Building SaaS with AI: what works, what breaks, and what not to build

Welcome everyone. We're here today to talk about building SaaS with AI, what works, what breaks, and what not to build. And the session is hosted by Qrvey. Thank you all for joining today. First, I'll just do a quick start um to give everyone an idea of what we're covering today and how to uh talk with the panelists. If you have a question, please put your question in the Q&A and we'll be happy to get to it at the end. so first up, we'll have our special guest, Bill Hostman of uh Dresner Advisory Services. He'll be taking us through the AI platform framework that he's been using with clients. And um And then after that we will start the discussion.

Quick note about Qrvey. We're the leader in multi-tenant embedded analytics and we provide embedded analytics platform specifically for SaaS companies. They have the ability to use our software to build AI and self-service data experiences and offer that to their end users. And we are excited to also celebrate 10 years in business this year. So let's talk a little bit about why we're here today. Now, it's no surprise now with AI that you've seen features get built within an afternoon. But the challenge here is how do you ship all of these prototypes that you're developing? How do they scale?

And how do you maintain them? And how do you make better product decisions about what Is ready for customers and what is not, right? Because ultimately we're using AI to save time, but at the end of the day, it has to deliver value to the end user, everything you're developing. And it also has to be secure, right? It has to be something that you can live with as a long-term approach. So we'll get into the conversation today uh to hear exactly from other SaaS leaders how they're also using AI to develop their products. I'd like each panelist to just do a minute or less introduction to get us started if we can. So Bill.

Bill Bice, CEO of Zebraworks, uh legal legal tech company, spent a lot of time building software for uh for law firms and I'm I'm still still doing that today. Mahesh. Hello everyone. I'm Mahesh Raja Sekran, president and CEO of Cleo, an AI-native supply chain orchestration platform serving more than 4,000 customers globally across manufacturing, distribution, logistics, and retail. Previously I led the high-tech business unit at iTunes Technologies, now part of Blue Yonder. For the last 14 years, I've focused on making global supply chains work better. And at Cleo, the journey has taken us from integrating supply chains to integrating ecosystems to building network context and now intelligent orchestration.

Our north star is helping companies get towards the holy grail of autonomous supply chain orchestration. And today we're advancing Cleo on two fronts. We're using AI to transform how we develop and deliver software, and we're building AI-native capabilities. that help customers make better decisions and execute faster. So it gives me two perspectives for today's discussion. One on how AI changes the way we build software, and two, how it expands the value that software can deliver. Thank you. Arman. Hi, Arman Eshragi, CEO and founder at Qrvey. We are embedded analytics and AI for SaaS. Before starting Qrvey in 2016, I also started another company named Bloggy Analytics in 2000.

And before then that I had two other companies. uh Happy to be here. Fabulous. All right, let's get started. I'm happy to introduce our special guest today, Bill Hostman of Dresner Advisory. Um, in his work, Bill partners with enterprise clients who use a multitude of SaaS products, right? And a lot of the conversations that he's been having with these clients are focused on where do they build with AI and where do they buy um products and solutions. So he'll be sharing.

Yeah, more about his experience and the conversations he's had with clients. And there's actually a lot of parallels between how to determine what to build with AI and what to buy on the end user side, as well as how SaaS companies are also approaching their own product development strategy. So please join me in welcoming Bill. Thanks, Kerry. Thanks for having me. And yes, Bill Hostman. Uh I'm a research analyst with Dresner Advisory Service. I work with clients as well as vendors as well. and uh my experience is uh 26 years as an analyst. prior to that, I was also in product management and general management with companies like Informex and then a startup at that time, which was Informatica. I spent 15 years at Gartner where I covered, I was the lead analyst for Oracle and their applications, and then SAP and their applications.

And now I cover the uh AI stack. And the thing that I talk with our clients about and give them a lot of advice on, and we got a lot of engagements, is how do I think about the the the AI stack holistically? And we use basically the AI platform framework, something that we talk about, which is basically uh three components. One is the runtime layers, which consist of the use case economics, the agent and the workflow and the orchestration layer and the context and retrieval layer. And then we also talk about the foundation layers, the ontologies, the semantic metadata, the data and integration, the compute and infrastructure. And the area that's getting a lot of interest right now is the cross-layer control plans.

This is a hot topic in most of the clients that we're working with when they're going out and they're evaluating a lot of new technologies. And they're saying, Well, where's the governance? What's the state of the market with respect to governance? How do I observe these multi-u-agentic workflows? How do I secure these these uh these workflows. So we use this framework for them to help organize their thinking about do I build an application or do I buy an application? So a lot of clients are going through that build versus buy right now because they think, I can just get a new software development lifecycle tool. It's agent-based and I can build a lot of this stuff myself. So they're going through that rationalization. If you can go to the next slide, we can talk about the the next set of challenges, which is The changes in these technologies move at different paces.

