Create AI Voice Agents, AI Tutors, And More With DocsBot

Josh Dailey and Aaron Edwards talk about how DocsBot has developed into a potent AI platform for agencies, companies, support teams, and course developers. According to Aaron, companies may develop AI assistants who can offer assistance via websites, Slack, WhatsApp, Help Scout, and other channels by using DocsBot to train AI agents on a company’s websites, manuals, support tickets, videos, PDFs, audio, and other information.

The use of DocsBot as an AI instructor for online courses is a primary emphasis. According to Josh, instructors may include data from a variety of sources, such as websites, podcasts, YouTube videos, PDFs, and course materials. In order to make the AI more tailored for students, Aaron also shows how DocsBot may link to LifterLMS to get course progress and other user-specific data.

They also talk about the distinction between AI that can act and AI that only responds to inquiries. DocsBot may be able to extract bills, assist with billing, make appointments, gather leads, and establish connections with other business systems. According to Aaron, companies may also develop internal bots for staff members, keeping confidential data and resources apart from helpers that interact with the public. Josh emphasizes the potential for agencies to employ DocsBot to create AI assistants and instructors for customers on a large scale.

Josh emphasizes the potential for agencies to employ DocsBot to create AI assistants and instructors for customers on a large scale. Before deploying a bot, agencies may test it against frequently asked client inquiries using DocsBot’s tools and MCP connection. He highlights that by managing the deployment, testing, and continuous improvement for customers, agencies may bring value. He highlights that by managing the deployment, testing, and continuous improvement for customers, agencies may bring value.

Lastly, Josh and Aaron stress that AI does not have to take the place of human assistance. Customers can escalate questions to a person as needed, although DocsBot can manage repetitive queries. They also talk about selecting more trustworthy data, such up-to-date software code over out-of-date documentation, and managing which sources an AI agent utilizes.

Here’s Where To Go Next…

Get the Course Creator Starter Kit to help you (or your client) create, launch, and scale a high-value online learning website.

Also visit the creators of the LMScast podcast over at LifterLMS, the world’s leading most customizable learning management system software for WordPress. Create courses, coaching programs, online schools, and more with LifterLMS.

Browse more recent episodes of the LMScast podcast here or explore the entire back catalog since 2014.

And be sure to subscribe to get new podcast episodes delivered to your inbox every week.

Episode Transcript:

Chris Badgett: You’ve come to the right place if you’re looking to create, launch, and scale a high-value online training program. I’m your guide, Chris Badgett. I’m the co-founder of LifterLMS, the most powerful learning management system for WordPress. Stay to the end. I’ve got something special for you. Enjoy the show.

Hello, and welcome back to another episode of LMScast. I’m joined by some special guests. They’re back on the show. It’s Aaron Edwards and Josh Daly from docspot.ai. We’re gonna nerd out on AI for course creators people building internal training, for the agency professionals out there loo- looking to provide more value to their clients, for e-learning websites.

But first, welcome to the show, Aaron and Josh. 

Aaron Edwards: Yeah. Thanks for having us. 

Josh Dailey: Yeah. Thanks. Always enjoy being with you, Chris. 

Chris Badgett: Yeah. It’s f- it’s fun to catch up. About two years ago, when DocSpot started or closer to the start, whenever it was you guys have done a lot since then. So for those that haven’t heard of DocSpot, what does it do today?

Aaron Edwards: Yeah. Essentially, it allows you… And the, kind of the name is a misnomer. Like DocSpot, really it’s like a bot trained off your docs, is what it is. And so we were one of the first companies that allowed you to create a AI agent that’s trained on your website, your documentation now your past support tickets, videos, almost any kind of content that your company has, and create kind of that company knowledge brain that’s used for customer support primarily.

You can deploy that as a chat widget on your website or merge it into your app using our APIs or connect with lots of different channels like Slack, Messenger, WhatsApp, things like that and provide those kind of AI agents that can help your customers 

Chris Badgett: So what’s like an example from some of your users?

No matter whatever they are, whether it’s a WordPress product company or like how’s it look out in the wild if somebody wants to see it? And I know it’s on your site too, but who- … who else is using it? Who, who else is using it? What are they doing? 

Aaron Edwards: Yeah. We have a lot of WordPress product companies, so a lot of big ones that you know are using us to help provide support.

Sometimes it’s like tier one, like front-facing support, like a chat on a website. Sometimes it’s just attached to their help desk, so it’s helping their team by drafting answers that they just approve and send, or maybe edit. And it’s kind of a surf- self-learning leap like that. But as far as like customers, we have…

It’s been really hard for us to figure out who to target because, we have all these like corporations in Japan, like probably 20% of our customers are Japanese. We have the State Bank of Venezuela. We have all kinds of things, telecom companies. Manufacturers, that’s a big one.

If they have a lot of like catalogs with tons of parts and things like that that can be really complex to answer questions for the customers or their sales team, and so they’ll use us for that as well. So pretty much any business that has any kind of incoming support load, they can help relieve a lot of that using AI with DocSpot.

Chris Badgett: Which is like almost all businesses, which is awesome and why I understand it’s hard to target ’cause people use it in all kinds of different ways. 

