SaaStr AI Day: How We Turned AI Support Agents in Revenue Drivers with Happyfox

Kick things off here. We are really excited to have Shaolin Jain here for our next session. Shaolin is the founder and CEO of HappyFox, and he's gonna be giving a great session that helps reframe how to think about support not as a cost center, but as a growth engine. Shaolin gave a talk at Saastr London just last month, one of our most popular sessions. And so we're really excited to have him here and sharing how his team's using AI agents, AI agents, and more. So I'm going to kick it over to Shaolin. We're going to have time for Q&A at the end of this, so please submit questions in the chat. You can also submit them on the live stream if you're tuning in from there. And without— with that, I will kick it over to you, Shaolin.

Hey, thank you so much, David. Hi everyone. Excited to be at yet another SASTR event. It's fun to talk about SaaS and AI all in one go. And today I'm going to talk about the hidden growth engine that we discovered ourselves, and then we put it to use for our customers. And usually support is the last place you look at for when you think about GTM, when you want more customer revenue, uh, more upsell, cross-sell, and so on. So Not only have we found that support data is extremely valuable from a GTM perspective, but it is also not very expensive to actually go after that. And in fact, every company should do this in one, some shape or form. So just a quick introduction if you haven't heard me before. I'm Shalin, founder of HappyFox. We've built a $20 million+ ARR bootstrap company with just over 2,200 customers. And we're largely in the CX and ITSM space where we offer helpdesk and service desk, which are the two large systems of records for us. We are just about 100 employees and we've rolled out a whole bunch of AI stack for our customers to make support easy. But when it comes to GTM, we have been— we are a very small GTM team as such. So when AI came in and we started doing all things GTM with AI, we realized that not having a very large team was actually a good advantage, a good starting point for us. So we started devoting engineering resources and built a lot of tooling, plus started using Claude prompts toward artifacts more aggressively to layer on all of our content, social, and our CRM activities with it. We also have built a workflow automation product since 2019, and that's our version of Zapier or N8N to do a lot of things that where we haven't got to it, solving it in a different manner built into our tooling. So, but we largely leverage a lot of our Zoom calls and use all of that data to, um, to power all of our sales and marketing. So today I'm specifically going to talk about the best GTM win we have ever had, and these are just collection of 8 agents. I think we have a couple of more agents, but I felt these 8 really were the best that actually gave us the highest leverage. And it was— we looked at our cost of doing all of this, and we probably spent less than $100 on our own product to do that. So if anybody tries to do this with HappyFox Autopilot, it will pretty much be a spend much lesser than this. Now, what is the opportunity with support data and why I call this a hidden revenue? That customers are reaching out you post-purchase or during their trial phase, or even when they're just inquiring about your company and sending you an offer or things like that. So a lot of times it does come to support conversations, and usually support data is where marketing and sales don't have a seat on the table, right? So you don't give all marketing people and all salespeople access to customer support data. If you do, that's great, but generally I've seen most companies don't. And support data is also considered unstructured. It's a whole lot of different issues. It's not necessarily any sales signal or a marketing signal directly. It's unstructured. It's not visible to sales and marketing team, and it usually doesn't get routed at the right time to the right person. Now, what is the hidden revenue map that I— we discovered was there was a lot of expansion chatter churn chatter, renewal chatter, and cross-sell opportunities, which were reading between the lines, uh, that support reps didn't have that, um, that natural inclination to do that. Uh, and also, uh, if you found all of that opportunity existed, the one of the best part of our support data compared to what we have tried to do on marketing and sales is there's a lot of signals. You have to really find which signal works the best, what weight each signal could give you the best result. But with support data, it's very high context. They're already— customer is chatty and talking to you, it's urgent, and it's directly from the stakeholder, right? It's not from an economic buyer or a sponsor, it's directly from the person who's maybe implementing your tool. So we divided these opportunities and into 4 different categories. We want to capture more revenue, we want to protect churn, we want to make sure any customer who's already in our pipeline goes through a sales acceleration where both the sales team and support team do a good joint effort in making sure