on what happened on that ticket and what the agent needed to improve. That is what we built in this case. When you realize you can build a human like this, then you start to realize how you could connect all those humans together.
So what we have done at Help Flow, remember we run customer service for tons of brands. And a big focus of what Help Flow does is like, yes, we provide agents. A lot of customer service agencies provide agents. What we do is we... provide like director of customer service level value and we provide the agents and then everything in between.
Right. So like this exact thing we literally do for clients because it's part of what we want to do is we want to say, look, what are all the insights from these tickets? What could be done better? How could you use it in your product development roadmap like those types of things?
And we do that for our clients. But what I want you guys to really realize and then I'm going to kind of open it up to questions and just shift because I feel like we're getting we're getting deep in the weeds. What I want you to realize is by creating a single human like this that does QA specialist work, it's going to add a lot of value to your business. In this case, if you want to do QA, but you could apply this to a ton of different use cases.
Where things become insane is when you chain these together. So I'm going to give you one chain link so you can really see the power of this. In this case, at the end of the slide deck over here, I think you can see, yeah, you can see my mouse. So this last step over here is the creation of insight about the ticket and what the agent could have done better, right?
So the QA specialist is creating that, in this case, the QA AI specialist. What we have done is we then take that output and we give it to a training specialist, right? So because the QA said, here's everything the agent needed to have done better, we then pass that into another OpenAI assistant that says, what type of, uh, it essentially creates training material on that skillset. So it creates training material on that skillset to say, okay, the agent needs to improve on these types of things.
So it will literally create the coaching material for the actual agent to then go through. Right. So it creates the training material for the agent to then go through. So the agent can, can be improved, right?
So now it's the QA specialist and it is the trainer, right? technically you can get crazier and crazier. You can go all the way down the line and say, okay, does the trainer now communicate that to the agent? Or could you create a, you know, a coach bot essentially that the agent interacts with to be coached on those things, right?
Like you can chain this out across an entire business. And that's literally what we're doing at Hubflow is like creating human specialists, essentially chaining them together and getting to the point where we have like entire teams that are literally AI that work together and That's why I'm so bullish on AI technology is there's just levels to this game that I don't think the world actually understands yet. And I think there's going to be a lot of new opportunities. But zooming out so we don't get too deep into the weeds because I want to make sure I don't hurt anybody's mind.
Think of AI as a human. Just think of AI as a human, as a specialist level human. And then start by building custom GPTs. to test what you want to do.
Take a ticket, analyze it from a number of different perspectives. Or if you're doing product development work, maybe you take a list of product reviews from a competitor, drop it into a chat GPT, and it gives you ideas of how to message your product better or something like that. There's a whole bunch of things you can do. But get comfortable building custom GPTs.
then get comfortable turning those into OpenAI assistance, and get to the point where you can test a simple process using Zapier to see it in action. Once you get to the point where you create a specialist, essentially, a human as AI, and then you see the whole thing work in a Zapier workflow where you push a thing in, or you tag a ticket in your help desk, And then magically some like super insightful things appears in a Google doc or a Google sheet or a Slack message or wherever you want it to be. Like when you see that happen for the first time, that I think is where your whole mindset will shift to like, oh my God, like AI could do tons more than like what people talk about, right?
So that's what I think would be best for you guys in terms of next steps. We did put together a resource. You can see the link on this screen. Let me see if I can put this, I can't easily put it into the chat, but you guys can see it here.
We put this together, it's essentially a video walkthrough of me configuring an OpenAI Assistant, configuring the Zapier and then literally seeing it run. It is not like super shiny in terms of like, look at this course and all these different things or something. Like it's pretty direct of like, look, I'm gonna build it now and now you'll see how it works. Um, but I think that if you're interested in this stuff, I think that'll be the next step for you guys to see how this can work.
Um, so definitely check that out. Angel and the team put it into the, um, the chat box, at least that's here in stream yard. Um, you can see it on the screen. It's ready to go.
Um, check that out. I think it'll spark a lot more ideas for you. Um, I'm happy to spend a couple minutes going into QA. Um, Angel, if you want to field some of those, um, in Slack or, uh, you know, in the system here, I'm happy to dig in.
I'm going to check the private chat as well. Cause I see some things there. Um, Yes, we can crank through some questions as they come in. But as far as how we can help at Help Flow, I do want to share a little bit of that.