Aaron Edwards: Yeah, 

Chris Badgett: exactly. And it’s very different from putting just a chat widget on your site to connect people to people. It’s actually trained and available 24/7 without needing a break.

Tell us about the training. Because a well-trained, AI, it needs a lot of context, so there’s like web pages, maybe other information. How do you kinda get DocSpot trained up? 

Aaron Edwards: Yeah. Josh, tell us what you can train it with. 

Josh Dailey: I’ll do my best, and you guys can jump in where you want as well. But really so the, the elephant in the room kind of conversation that we get from a lot of people about DocSpot is, “Why don’t I just build my own agents at this point?”

We’re to a spot in AI where people can self-code things, and everybody’s talking about vibe coding and releasing new things. But some really big differences and where we see people still going like, “Oh we need a solution like DocSpot,” is people who have knowledge spread across multiple different platforms.

And so the amount of things that you can connect with now in DocSpot that early on might have just been like a website and a few sources, we have over 30 native sources that we’ve built in. Your web content, whole YouTube channel videos for off of YouTube or other videos that you wanna upload it, we’ll do all the transcribing and pull in all of that kind of stuff, supports TSV files, which would be really big for things like a, a online course.

People who are using online courses often are wanting to do media sources. Your audio type files, PDFs. When Aaron was talking about the manufacturing-type sites that use it we do both basic PDF extraction and advanced PDF extraction, so it maintains the structure- Yeah … images tables, all of that kind of data can be maintained.

And then some of the funner ones that we have too are like- Because it’s primarily customer support would be your help desk. Help Scout, Zendesk, Zoho Desk, like all of these different platforms we’re able to connect into as training sources, and they can update automatically. So this has been the if you’ve been in around AI or unless you’re living under a rock, at this point the, the real thing is that it’s not that companies don’t have a lot of knowledge already available, it’s that it’s spread across a lot of sources, and how are you aggregating and pulling all of that information together?

My personal favorite things that’s taken place more recently is we’ve added a skill builder that allows you to connect into a lot more sources than just what we have natively on the site. So now it’s moved from just whatever we can natively provide and how fast we can churn them out to whatever our customers can dream up.

And Aaron could talk a little bit about this because it’s super fun to, to play with and use. But it literally is spinning up a coding agent into the cloud and writing the code for custom integrations to be built back into the widget so it can access and provide answers based off of external information.

Chris Badgett: How does that work, Aaron? 

Aaron Edwards: Imagine, let’s say, you’re making a bot for your course, right? You’re using Lifter to power it. And Lifter I assume you guys have some APIs. Yeah. Yeah, or exposed endpoints and stuff within WordPress. So you can just tell our skill builder, “Hey, I have this online course using Lifter for LMS.

I want you to build the query live, like course status and maybe the progress and membership information about the person who’s chatting with the bot.” And so you just tell it, and it would build all that using its internal coding agent essentially, and create that integration for you. And then it would just say, “Hey.

All right. Create the WordPress app password. Here’s the link to get it, and paste it in here.” And then it’ll be able to do that live. Create those kind of like live integrations, and they can be just like informational. It can be connecting to some external data source, such as LifterLMS.

And it can be scoped to the specific user who’s chatting with the bot. So it’ll check that they’re logged in with the widget, check what their WordPress user ID is or whatever, and then from that, it’s able to look up information only about that user or perform actions in their course, whatever it may be that you want it to do.

Chris Badgett: That’s amazing. In the e-learning space, one of the things that is like the holy grail is having an AI tutor. But spoiler alert, it’s already here with DocSpot. You just have to get it and set it up- … because there’s a lot of things that, like you mentioned, the data in multiple sources. If a subject matter expert is teaching on a topic, sure, they’ve done their best to make the gr- a great course, and maybe those videos are on YouTube or Vimeo or something else.

So there’s like transcript data, there’s probably PDFs and other things in the media library. They may have a separate YouTube channel that they’ve been building for 10 years with all kinds of helpful data for the tutor. They could have a long-running podcast. They could have a blog somewhere else. And so to really try to represent the human expert, the data’s gonna be spread all over the place, and DocSpot is ready to grab all that in.

So for, let’s say somebody non-technical is watching or listening and they’re like, “Hey, a AI tutor sounds really good. I’ve got the YouTube channel, the podcast, my course videos, and a blog somewhere else,” how would they get started? 

Aaron Edwards: Yeah. Essentially you just on, in our dashboard, that we have common connectors for all those kind of things, so for web, crawling websites uploading document files, PDFs, whatever it may be if you’re using that in the course we are able to support. Also this is good for LMS type stuff is audio and video. So if you have like those files that you’re using in the course, you can just add those and we use AI to transcribe them and add it to the bot’s training.

And if it’s YouTube, we’ll actually preserve like the timestamps and everything. So when it answers a user’s question it’s actually able to provide like the timestamp where it got that information. And we do actually have a feature where it could link directly to that timestamp so they could play that in the video, whatever it may be.

Chris Badgett: What about scale? Like for example, you mentioned like a manufacturing company, they may have a ton of products with a ton of instructional PDFs. So like- … for example, it’s one thing if your website has 100 pages on it, including lessons and things, or you have 100 YouTube videos.