And so the organization has got to learn how to align and integrate all these different technologies and stacks and disciplines at different paces. And that's a challenge for a lot of organizations. The other thing that they're recognizing is where is our power? Where is our buying power as an organization? And you know, the power is not in the flashy, exciting sort of presentation layer things, but it's actually down at that foundation layer. And that's where most of the organizations we're talking to right now are saying that's what we have to we have to own that. We have to uh you know either build it or buy it, but we have to be in control of it. That is our destiny. The the ontologies, the data integration, the computer and infrastructure, having that foundation being the correct foundation for the consumption by all the different kinds of applications is the area where most people are really struggling.

And also on the governance layer. Governance is very slow. It's almost always a uh a technology interval behind the the actual applications themselves, and we're seeing that right now. The next slide, if you can, is I get asked a lot, well, what about vibe coding? Why can't I just vibe code these different layers myself? And you know, why do I have to go out and buy something when I can just build it myself? And we're seeing organizations that haven't been in the building business for quite a while standing up new teams saying, we're gonna get rid of all of these expensive uh applications that we've been paying a lot of money for, and we're just gonna build them ourselves because we can buy code them with these new SDLCs. And so I think we go through this exercise with them and and suggest that there are places for you know doing that kind of coding yourself.

And you have to really look at it at the different layers of the framework. When you look at the use case and economics, great way to get prototyping, great way to get proof of concept. But when you start to get into things like context and retrieval and ontologies and semantic and the data integration layer, it's more confusing. It's not as straightforward in terms of I can just I can just vibe code something. So we're trying to get them to realize that there's more than just the presentation layer and sort of the consumption layer. There's all these other different layers that have to be integrated and aligned. If you're going to build something yourself, and that's a challenge that's daunting for a lot of organizations. If you can flip to the next slide, the topic that really comes in the biggest is when they're building things, is they're realizing they're exposing a lot of security surfaces that they hadn't exposed in the past.

And that's causing a lot of heartburn in terms of the amount of code that's being developed. They're getting saturated in terms of trying to be able to examine it and inspect it. and so they're getting actually saturated by the amount of code and they can't the the the security checks and the governance checks are are quite a bit far behind. The other thing that we're finding in in terms of vibe coding is, you know, the the observability and observing and debugging across multiple platforms is a real complicated area that most vibe coding hasn't yet addressed, or most debugging tools haven't addressed. The ecosystem and integration is the last place where people really are getting tough, where it's like, do I build this or do I just buy this? So what we're talking to our clients about is use this AI framework to do an inventory of what should you be building, what should you be buying.

Think about the pace and the power and the change of these different layers. Understand there's different business risks and different business risk risks at different layers. uh Use vibecode to learn, buy, buy to run, and own what laps. Own the things that where the power are, power is things like the ontologies, the data integration, that foundation layer, and the governance and security and observability layer. Keep L5 fairly swappable. There's, you know, it's getting commoditized very, very quickly, but own the data and the meaning in those lower levels. Um, that's that's really the essence of the conversations we're having with clients when they ask about should I be building this or should I be buying it? That's fabulous.

Thank you so much, Bill. That really sets the stage for the rest of the discussion. And I think we'll probably find some similarities and also based on our experience from our SaaS leaders here on what has worked and what has not. So thank you again, Bill. We're now ready to get into the heart of the discussion here. Uh so the first question when we're talking about building SaaS with AI, what works? What has worked for you? Where is AI creating real value in your SaaS product development? Bill Bice, would you like to kick us off? Yeah, so it's it's it's a great accelerator. And and I mean that in both positive and negative terms. So if you have a really great development process and you're building on a great foundation and framework, then AI is going to speed that up and is going to really enhance your development process.

Uh if you don't have that foundation, if you don't have that framework, if you have a mediocre process, then it's going to greatly speed up your mediocre process. Which is exactly why I'm not a fan of of vibe coding, because that's essentially mediocrity on speed. And so great for prototyping in order to see things, but it's it you just create a lot of stuff that never goes anywhere. There is value in doing that because it is wonderful to be able to sit down from a design standpoint and actually use something. That's why you've always been able to prototype things in in Figma. This is just a better version of that. It's that next step of where You think because you've built something that's basically a prototype, do you have an application? That jump is where the real issue is.

So the real value in AI is if you've got a great foundation and you have uh a really good solid development process and you energize it with AI, well, that 30% of your development process that was really writing code, that's now that's now taken care of. And so it just changes where the emphasis is. Your design's faster, your writing code is faster. So now the architecture, the scalability, the security, the design choices about what goes into the product becomes so much more important in the in the process. So the problems we've long had about product bloat, like that's just gonna get much worse.