At a high level, we run customer service operations for hundreds of brands. What that means is, yes, we provide the customer service agents, but we provide that entire department support, like I mentioned, right? So if you are running customer service yourself in-house right now with a couple agents or something like that, Where we would help in that particular case is basically going through your entire help desk system and the way you do customer service and say, look, here's all the things that can be improved. Here's the right way to structure it.
Here's how to layer in automation on some parts. Here's what you're missing from a infrastructure perspective. So we would do an audit and then we might help you to actually build that up. Right.
And so we can help if you're already running customer service yourself, we can help you have a better infrastructure. Right. Obviously, we can also help run your customer service for you, right? So if you're thinking of getting outside help for customer service, we can help with that.
What I will also say, and I will say it very bluntly and very directly, if you are already working with a customer service agency, we should have a conversation, not because I want to sell you or I want my team to sell you on working with us, but I want you to understand something very specific that is happening in the market right now. And I know this because I'm very involved with a lot of people in the industry. I'll just say it like that. Your customer service agency It's probably scared of AI, but it might not actually be that simple where it's like, oh, they're just scared of it, right?
They might actually understand AI at a very deep level, right? Kind of like we do, but they are not incentivized to take your 10 agent team and run it with six agents. They're not incentivized to do that because you probably pay them per agent and we sell our services per agent also. So I want to be crystal clear, like our model is the same, but our mindset is not.
Our mindset is not the same. So if your customer service agency has 10 agents, they can probably do the work with six, but they will never tell you that. And you are probably not far along on AI land enough to say, how do I take it from 10 to six, right? So it creates this dynamic with the agency and the client like this, where they're disincentivized to do what is best for you as a brand.
We take a different approach. We are literally out in the marketplace, having conversations with brands that have 20 agents, 30 agents, 15 agents, some number. So a very typical situation for us is we come in and we say, look, you got 20 agents. We're running with this agency or that agency.
Great. Right. Let's run an audit. Let's run an audit and see how things are working.
And every single time we come back and we say, you don't need number one, you don't need 20 agents now. Anyway, you really only need 17, like literally based on the work they're doing. They've sold you too many agents, first of all. So like we always see ways to decrease just purely based on waste.
Right. But what's more common right now, and we've been going very hard at this behind the scenes, is we'll go to a 20 agent account. We'll run the audit process. We'll say, look, if you get some of this automation stuff in place, you use AI effectively, you know what you're doing.
You can run your account at the same or better performance level in terms of CSAT and revenue and all those things. Same or better level with 12 agents. And then typically in those cases, we end up with a 12 agent account. So we will go to a big brand that has 20 agents.
the audit realize they only need 12 we'll help them get to that point we will essentially do what the agency is not going to do and then we will win that 12 agent account so if you're in a position where you're working with an outs a separate customer service agency we should have a conversation at a minimum you will get a lot of insights um of how your systems are working and how to manage that agency better um but it may lead to realizing like I think it's time to make a switch and you know obviously I'm biased but I think we can do a great job so that is one thing I would share um of how we can work together you can check it out at healthflow.com
uh for now let me dig into a couple of these questions um so have you built an agent to work with shopify data like order tracking refund exchange processes with ai um yes we have done that my view right now with where um the ai space is at is to use the tools that are already out there so a couple of the tools that we really like sienna uh sienna is a very very good tool uh digital genius is another very good tool they have uh slightly different financial models in terms of how they work uh four thoughts pretty good as well there's a number of different third-party tools all of like sienna for example can do this type of stuff that we're talking about um What is starting to happen in the e-commerce space is you have Gorgias, and then you have Sienna, Digital Genius, and the other ones, right?
So you have the foundation help desk, and then you have the third-party AIs. What is going to happen is the help desks are all gonna start offering their own AI service, which they should, and it totally makes sense. That's gonna create some competition between the third-party tools. But what I see happening is there are tools that are best for certain use cases.
For example, Sienna is going to out-compete Gorgias for certain use cases. And in those types of use cases, you should be using the Siennas. You shouldn't be using the Gorgias, right? So it's just complex.
That's any sort of disruptive market. That's what happens, right? And I've had these conversations with people at Gorgias too, right? So this is not like some competitive thing.