But what if somebody’s scaled out and they have 20,000 pages on their website and just like tons and tons of source material? 

How does DocSpot work at scale? 

Aaron Edwards: Yeah, we handle that pretty well. We do have plenty of people that have that many like website pages and stuff.

So you could just point, like for example, your site map to it, and so every week or whatever, it’s recrawling the site and it’s finding those new blog posts or content that you’ve added to add to your bot. All that kind of stuff. And now we have a new we have a full MCP server, and we have a agent within our dashboard, which has access to that same MCP server without needing any technical know-how, and so that’s the best way we’ve found to manage this stuff now, is don’t learn a new interface. Don’t learn how like the software works. Just drop it in your cloud or your Codex or whatever, and then say, “Hey, I need this AI agent that does this for me,” and then it will go figure out all your data sources, figure out the best way to train it upload and add all that information to your agent, and configure it to work for you.

Josh Dailey: Yeah. I think something that when you say a beginner who doesn’t, may not know how to do this stuff, I think they’re … When you start talking about all the sources that it can connect to and all the information that’s spread out across all these places, it b- gets really overwhelming fast. Like you’re like, “Oh, I’m never gonna be able to figure this out.

It feels like such a big task and a big lift.” And when we get asked “How quickly do you think I could have a bot up and running for my site?” And we’re … Literally within a- an hour you could be deployed and running something that probably is really knowledgeable and more knowledgeable than most people about the topic and the information that you’re providing.

Even faster than that, if you wanted to, you can be in the environment testing right away. The longest part of the experience is really, like people wanna refine that answer, the way that it’s answering and that kind of stuff. And we find people more reluctant about the, Wanting it to be able to answer more accurately more questions faster than we do about actually being able to get to deployment.

So within 24 hours generally if somebody’s working with us to deploy, we already have a bot that was created and ready for them to start testing, and then it’s, that human time where they’re doing their own testing that takes the most time in getting it to production. 

Chris Badgett: That makes sense. Like, when you work with AI, you It’s not just like everything’s a one shot, you gotta refine it. Yeah. So for example, let’s say somebody had an old blog post but there’s a newer one that has more updated information, and the bot references the old one. You can refine it and be like, “Hey, ignore that old post or whatever.”

What’s that called? Are those rules or, It’s not r- personality. What are those kind of fine-tuning you’re talking about?

This episode of LMScast is brought to you by Popup Maker, the most powerful and trusted popup solution for WordPress. Whether you’re selling online courses or building a membership site, Popup Maker helps you grow your email list, boost conversions, and engage your visitors with highly targeted, customizable popups.

You can create smart popups for things like LifterLMS or WooCommerce cart abandonment recovery, upsells, course engagement, activation, targeted offers based on exactly what your students are doing inside LifterLMS. I personally use Popup Maker on all of our websites at LifterLMS for lead magnet opt-ins, cart abandonment, upsells, downsells, and guiding users to the right next step.

Popup Maker is essential for growing my email list and making more money through my website. Are you ready to take your website to the next level? Head on over to wppopupmaker.com/lmscast to save 15% off on your order. The discount automatically applies when you visit through that link. Popup Maker also has a free version, so you can start with that too.

So go to wppopupmaker.com/lmscast now to save 15% or sign up for the free plan, and start getting more leads, sales, and student engagement today. Now back to the show.

Aaron Edwards: yeah, there’s different kind of ways. So we give you full access to the instructions to where you can add kind of behavioral kind of instructions or maybe the way it handles your sources, like the way it prioritizes them and searches through them.

That can be easily added to the instructions for those kind of things. Another common use case is you’ll look at those chat logs. You’ll see, like, how the bot’s responding to your users, and then you can just go in, you can click “I’d like to revise this answer.” And just right there you can just type in the new answer or adjust it, and that kind of goes, recycles back into your bot training so it knows better how to answer those similar questions in the future.

And then we also have AI agents built into our platform. So if you’re, like, wondering why did the AI answer this way, or I didn’t want it to answer that way, you can just click analyze and that will spin up a full agent with access to everything, and it’s specifically trained on how to analyze and improve the responses.

So it’s gonna look out and see is there a problem in the instructions? Is there a problem in your training data, like the sources, like a conflict like you mentioned? Or is there… You’re just missing the training data to be able to answer this, and it’s able to identify that and actually fix it for you.

Chris Badgett: In terms of public-facing versus having it be, like, a, a resource internally to the company, like an employee, a digital employee that helps with- … training or onboarding, can you do both a public-facing bot and a internal bot for the team? 

Aaron Edwards: Yeah, that’s a very common use case. You can create, depending on your plan, you can create any number of them, and they can be trained on different content or the same content, and they can have different rules and different tools.

So often we’ll see the sales team or even the support team have a, a internal one that’s maybe trained on similar set of docs, but also private docs and information, and it may be connected via MCP to their internal tools, like their help desk or their billing system of record or whatever.

So they’re able … The AI bot is able to do more than just answer the questions. It’s able to actually do actions, like deal with billing issues and things like that. And that can be scoped to a private bot that’s only deployed, for example, in their Slack or something like that. 