the customer gets the most VIP treatment possible. And oftentimes, you know, building case studies or getting a testimonial from somebody becomes a seasonal thing. But we wanted that to happen on every opportunity possible when a customer is reaching out to us. So, and I'm going to dive into demo and show you exactly how we did it, but I want to put some proof forward before I actually show you how it's done. So we captured a lot of opportunities, both upsell and cross-sell opportunities. We prevented churns, uh, when companies got acquired or when, uh, leadership changed and somebody came back and said, hey, I'm relooking at Happy Fox and want to get an intro to your team and things like that. We, we, we paid a lot more attention to those things because those are usually the turning points in a relationship, even if it's like 8 years later, uh, of that customer being with us. And then we wanted to influence more pipeline and essentially save a lot of work between support communicating to marketing, support communicating to sales, which often doesn't happen. Support is busy, uh, with tickets. They have their work cut out really, really hard, and, uh, and they often don't have the time to just hand off everything in the best possible manner. So we saved a lot of time when some of those activities weren't happening, and if they were happening, all those meetings have now disappeared. Now it would be all very not worth it if it was taking a lot of time to set it up and a lot of tweaking and, you know, you spend a couple of months trying to make this happen. But because it's first-party data sitting in the first-party tool, the whole implementation is a lot more easier. You start seeing results right same day and the cost is super low. And there's no engineering effort required, no apps to build, no flows to, to write, even when it comes to multiple apps in picture. So let me dive right into the product and the demo and show it all to you. And we'll have time for Q&A at the end. So if you want me to walk through anything else towards the end, feel free to ask. Now, so this is our product, HappyFox Helpdesk, and I've just logged into our Autopilot tooling. So we have a bunch of AI products, of which Autopilot is our latest AI tooling, and originally built for support teams to route tickets better, tag tickets better, find out different, you know, sentiment opportunities and data compliance and so on. But it was an aha moment for us when we said, okay, Hey, why don't we— and I mean, why I call this a hidden revenue, because even for us it wasn't very obvious when we started implementing AI. It was all focused on all the marketing tools, all the sales tools, all the content writing tools. Um, but I think implementing it on our volume of support data was extremely helpful. So, so HappyFox Autopilot allows you— so we built a lot of AI agents specifically for SaaS, and today I'm going to talk about 7 to 8 agents that we've hired, and I can hire multiple of them if I have different products and different scenarios. But let's look into some of the agents that we have. So we have a testimonial identifier. Our marketing team loves that. It flags any, uh, aha moment that our customer has shared and turn it into a quote that we can use on our website, on our social. They also have a case study finder where we can discover some case studies from existing customers. Where they talk about how they implemented something or what got done in the process of going back and forth with support. And even if support didn't do the work, the success team did the work, or even sales team did the work, the mention of it just brings up that opportunity for us to do that. So, so let's, let's look into one of these agents. I want to dive right into it. So let's go into, let's say, now most of you are probably using Salesforce or HubSpot. We use HappyFox CRM, but if you're using, let's say, HubSpot, and we have this agent which basically gives context for a deal, right? It can give context so that agents immediately understand the deal stage and urgency and the business importance. So one of the things that can go worst in a deal for sales reps is the client is doing a POC or they're in a trial phase for 14 or 30 days and they send some questions to your support team and your support team answers that in, in, it takes their sweet old time because it's, it's a trial customer. They don't know how urgent or what plan they wanna sign up, how many agents they have. They don't have all of that information because the customer themselves haven't uploaded as many agents yet. So it's very hard to give context and the best context is there actually in your CRM. So what I would do is basically go in and let's implement an agent which informs the support team on how valuable this deal is to sales and increase the prioritization so that this is answered within the first couple of hours than putting it in the regular queue. I'm just going to click on Hiring Agents. So essentially we onboard an agent and we deploy it. And this agent specifically is costing us about 4 cents for completed task. Now completed task is essentially when we actually find an opportunity, right? When, when