I'm trying to put a stake in the ground. But I am putting a stake in the ground to say, If a company, if a agency comes to you and says, you should use Gorgias, or you should use Sienna, or you should use Digital Genius, ask them what the partnership structure is between them and that tool. There is some stuff that goes on behind the scenes, and this has always happened in e-commerce, but it gets very aggressive when there's market disruption like it's happening now. It gets very disruptive or very, very aggressive, I guess I would say, that creates bias in the system.
So if an agency or if a consultant, or if somebody comes to you and says like, this one is the best way to go, um, ask them what their financial relationship is. And if you're going to sign up with them, make them put that in writing, like make whatever they said, make them put it in writing so that if it's not, if it's not accurate, they will refuse to put it in writing. Um, and that's how you know that it's bias. Make sure you avoid the bias, but to answer your question directly, Sam, yes, you can do that.
I think Sienna is a great way if you're on, um, If you're on Shopify, Sienna is a great tool to do that. I think Gorgias' tool will probably end up doing that pretty well too. I'm not exactly sure because they haven't gone deep in terms of releasing it. I think it'll go pretty well.
Usually what I find happens is point solutions like this of, you know, can an AI tool issue a refund? Like, yes, most AI tools. It's just a data problem, right? It's an integration problem with the data.
Where the tools start to differ a lot is judgment, right? So if your policy is issue a refund, no matter what, in every circumstance, like every tool is going to perform the same for you, right? But if you want it to operate like a human, for example, if this is our best customer that's ordered a bunch of stuff over the years and they tell all their friends and they get featured in publications all the time mentioning our product, like a solid customer service manager is going to know that when that person's name comes across the screen and all their order data comes across the screen and says, give this person whatever they want, right?
They get a different refund policy. And I know that might sound kind of sketchy, but that's how business works. A good AI tool can make that judgment. Not all the AI tools in the marketplace can make that judgment.
So that's what I would say on that piece. Another question that's in the chat over here, Angel sharing with me, is the data. How do you train and manage an AI? If you're going to onboard some of these AIs, essentially, how do you train the AI?
And again, think of it like a human. Like that's what I always tell people. Think of AI as a human, right? So if you're trying to train an AI to automate your customer service, think of it like a human, right?
You need to onboard it and integrate it into your systems. You need to give it access to your different systems. Like Sam mentioned, if you want it to do order tracking and refunds and all that stuff, like it needs to be integrated into Shopify, right? At a pretty deep level, or maybe integrated to loop returns or whatever the tools are.
You got to integrate it. Then you need to give it the training data, right? You need to give it the actual training data, such as your FAQs of like how to answer the questions. You need to give it that information.
You need to give it example tickets of good answers and bad answers, just like you would give a human, right? And you give it that access, right? So you need to give it all that data. Then it's trained.
Now it knows what to do, right? Then you need to monitor and manage it and coach it just like a human. right so that means reviewing the chats and qa-ing all the chats you need uh that's actually part of why we built these tools by the way this qa assistant thing that we built we qa a hundred percent of the tickets that happen in our business the reason we do that is because you do need to do a hundred percent of the tickets of a of ai tickets um you should do for um human tickets too but um that's probably more that's deeper than we need to get. What I'm trying to say though, is you need to review the AI's tickets, QA those tickets, determine how it did, and then coach the AI based on what needs to be improved, right?
So again, you're treating the whole thing like a human. That is what most of these companies get wrong when they're implementing AI is the brand will think, oh, just like put AI in, put Sienna in, put digital genius, put whatever it is. put the tool in, and then it'll just work. And it doesn't work like that.
And so brands end up like disappointed, like, oh, AI doesn't work. The tech's not there yet, right? AI 100% works. You just need to know how to integrate it and train it and monitor and manage it exactly like you would monitor and manage an agent team, right?
So that is like the high level of how you train and manage an AI. The metric, because some people have asked about like different metrics for different aspects of AI. I'm seeing that from Angel as well and Ray. And I actually don't know what you guys see.
Like, I've got comments here. I've got different comments there. So I apologize if you guys don't see these questions, but Angel's kind of consolidating them all for me. The metrics, how do you measure the effectiveness of AI?
You measure it just like a human. Everything's going to come back to like, just think of it like a human. So you measure it with customer service satisfaction, right? CSAT.
You can measure it with other things too, like, you know, revenue per ticket or those types of things. Like you should definitely track revenue when it comes to customer service. But essentially what you're doing here is you're saying, okay, what is the CSAT of those AI tickets? And the benchmark that matters is when you implement AI, those tickets need to be the same or better in terms of CSAT compared to your human team on the same types of tickets.