Chris Badgett: So for example, if you wanted to automate a refund process, you could.

And it would be human hands-off if you were happy with it and good to go. 

Aaron Edwards: Yeah. No, and we have built-in integrations for Stripe and EDD and Freemius and WooCommerce and things like that. So you can just add one click or add your API key, and then it’s able to do some of those common billing actions that users ask for.

Chris Badgett: Wow, that’s really cool. And explain again, you k- you did, but I just want it to be clear for people out there watching and listening, the difference between answers and actions. 

Josh Dailey: Yeah. So in the early stages of AI it did a really good job when you would ask a question of agents. They would come back and provide an answer to you, but it usually gave you a list of tasks that you had to perform, and it’d be like, “All right, here’s your next steps.

Go do one, two, three, four, five, and then you’ll be a millionaire at the end of it.” That kind of output kind of thing that it would tell you to do. But it couldn’t go in and then perform those actions because it didn’t have access to your computer, it didn’t have access to the information that it needed to actually go in and do the whatever that one, two, three, four was.

Now, with an action, when you ask it to do something, it can actually check, do I have the skill or the tool that I need to go out and perform that behavior so that the customer doesn’t have to think about it and do that? So if I come in and I say, “I wanna get my invoice,” instead of saying, “Here’s your account.

Go- click on that, it’ll take you over to your account page, find this link, and then you’ll be able to download your invoice,” now we can actually say, “Oh, I have access to their account information in Stripe, and I can generate that invoice right inside the widget with a clickable link so that they can download that, never have to leave the conversation that they’re in.”

… Another one that we have common use case would be scheduling an appointment and somebody would say, “Hey, I wanna meet with your team.” Instead of saying, “All right, here’s the Contact Us form or the link to Calendly or TidyCal,” we can actually display your TidyCal, Calendly, cal.com right inside of the widget.

They select their time and date, and then they say, “All right, now we have context and awareness of the schedule and are able to make sure that you have a date and time that meets your requirements.” So those are some examples of actions that now have become easier over the last couple years.

I would even s- argue probably within the last six months or so where that became a natural thing for us to say, “Oh, look at my email,” where before it couldn’t do that, and perform those actions before y- for you. 

Chris Badgett: … Tell us more about, like, where the bot lives, ’cause it’s not just on the website.

You mentioned as an example the same bot could be in internally in the company Slack- … or is it, am I correct, it could be sitting in your Help Scout as an example drafting replies that you can review as a human in the loop and then send it? A- Yeah … tell us about those locations. 

Aaron Edwards: Yeah. Mind if I share my screen?

We’ll see. 

Chris Badgett: Go for it. Here, let me pull it up. 

Aaron Edwards: See that. Yeah, you pull it up. 

Chris Badgett: All right. 

Aaron Edwards: So here’s kind of a bot. As we were talking, I actually had a AI agent spin this up for us. And this is essentially what did it train with. It pulled in your LifterLMS quick start course so just to see what it looked like training from a course and a little bit on the website.

But you can deploy that bot in a lot of different ways. So we have our chat widget which that’s easy to just copy and paste code. Share links, so you can have a chat page that looks like this where it’s just a public URL that you can share with your users. You can connect your phone when we get into voice agents.

Slack, we have a native Slack app and a Teams one coming soon. Facebook Messenger if you wanted to reply to incoming messages to your business. WhatsApp, Telegram Help Scout if you wanna hel- connect it to your help desk to write draft replies to your users. A lot of different ways and channels to deploy it.

But voice is a fun one for sure. That’s something that we launched about a month or two ago, and so actually I’m excited. OpenAI today, they released their new live voice model. I don’t know if you’ve seen the ads with the grandmas that are, like, talking to it. 

Chris Badgett: Haven’t 

Aaron Edwards: seen 

Chris Badgett: it yet. 

Aaron Edwards: Okay.

They’ve been pushing those everywhere. I’ve been seeing them online and streaming and stuff, but essentially it’s the first voice model that’s agentic that al- allows it to talk and hear at the same time, so there’s no, like- 

Chris Badgett: Latency … 

Aaron Edwards: turn… Yeah, no latency, no turn-based detection. If you’re translating, for example, it can speak in the language it’s translating to as you’re talking to it.

There’s no turns which is pretty amazing. So I’m looking forward to integrating that, but we’re already using the previous generation model, which is only a few months old, which is their real-time API. And that is essentially the same thing as voice tokens in, voice tokens out. There’s no turning it to text and then back to audio step that most voice agents have, so it gives you a really natural conversation that you can have with users 

Chris Badgett: This is so cool.

And I’m somewhat of a Power AI user, and the idea of trying to build all this ourselves internally is- … even as a power user is complicated, and you guys are just there ready to roll with all these connections and everything. 

Aaron Edwards: Yep, that’s the idea. 

Chris Badgett: In terms of voice, like I, just for me personally, where I really wanna get into voice, I haven’t done it yet, I’ve d- I’ve tested some things, but is for sales, like pre-sales basically.