the escalation happens, the priority change happens, only then you're charged. So even if you have 2,000 tickets coming and this scenario happens only 10 times, then you are essentially sending— spending only $0.40. So it's truly not usage-driven but success-driven. And I think, um, so, and then I can build the agent, I can give it a name, uh, have my description. Now I can run this in supervised mode or autopilot mode, and in this case I want to run Autopilot. So anytime there is an opportunity detected, we are going to basically make sure there is a summary with HubSpot data on how valuable this deal is, what is the MRR on it, how much— what is the context of that deal— is actually sent to the support system. So this is great when sales want to make support aware of what's going on. And the other opportunity we have with another agent would be the other way around, right? So when, when, uh, when you have a customer that is, um, uh, that has a— basically raises a ticket, and, uh, this data is now sent to HubSpot, uh, to let them know that, hey, while you're talking, you're about to give this customer a call and he's just reported that the Workday integration is not working, right? And it's the worst time for you to give a call because that issue is not resolved, or you not being aware as a sales rep that they're actually struggling with the implementation right now and you're just pushing on that order form. So here the sales rep will get context that, hey, Happy Fox, who's trying to buy our product, has raised like 2 tickets related to Workday integration. And somehow that integration is not working for them, so they need to triage. So they can be more aware of it and I can make this work. So all I would do is, again, I want to run this in autopilot mode, select all my products, and then here I can specify what HubSpot action I want to take for updating the priority and basically set the priority level and, and save it, right? So deploy this agent. So one of the, the most successful though has been our expansion tracker, where we are able to look at potential expansion opportunities. So this has been the highest buck for, uh, uh, the money that we've spent, which is probably— we probably spent about $8 or $10 on this agent, um, and, uh, and this basically finds potential. If I hire this agent, um, it will run on all ticket creation and it will find all expansion opportunity. Now, HappyFox and a lot of SaaS companies have multiple additional products that come later in the sales cycle. Not everything is front-loaded and one big package is sold initially. And expansion, whether it's seat expansion, cross-sell, upsell, all of that is really, really important. So here you could, again, it will tag it. And the moment I come back to the ticket, so I logged in as a sales rep, I just click on this expansion opportunity. It just shows me only the tickets where there are exact expansion opportunities. So I don't— my, my system might be getting about 5,000 tickets a month, but we're only looking at expansion opportunity, which might be 13 which are still active, right? So it brings you all of the things that sales and marketing to look at without having to get thousands of notifications and trying to communicate hard with sales, uh, and with support teams and, and not losing context. Um, so, so going back to, uh, to summarizing all of these agents, I want to quickly walk through these agents again. Um, so we put the Expansion Signal scanner. We created a Stripe license request. So what we did with this is every time somebody said I wanted to add 2 seats, you make sure he went on a call and also upsold additional product. Kind of like they already added something onto the cart, we want to add on a couple of more potential things that they haven't considered. So we did this for revenue protection, sales acceleration, the example I talked about, and then the marketing team as well. So, um, yeah, so with that, and, and one of my favorite things which we lacked a lot was on the marketing side and just finding that single quote that customer appreciated a certain feature. Somebody just moved from a competitor, came on, and they said, hey, schedule tickets really changed the way we operate our whole IT operations. And that was like a huge— so we know how important or valuable that feature was to that customer, and we are able to use that quote on that particular landing page and hence creating a lot of unique first-party content, which was otherwise, you know, uh, marketing team had to guess to find out what real chatter was happening with customers. So with that said, I think I knew the demo was really, really short because it's super easy to implement. It's literally a 2-step process. And we do have a way where you can actually say, hey, Shalin, I have a new idea for what my AI agent needs to do on my support data. We are turnaround time for building an agent for exclusive for a customer or for our entire customer base on an idea given by a customer. Customer is about 2 days turnaround time, and we take it to production in 2 days and you get to use it right away. And generally the cost is between 2 cents to 10 cents per successful run.