So tier one agents, tier one AI, CSAT needs to be the same or better. If you over AI something, the CSAT will be lower. The CSAT will drop, it'll show up in the CSAT. So that is basically what to do on the metrics side.
Another question we see a lot is, should you disclose that it's an AI, right? And I want to be crystal clear, there's some legalities around this too. So I'm not a lawyer and I don't play one on TV or anything. The AI regulation space is changing a lot and it's state by state.
It's actually pretty complex. But I'm going to speak just from a business strategy perspective. Should you disclose to your customers that they are talking to an AI, whether it's on chat or whether it's on tickets or whatever it is? Should you disclose that?
We have found that you should not disclose it because it's in the customer's best interest not to disclose it. And that's, some people are gonna look at that and be like, you have to disclose it, it's an ethical thing, right? Let me explain like two parts of this answer. What's most important is that when you disclose that it's an AI, it causes the CSAT to go down for that customer.
And it's not because it's an AI. It's because they assume that it's going to be a bad experience. And they say, take me to an agent, talk to an agent, those types of things. I'll go to the agent, which is fine.
But usually they're already a little frustrated by that point when they get to the agent because they're like, I didn't want to talk to the AI. Right. What we found is that if you if you and we go we go into some of this, this mindset stuff, we go into that in the in the downloadable resource. So definitely check that out.
I think you'll learn a lot from it. But what we found is that if you disclose that it's an A.I., the CSAT is lower than if you were to not disclose that it's an A.I., Let me think of how to word it.
This is a little hard without a visualization. If you don't disclose that it's an AI, the CSAT is better on the AI chats and the chats that get escalated up to a human. Both of those chats, they start on AI, some get all handled here, some get escalated over to a human. If you don't disclose that it's an AI, the performance on those is better, on both of those, than if you did disclose that it's an AI.
If you disclose that it's an AI, some will be done by the AI, some will get moved to human, but the CSAT on all of those will be lower. Um, that is essentially what we see the recipe for disclosure and like how to do it effectively is number one, you don't disclose it. It's an AI, but again, you gotta get the regulation, right? So double check that things change a lot.
Don't disclose it. It's an AI, make it look and feel like a human. Number two, make sure the AI performs well, like don't, don't drop the ball in terms of how to make your AI work and have a clear process for the AI to flip to a human. if it's too complicated for the AI.
So you need to proactively have it switch to a human if the AI is starting to not do a good job. When it switches, have the AI say something along the lines of like, hey, I need to check with somebody on my team. Give me a second, right? Like it literally says something like that.
And then it flips to Jack on the team or Joe or Sarah or whatever. And...
Sarah just keeps the chat going. You know, I was able to look into this and, you know, here's what I'm seeing. We should be able to get a refund tomorrow. Like, you know, let me know if you have any questions.
And then they say, great, thank you so much, Sarah. Glad Sarah's here. Right. So it just feels natural to the visitor.
But it causes the entire experience to be better for the visitor. So the key thing I want you to understand is what we found is disclosing that it is an AI does not perform well for anybody on both sides of the table. but you need to get the whole architecture of the strategy right. And again, I'm saying this every single time, check your regulatory environment, make sure the legal part's good.
Again, in the downloadable resource, it'll walk you through basically like actually building an open AI system, actually building Zapier part, all those types of things. You'll be able to see it in action. We can only go so deep in a session like this live of like actually building a tool, but everything's in that video. So definitely, definitely check it out.
I know we've got a lot of people that are watching that aren't on the don't they didn't rsvp or they're not going to get like the follow-up email or whatever so if you're watching and you didn't like sign up for this via email uh look into that uh resource and you can basically download it there but again that will walk you through some of the things that we talked about uh building it'll literally walk you through the actual building of this um angel I'm gonna get wound down let me know if there's any questions uh on our slack thread here just let me know if there's any questions um that are remaining that you want me to hit otherwise I'll get us kind of wound down but again as far as takeaways for everybody and I actually won't go to the other section but As far as takeaways, the most important thing for you to do is first, don't ignore AI.
Don't assume the tech's not good enough. I was at a conference with massive, massive companies last week in Vegas, like the agencies that do Verizon support, Walmart support, like big, big, big companies. And they're ignoring it. A lot of them are ignoring it.