If somebody wanted to set up a 24/7 voice agent bot that- … was trained on the product, the offering, you know, all that, the, where sales… ‘Cause a lot of people don’t necessarily wanna get on the phone or have time, or they’re scared of sales. But if a bot can do it for you 24/7 in any language on Earth, why not?

Yeah. So how would somebody set that up? 

Aaron Edwards: Yeah, it’s the same process that we have. So you’re just adding your sources however you wanna train it. And then we do have separate voice instructions. So we’ve found that the way your bot behaves in a chat widget is very different than how you want it to behave, like, when you’re talking to it via voice.

And we have some templates that we’ll start up, and it’ll build that out for you, and then you can edit it from there. So for example we made this like a tutor, like we were talking about before the call. So it’s to act like a tutor and help users and things like that. And you can also add actions and things to your voice agent.

So we support all the same actions like web search, MCP servers, skills. If you wanna add a skill for it to do a specific task, like we talked about billing. Lead collection, you can add lead collection fields there. And we have a full form kind of builder. So you can say, “Okay, I want it to collect their phone number this text area about this,” or whatever.

And so it’s actually able to depending on if it’s voice in the widget, it’ll show you that form to fill out, or if it’s over a phone it’s gonna actually interview the user to collect the details it needs. And if it’s phone, you can also do things like call transferring, so have it transfer to your human phone number, right?

After it acts as a receptionist kind of thing. You can even put rules for when it’s okay to transfer you and even an availability schedule if you’re, like, a business that has certain hours. You only want it to transfer during business hours to that phone. So that’s some of the initial tools that we’ve added to our voice agents.

But it’s pretty user-friendly and simple, and our agent kind of sets it all up for you during the onboarding. 

Chris Badgett: In terms of the voice, in my investigation, one of the places I got hung up, and I understand the widget idea. It’s like kind of a web view of the voice. But then if somebody- … wants to connect a phone number even if it’s their cellphone number, or do they need to go to Twilio or something and buy- a number? How- Yeah … do you get to the actual there’s a phone number that peop- some people are used to calling phone numbers. 

I would say people are more used to calling phone numbers than interacting with voice agent widgets- 

Aaron Edwards: Yeah, on a 

Chris Badgett: website … at least today. Exactly, 

Aaron Edwards: yeah. 

Chris Badgett: Like, how do you get the phone number connected?

Aaron Edwards: Yeah. So we support right now, we support Twilio, like you mentioned, so you can purchase a phone number there. I think it’s like a dollar or two per month. Or you can forward phone numbers there, so if you have just your cell phone number, and you can turn on call forwarding, it forwards to there, and then we answer the Twilio number.

So I can see I have some numbers here that I bought. This one’s available because we made it for WordCamp US specific or this one is assigned to another bot, and so I just click assign to the bot, and now it’s gonna start responding to that phone number. Or if you have a business phone most business phones or PBNs, they support, it’s a standard called SIP, and essentially you just follow these instructions to put in the specific address, and then it will redirect any calls to that address essentially.

So we try to make it as user-friendly as possible. 

Chris Badgett: And a question which may be a feature request, I apologize. Is is WhatsApp an option if you wanted a WhatsApp number? Or if- 

Yeah 

Aaron Edwards: That’s a good question. I don’t know. I, we support WhatsApp via chat, and you can also send audio or image messages that way, and the bot will respond, but it’s gonna respond in that same text-based format.

D- WhatsApp phone number, I think it’s the s- same kind of thing, like a business phone number. I’m not sure if they allow forwarding or things like that. 

Chris Badgett: The reason I ask is, whenever I leave the United States, I realize that the rest of the world uses WhatsApp. And also you get around the long-distance calling fees or whatever.

So it’s just- Yeah … it’s just something I’ve been trying to figure out- … on my own. 

Josh Dailey: Yeah. 

Chris Badgett: In term- in terms of the bots let’s say we want a sales bot, a tutor, a customer support bot, and a bot for our team. Is it better to just use one bot and let make sure it knows where it is? Or are we building four different bots which f- on the same account?

Or how do, what would you recommend there? Yeah. 

Aaron Edwards: I think it’s gonna depend on if it needs separate rules and instructions if it needs access to specific data that’s maybe restricted between audiences or certain tools that it’s connected to, whether that’s a skill or API or MCP or whatever.

That’s what’s gonna determine it. So you can have one for example we do allow if you create one bot for customer support, you can have different voice instructions for that channel. You can have different instructions for when it’s responding via help desk versus chat. But it’s still the kind of the same underlying core training information and stuff.

But we make it really easy to create multiples, and you can even copy it to create a copy of that bot with the same training content and everything like that. And you can adjust the settings and permissions of each one individually. 

Chris Badgett: Awesome. And then for an agency that’s building LMS sites for clients- 

Like as an example, like WordPress has an MCP, LifterLMS has 100 MCP end- endpoints that you can do stuff with. And let’s say we’re like a high volume agency and we’re doing tons of sites and getting lots of new clients, how could we leverage DocSpot as an agency? Are we like Slacking with our bot and it’s spinning up sites and building stuff?

Or what’s possible for the agency?