Alan, it's Amelia. Can you hear me?

Yep, I can hear you.

Hey, cool. Can you go back to the screen? I don't know if it's your live screen or your demo screen where you were showing some of the different agent use cases. There's a few questions on that. Yeah.

Sure. So, so I can, yeah.

Yeah, the other one, the screen that was actually Happy Fox. Yeah.

Okay, got it. Okay, sorry. Let's go back to Autopilot.

Yeah, this one. Yeah, this is cool. So you were saying, I know you, you went fairly quick through the demo, so we have a few minutes though. You were saying, how do people mostly plug into this now? They connect HubSpot or Salesforce?

Yes, so I mean, now one caveat here is your support data has to be on HappyFox Help Desk, but if you are not doing that, you can still come and take a trial of HappyFox Help Desk or look at this presentation, steal our, all of our agent ideas and make sure whether you want to implement it via anything or Zapier or any tooling that you're really comfortable with. But here it just makes it first party where the data the system of record and the implementation is in one place. So if you're using a system that doesn't have this today, you can bring, you know, consider using HappyFox Help Desk so that you can launch this. And of course, the cost of operating that also like goes down because we are all using internal APIs to just, you know, go over hundreds of tickets. So the cost of us doing it on our own server is much lesser than doing it all by a third-party software.

Yep. Yeah. And if they don't, like say they're using Salesforce maybe just for like records but not for, for help desk or HubSpot for help desk, can they, I'm sure you have it 'cause a lot of these agents have similar functionality. Where do you have like an AI studio where they can upload like contacts and like kind of like talk us through how they train the agent?

So again, a lot of the— so a couple of things to clarify. You can use any CRM. We are— we are— so the HubSpot was an example, but you could be on Salesforce, any other CRM. And as far as you connect that within your help desk, so we have a bunch of CRM integrations. So if any of these CRMs are connected to your help desk, the AI agents automatically get access to it, right? So you don't have to do anything special. To connect the CRM after that. Now, in terms of guidance, every agent that we have built has, have industry context. So we've built a lot of similar agents, and sometimes, like, for example, we have agents that are built specifically for e-commerce companies that have e-commerce context, right? It knows what are the topics in e-commerce and all that. So you don't have to train like a general purpose AI where you have to set all the context. And we also take context from existing ticket data just to know what you do what you support, what your products are, what your tech jargons are. All of that is already considered. And then you have, you go in and for example, in this case, let me use a different example. In this case, I can give guidance here on what agents need to do or yeah. So there are some places where you can set guidance to make sure it's doing what you really, really want. Now there was an agent where I specified that I want this to happen for deals. Let me go to one of my higher agents actually. So I can go in here, I can see all the activities it's done. But if I go to configuration, I have a very simple prompt here which says deal above $4,000. Yep. Right? Now you can say complex multi-product deal or deals that involve this. So you can do all of your prompt engineering or context guidance here, and this will then just work on that, and it's not going to just general purpose decide something based on that, right?

So yeah, and can, can, can folks upload their own context? Like, say, you know, they have some stuff on their website, or they have some stuff in there, um, maybe they just have a bunch of Google Docs of like, hey, we have a bunch of like FAQs that are just Google Docs, Can they upload that to the agent as context?

Yeah, so yeah, all the knowledge can be brought into the knowledge base section of the product where you can bring in all of the knowledge context. And then if you bring your ticket volume, generally the moment you turn on your email to come to HappyFox, you're gonna get— start getting so much content for training right away, right? So that's a huge advantage where And, and we keep building all the con— as soon as tickets come, you know, your contact database is building all the time, uh, and you are— so all your customers are here, right? So you're— so in, in a SaaS company, especially for us, we've grown to $20 million, the cost of acquiring a new customer is still very high. The cost of retaining and cost of expansion is probably the lowest. And we felt that, you know, we could spend unlimited money in the land and expand motion rather than always trying to spend all of our top dollars for like extremely high, you know, not high risk, but high, you know, high probability of that thing not to work. And you have to attend to multiple things or just, just doing that very large volume to make it work. Here, this is like your existing customers. They want more things from you. Or some customers that are happy, they're telling you that and you wanna capture that emotion and, you know, take that to your next customer as well.