A lot of them are getting slaughtered in the public markets in terms of market cap. Um, and it's because AI has kind of been around for a long time in their space and it wasn't very good. It was actually like pretty bad. Didn't work very well.
And so they're missing what changed in the last two and a half years or so. And it is fascinating to watch, you know, C-level guys at billion dollar companies say like, yeah, it'll never replace a customer service agent. And I'm just like, we've already done it, dude. Like it's already happening.
Um, I didn't say that, but that's, that's what we're realizing. And so the key thing I want you guys to do is just don't ignore what's happening. Don't assume it's not going to work. And also don't, um, don't write it off when you'd like, if you test chat, GBT and you go, oh, it's kind of good.
I don't know. I don't know how you'd use this. Like you got to go deep because it's a messy, messy space right now. The point in the future where.
Everything's easy to use and the tools are clear and you sign up for a tool and poof, it works, right? Like that's the point where there's no competitive advantage to using it. You're still going to have to use it at that point. You're going to get left behind or get crushed, but there's not like you're not on the cutting edge if it's easy.
I think that's the key takeaway I would say is you're not on the cutting edge if it's easy. right now if you start to understand how these tools work you can do some insane things in your business um and that is that is what we're trying to do at help flow both for our clients on the customer service side but also for ourselves we've got some insane stuff that operates behind the scenes using ai just because we're so deep in it so check out the resource dig into it it'll walk you through um like how to build some of this stuff um you can find more about us at helpflow.com And again, for those of you that joined halfway through kind of this Q&A part of the session, if you're working with a customer service agency that is doing customer service for you, they are almost certainly not investing enough in AI.
Even if they understand AI, they are not incentivized to sell you less agents. If they're running your account with 10 agents, with AI, you could probably do it with six. They're not going to tell you that because they're in the business of selling agents. And those are the agencies that will die in the next three to five years.
Our company. at a high level has the same business model. We sell agents too, but we are also aware of how these technology things play out. And frankly, I'm just excited about where the space is at.
And so we are disrupting ourselves. We are also going out and disrupting the market and saying, you don't need 10 agents on that account. Let us do it for six. Here's how it's going to work.
And maybe you don't even hire us, right? Like maybe we just do the project to say, let's take it from 10 to six. And that's the extent of how we collaborate, right? And maybe you stay with that agency with six agents.
Maybe you have six agents in house at that point. Doesn't really matter to us. We know where the world's going. I think those of you that are paying attention to AI know where the world's going.
And we'd be happy to partner with you guys. So check out the resource that's on the screen. Check out helpflow.com for any other questions.
We will be doing more events like this in the future. And thank you guys for sticking around. Thank you for the questions. And I hope that this has kind of opened your mind in terms of what's possible.
Dig in, let us know any questions and we'll talk to you soon. Thank you so much.
Jon TuckerWebsite: ecommercetownhall.com
Email: jtucker@helpflowchat.com
Facebook: Jon Tucker
Twitter/X: @JonTuckerUSA
LinkedIn: Jon Tucker
Ecommerce AI Mastermind
How The AI
Mastermind Works
This will be a laid-back, no-pitch session where you can connect, learn, and discover actionable AI tactics. Expect a lively discussion filled with actionable insights that'll give you those "aha moments." You’ll walk away with innovative AI strategies you can directly apply to your business, ready to spark real change.
Next Ecommerce AI Mastermind Session
11
22
38
32
Meet Your Host
Jon Tucker is CEO of HelpFlow - and a huge nerd when it comes using AI effectively. Jon has been leading the charge at HelpFlow in guiding our client brand partners to leverage AI over the past 2- 3 years. In addition to client use cases, Jon has spearheaded numerous ways that HelpFlow applies AI behind the scenes in our operations - so he knows how to apply AI in a wide range of business scenarios.
When it comes to the AI Mastermind, Jon is the perfect moderator to maximize insights and impact. Not only is he deep in the AI rabbit hole, but he's got years of experience being a member and leader of various business groups. With this nerdy knowledge of everything AI + ability to drive a group discussion, these masterminds are sure to be valuable for everyone that attends.
Ready to dive in? Register today!
Be A Guest on our Mastermind or Podcast
Contact us now if you are:
- An Ecommerce Brand with an inspiring story and best practices to share.
- Or a Tech Company who wants to showcase their technology as a useful ecommerce solution.