This episode of LMScast is brought to you by WP Fusion, the powerful WordPress software that connects your website to your CRM and marketing automation platform, whether that’s HubSpot, Salesforce, ActiveCampaign, GoHighLevel, and many more. If you’re selling online courses, memberships, coaching, digital products, and more, WP Fusion turns your WordPress website into a smarter, more personalized, and more automated online business.

With deep integration into LifterLMS, WooCommerce, the popular form plugins, and many more top WordPress tools, WP Fusion lets you automatically create and tag contacts, unlock course and membership access, track student progress, recover abandoned carts, and trigger powerful automations based on what your users are actually doing on your LMS website.

We use WP Fusion ourselves on our LifterLMS website, which is powered by WooCommerce for selling software subscriptions, but we also use WP Fusion on the LifterLMS Academy, where we are selling access to courses and memberships powered, of course, by LifterLMS. WP Fusion connects our websites and CRM seamlessly together so that we can create a better customer journey, deliver the right offers, and automate follow-up based on user behavior.

If you wanna sell more, personalize the learning experience, and build a more connected, automated online business, check out WP Fusion today. Go to wpfusion.com and use the coupon code LifterLMS to save 15%. Now back to the show.

Aaron Edwards: Yeah. Josh? 

Josh Dailey: Yeah. So this is actually one of my favorite ways to interact with DocSpot right now is we have our own admin MCP for DocSpot, so you can connect it into your coding agents, the places that you work from, however, wherever you wanna connect that into.

And then you could pretty much do anything that the operator can do that we have built in. So if you’re like, “Oh, I don’t know enough about MCPs, and that scares me,” you could just talk to the widget on the bottom. But if you’re talking about as an agency, I’m wanting to spin up bots, essentially you could just connect to the MCP for DocSpot and say, “Hey, I want to build a tutor like Aaron did for this customer over here.

Here’s some links that are really important to train off of. Here’s the sources we wanna train off of,” and give it those instructions. Then it will go out and do the work and perform that, that work for you and you don’t have to go in and do it. And then like I said, then that becomes more of an exercise around testing.

I will say we’ve got a really good case study on our site right now about, about this kind of what’s the best way to perform the, the, the tests. We’re building in a feature called Evals that’s coming out pretty soon. But the– essentially, the best mode is to come up with 50, 100 common questions that you’re trying to resolve, and then to feed those in.

And you can use something like Claude using your MCP connection to do that, say, “These are the questions you have to answer. Tell us how you would answer these.” It comes back and shows you. And if you say, “Oh, only 70% of these are getting right. I need these other questions to be changed. How can we perform that thing?”

Then you can keep running that same set of eval questions back through your system so that you have a baseline to function from, and then a way to improve the messaging. And we have essentially our customer said, “We made DocSpot earn its place on our support team.” We didn’t just release it, and we had it run through two months of rigorous testing against that, and it had to perform at, I think it was 80% or better before they would actually deploy it for themselves to, to utilize it on their live site.

And so that’s some of the stuff where an agency becomes highly valuable Where a lot of end users are saying, “Hey, I don’t even want to think about this.” It’s we heard this for years and years around website builders and, block editors and all that stuff. It’s like everybody’s just going to go build their own sites now.

Nobody’s going to use an agency anymore. 

And it would really quickly became, oh, everybody’s still going to need an agency because they don’t want to build websites. They want to run their business, and they want to take care of the physical pr- products and the needs for the, the customers, right?

And so it’s the same with this AI stuff is going I know I could probably go do this myself, but I need an agency because they are experts in this. They’re going to be able to do it faster. They’re going to be able to run the evaluations that I need and do the implementations. And so as a partner to our agencies, like as DocSpot, we think agencies are the mode for mass distribution and adoption of this agentic stuff.

It’s really important because they have the client relationships. So we’re trying to make our tooling as valuable to agencies as possible. 

Chris Badgett: Yeah, and that’s what- Hope that 

Josh Dailey: answered your question. 

Chris Badgett: That does. Yeah. And that’s what agencies have always done. Like they lead- Yeah … and then bring the clients along.

Yeah. And the client’s usually really busy and doesn’t want to become AI expert or coder- … or whatever. Yeah. 

Aaron Edwards: That hasn’t changed. 

Chris Badgett: Yeah it hasn’t changed at all. And I just wanna put it out there to somebody listening or watching, if you figure out how to build just a workflow and a process for spinning up AI tutors for a WordPress LMS website, that could be a huge business.

And for multiple agencies. It’s not just somebody might do well there. Tons of agencies could do well by solving the AI tutor situation with DocSpot, and then, going after the market of all the people that are already using these LMSs, not just Lifter in the WordPress space. That would be a huge opportunity, which is gonna help the students have higher completion rates, and- 

The, the site owner sell more. There’s just so much. You could- Yeah … then char- charge more if it comes with tutoring. It’s infinite opportunity there. 

Josh Dailey: We are working with quite a few education institutions, which is a huge value side of why people still use a DocSpot security. I think we covered that a little earlier.

But they don’t wanna deal with HIPAA, SOC 2, GDPR, all of those things that, that we- Accessibility … take administration and security risk away from the organizations to do that. And so 100% on board with you, Chris. If there’s somebody out there right now watching that wants to do that- … and wants to partner with us, we do have a partner program that we’ve put in place that we’d love to chat more with you about the implementation process of that.