So, yeah. Related question, 'cause you touched a little bit, I don't know if you touched on this, of your internal structure, but you just mentioned, you know, you guys are at 10 or 20 million. I know you're bootstrapped. So this is all through, like, for those who don't know Happy Fox, that they're bootstrapped, like this is all through revenue. This is, you know, not, not a venture-backed company doing this. What— how does this then trickle down to how you've structured HappyFox, right? Because obviously you're using your own tool for support, but how does that impact your hiring across the board for like where you deploy maybe other third-party agents? And then obviously you have agents yourself because this is your own tool. How has that impacted like how— like what's your— I guess just walk people through like the dynamic of your team right now.

Yeah, so, uh, and this is something I mentioned in SaaS to London, that we were at $1 million, we were 5 support agents, and I think today we just hired our 6th support agent at $20 million, right? So we—

a 6th person—

yeah, so we just have one extra person going from $1 million to $20 million. So I mean, before AI, we did a lot of optimization with our automation tooling and We've been efficiency freak as a bootstrap company. And then on the sales side, we have about 6 sellers right now in the company, and we were at 6 sellers when we were $3 million revenue, I think. So, so, and my goal is all the time take away all internal communication, internal meetings, internal clerical tasks, administrative tasks, just offload as many of that. And for us, the amount of time sales reps spends with customer is important. And for marketing people, it's like we just built our first marketing team in 2024, I think. And we basically said, hey, we don't want to sit and train you on the product and all that. You have to learn from support data, and we just built a lot of tooling for them. If they're writing a new landing page for a feature, they get educated almost— they're supposed to use all AI just to educate, and it will refer to a knowledge base, it will refer to what competition does, and just do it in a completely self-serve manner. And, and we are a flat organization, not too many VP-level or C-level titles in the company, uh, and And teams are, you know, encouraged to just cross-function and look at data across sales, support, very, very freely to find their context.

So yeah, it's like, as I've gotten to know you, I think it's such a, it's such an interesting story that I think you don't give yourself maybe enough credit for to put out there. Like, you know, we just heard from Salesforce, you know, obviously Salesforce is huge. Google is huge. They were like the two presentations before you. And then some of the other, the other AI go-to-market folks that are presenting today are like more so venture-backed. You, I think other than Saastr itself, you might be the only other one that's bootstrapping today. So I always like to highlight that to folks because I'm like, no, like you're, you're, you know, you're probably the same-ish ratio as Saastr of like how many humans to revenue you have, which is kind of, you know, again, I think that's the, I think that will become sort of the new standard going forward. But it's something I like to highlight with you because I think, you know, sometimes I think because you are an AI-native company, folks can automatically assume like, oh, okay, they probably have like, you know, a bunch of, bunch of people working there and that's why. And they've got all this VC money. I'm like, nope, they haven't. Like, this bootstrapped, dude. Like, this is all like of their own, you know, fruition and hard work, which is why I like to, to call that out. Not to put you on the spot, but I think it's just good.

Yeah, and we have competition on how— who can spend me, especially me and the CTO and our head of success and our sales team, that who can burn more AI dollars and automate as many things.

So that's a good internal competition. Yeah, yeah.

And we, we all take ops role. I have implemented most of these agents myself along with sales team on the call saying, hey, what would be useful? So we've done it in a, like, a team effort, and generally it doesn't take time. And what to do becomes very clear. But as I hear ground reality and they see me actually sitting and implementing, yeah, they know that, hey, things move fast here. And I tell them, where else can I cut your work next, right? So you can—

yeah.