But we could be a foundational layer. We’ve taken that kind of in two, two approaches. One is you can resell the service, like on our business plan, is to buy it and just pack a few different people on there, and you can separate out so that none of the other people can see none of your c- clients could see the other person’s bots, et cetera.

But then we also have our partner program where we do a, a pretty generous referral commission as well. So they could just set it up and spin out agents for for their customers, and then we’ll handle, help with kind of the support side and be more of a partner to them. 

Chris Badgett: That’s awesome. When it comes to let’s say a human handoff, like where a we’re using DocSpot as a sales agent, but it’s doing the qualification step, and whatever happens, like there’s enough budget or the right person’s on the phone, it’s transfers it to a live human during business hours or whatever.

How would you do that qualification piece, train the bot to make sure the person’s a fit, then transfer to the live e- human? 

Aaron Edwards: Yeah. Honestly, I don’t know that, so I just type it into our agent in the dashboard and say, “Hey, this is what looks like a good customer and not.” Change the qualification based on that, and I’ll just let it decide that for me.

That, that may be a cheater answer. But,

Chris Badgett: But that is the answer … we- 

Aaron Edwards: Yeah … 

Chris Badgett: that is the answer ’cause you can just … if you don’t have all the pieces, just describe- Yeah … the solution you want. 

Aaron Edwards: Yeah. And we do have a template as, like lead qualification. And so essentially this is a workflow that we’ve done to where it first like asks the user their business website or about their business, right?

And then it actually is able to use our search tool, so it’s using search behind the scenes to actually get context about that business. And then based on your criteria, it can use that context to judge whether or not they qualify. And if they do, then it’s able to pull up the, their booking calendar to book like a call or submit the lead via the lead submission tool, that kind of thing.

And that can all happen over a phone channel. 

Chris Badgett: That’s awesome. Hopefully you can’t hear it, but there’s somebody with a chainsaw outside, so I’m gonna close my windows in a second. But I’m gonna queue up with a question while I close the windows, which is I know people get excited about this stuff, but then they feel overwhelmed about getting started.

What’s the best way and like first action steps to come into the DocSpot universe?

Aaron Edwards: Yeah, Josh? 

Josh Dailey: Yeah, I can go with that question. We actually try to make that super easy. If you go to our website, and we’re set up to be able to self-serve your plan. You just pick the plan that you want to be on. If you’re coming to us with a business, from a business, you can actually do a 14-day technical evaluation trial or f- for free, and so it will give you access to the business plan, and you can use all our features without paying a dime.

You just have to put your credit card information in to validate, to help us validate, like your organization. But we won’t charge that, and then if you cancel before the 14-day trial ends, then you obviously we’re not gonna charge you, and you- there’s no refund or anything necessary at that point.

So we make it really easy to get started for free and just test this stuff out and would encourage anybody to do that because we think it sells itself. As soon as people start playing with this stuff, they’re like, “Oh, this is a lot easier than what I thought it was gonna be.” The one other thing I was gonna add about the lead kind of qualification pieces, we do have…

I think you’re asking the hot topic questions, Chris. I don’t know if you intentionally are doing that, but it is the number one thing that people are concerned with is I don’t want my customer to get caught in a loop where they have to just keep talking to AI, and how do I make sure they have a way out of that?

So human escalation is a massive, important thing to us. We’re not believers at this point anyway that AI’s taking over every role in customer support. There’s still need for human support and it just, the repetitive questions that we get over and over, like billing, what you mentioned is a massive percent of questions that come through.

So if AI can answer that effectively, then you don’t need to. But for pre-sales, we also have not just that qualification one, we have what we call lead capture in there, and you can set it up so that your agent has a– We have a full form builder, and you might already have your form on your site somewhere.

You can just copy that information and say, “Hey, build me this form that shows up in the widget so they don’t have to go to somewhere else on my site.” And when they start the chat, say, “Hey, I wanna capture some information about you ’cause I– if we get disconnected, I wanna be able to follow up.” And that could be the first thing they see in there.

Or if you say I don’t wanna scare them away by trying to capture information. I want them to be able to chat with us you can select to not have that form appear until right before escalate, human escalation. Say “Hey, before I escalate you or connect you with one of our human representatives, can you provide your email address so that if we’re unable to assist you right now, we can help you later?”

And that has become a really valuable sales pre-sales kind of tool for a lot of our customers, so I think that’s the approach we’re taking at this point. 

Chris Badgett: Very cool. And this is a technical question, I’m probably gonna mess it up asking it, but as an example if I’m using DocSpot and I have somebody asking support on my website, I may have trained it on everything Lifter, but it’s running into a core WordPress issue.

Does it have the ChatGPT or Claude or whatever the large language models know a lot about WordPress? Or is it- 

Josh Dailey: Yeah … 

Chris Badgett: so there’s like the information we trained it on, and then there’s this like huge amount of content in the large language- Yep … models. How do we think through But of course sometime- like as a tutor, we want the tutor to be w- like, just focused on our material and not necessarily surfacing- Yeah

Wikipedia answers and stuff like that. So how’s- 

… 

Chris Badgett: Where’s the line of just the huge public data versus your private training? 

Josh Dailey: Yeah. By default, we’re pretty tight with that, so like it’s not going to go out and search the broader web. But we built in the option to expand that out, so you can actually enable specific websites or other knowledge sources that are public sources that you could connect in.

The WordPress ex- WordPress example is a great example because because there’s so much public information about WordPress as an open source project that you could say, oh, if somebody asks… If I’m a, a WordPress plugin and I want general WordPress knowledge to be able to help provide an answer, you could connect that as a source, the public documentation as a source, and then say, “Go search this if they ask for this, this question.”

Yeah. And you can actually narrow or expand depending on what you’re trying to achieve on the site. 

Aaron Edwards: Yeah. Or in your instructions you can relax its restrictions on only answering from your sources and saying, “Hey, if the question is about WordPress, it’s okay to answer based on your general knowledge, as long as you tell them to check that information is accurate,” or something along those lines.

And so you can relax the restrictions that are there to prevent hallucination for things, maybe your product, that the AI models have not been trained on very well and don’t know the facts, you know? 

Chris Badgett: And a question on I’m sure there’s probably a GitHub integration maybe.

For example, this would just be a question for a software company, which I imagine are definitely early adopters of DocSpot ’cause they’re- 

… 

Chris Badgett: Technical, is if we’re integrated with GitHub, and the agent can see all the code that we’ve written that runs our app and let’s say we have– we also trained it on our documentation.

A- and as it happens with software companies, sometimes the documentation gets out of date or stale- … or even might have some errors in it. 

How do we… I guess we’re looking at the response and seeing if, oh, that’s not quite right, and then we train it. But how does it… I’m just thinking about if you have two sources- Yeah

like past Help Scout tickets, it’s training on our docs, and the code in GitHub, and maybe those aren’t always lined up. There might be some conflicts. How does it- … how do we work through that to optimize it over time? 

Aaron Edwards: Yeah. Ultimately I think code is probably, for a software company, code is the greatest source of truth, right?

And docs maybe log behind lag behind that. And so we do have a GitHub live search integration, so you can connect to your GitHub repo and then… And this is especially good for open source software. So if it’s closed source, you probably don’t want users to be able to prompt inject and get them to print off the code to you,

but if it’s an open source thing and that code’s open anyway, then it’s fine. So it can actually go and search GitHub to read the code itself and answer questions based on how it’s actually built. And then in your instructions, you can have those priorities. If sources conflict Help Scout or code is the most accurate, and then Help Scout is the second at most, and then our docs are third most, or whatever it may be.

You can add kind of those things to help it judge responses or even provide caveats to users 

Chris Badgett: This is awesome. I’ve really enjoyed this, Aaron and Josh. I know as kind of AI tech people, we’re ahead of the curve, but I feel here comes everybody. We may be in the top 1% of AI users- 

But as the other 95 or 99% starts catching up, DocsBot, in my opinion, is just gonna take off when people get it, and you remove all that friction of where a regular non-technical business owner or whoever can just get going and leverage all these capabilities. It’s so impressive what you’ve built. 

Aaron Edwards: Oh, thanks.

Josh Dailey: Really appreciate it, yeah. 

Aaron Edwards: Yeah. 

Chris Badgett: Thanks for coming on the show. That’s docsbot.ai. If you have any questions, just talk to the DocsBot, agent. But of course- … I don’t know. I’ll leave it with you guys to close it out. If somebody’s super excited, they can talk to the, the DocsBot on docsbot.ai, but- how else can they, get their questions answered and figured out how to move forward? 

Aaron Edwards: Yeah. You can talk to the agent on our site, and that can forward you to us if you need to. And we do– if you’re an agency or you’re you have clients of your own, and you’re looking to maybe add these kind of AI tutors or things to your courses we would love to talk with you and give you a demo and show you what’s possible there.

So please reach out to us. 

Josh Dailey: And social channels, Aaron and I are both pretty active on X. So I’m @JoshDailey, Aaron is @UglyRobot, and for LinkedIn, you can just search our names. But Josh Dailey is my last name is D-A-I-L-E-Y, which is right down here somewhere. So you could see that, but that’s the spelling of it.

But yeah, so if you wanna DM us or talk to us that way as well we’re pretty available. 

Chris Badgett: Thank you for coming back- Yeah … on the show. We’re definitely gonna need to do it again in about a year. As you said the past six months, a lot has changed. I can only imagine the next 12 months of where we’re gonna be at.

Yeah. Yeah. So- 

Aaron Edwards: For 

Chris Badgett: sure … thanks for coming on. I really appreciate it. 

Aaron Edwards: Yeah. Thank you. 

Josh Dailey: Right on. Thanks, Chris.

Chris Badgett: And that’s a wrap for this episode of LMScast. Did you enjoy that episode? Tell your friends and be sure to subscribe so you don’t miss the next episode. And I’ve got a gift for you over at LifterLMS.com/gift. Go to LifterLMS.com/gift. Keep learning, keep taking action, and I’ll see you in the next episode.