So the companies that are really good are gonna be the ones that truly understand how to solve the problems for their clients without just like throwing everything up against the wall because now you because now you can't. Mm-hmm. Absolutely. Thank you. Mahesh, do you wanna go next? Yeah, th thank you, Kerry. um You know, for us, we're seeing AI work at two levels. One is accelerating product development, and second, enabling better customer outcomes. from a development perspective, we started initially a couple of years ago with coding assistance and then progressed agent to coding. And as coding accelerated, uh the bottlenecks moved into requirements, into testing, into verification, and how do you deploy in production. For example, we're now building and built a development harness that connects all those stages with very clear specifications, automated checks, and human decision gates.

We call it SNAP. And here we're able to get uh any code developed by any developer anywhere in the world at Pleo in production within 30 minutes with all those checks and gates done. And we're able to actually build this capability in two months against an eight-month estimate. So uh another example, we delivered a uh very important critical module called Frade Order Execution in three months against a nine-month estimate. So that's those are huge. And from a customer perspective, everything I said so far is about Cleo. From customer perspective, AI creates an opportunity to interpret thousands of supply chain signals at scale and act in real time while there is still time to make a change to change the outcome, right?

Which means As an example, we can identify an order at risk and take action before it triggers a chargeback. It can be using warehouse readiness signals to improve trade execution, reduce dwelling tension, uh, and free trucking capacity, which can be monetized by asset-based carriers, uh, drive additional revenue generating loads. Um, same thing in procurement. We can look at procurement risk risk across the entire supply base and apply dynamic. information based on evolving situations like the state of hormosis shutting down or you got tariffs rolling through the economy and make sure the right supplier commitments are in place so we can help customers prioritize exceptions and improve execution across the supplier base.

So these problems have always existed. uh And so what AI has done is it has really expanded our ability to address them at scale and at a speed that was previously very difficult to achieve. So I would s sort of summarize this question on what works into two buckets. There is value in two fold. We can deliver capabilities sooner. That's one. And second, those capabilities can help customers make far better decisions and act sooner. Right. Yeah, absolutely. There's a lot of power with proactive alerting triggers and actions in in your industry. Absolutely. Um Arman, what's working for you at Qrvey? Yeah, so as uh you know a number of uh points that rightfully so Bill and Mahesh pointed out.

And I would say, you know, in summary, we are using AI in order to save time, number one, or uh provide a better product. So AI by itself, I think one of the most important roles that it plays, it would increase the quality and it will increase the in a way. the bar. So right raising the bar, it's just, you know, it just will put it in a place that maybe in the past market expectation was here. Now they are going to really expect much more. And it will improve the products based upon that because everybody needs to create a better product at the faster time period. Now if we look at you know where really AI can create that kind of value, Definitely it comes to mind like prototyping, for example, as Bill mentioned, right?

So it's the time that you don't need to review to make sure everything is correct. So that saves you that time. But you can't really bring those concepts to life. As a real example, you know, our UIUX designers used to create 250 slides and then show us the flow of what happens if you click here and if you don't click here and click on the other side. But now They're really in a much better way and faster way. They can really put together the prototype that looks like so real that you can really play with it. And what it does, in the past, we could afford only to have these prototypes and concepts to show to customers and also internally collaborate on it once a quarter, maybe twice a year.

But now we do it every month. So that acceleration is a real value. much shorter time bringing that prototype. But I would say that's where I see that when you save time or you improve the quality or the value of your product for your customers, those are places that AI plays the much bigger role. Um and if it doesn't, then those areas are not the good areas to implement AI. Yeah. I mean in fact, Arman, just a quick example, uh just piggybacking on what you said about balancing speed and quality, I clearly across a broader development transformation, our releases have increased 46%.

Right. 46% increase. And uh what we call the production incidence, when you launch new stuff, it fell by more than 80%. So when you measure speed and quality together, it becomes powerful. Right. And for that you you definitely need to have the right harnesses, the right reference data data set, what we call the golden data sets, so that you can balance answer quality with operating cost. And that's been the biggest benefit we're seeing to date. Yeah exactly. well, you end up you end up with a much much higher quality uh product, a lot less bugs, a lot less rework. It's really it's really quite impressive the results. Yeah, and one one more example I can mention is that, for example, on the QA side, we first started utilizing agents, but then we realized that there are some cases that maybe agents are not as predictable as we thought.

And sometimes, you know, we need to double check what agent is doing and it doesn't save us time. That's another example that we said, okay, if it doesn't save us time, why should we use them? Instead, we used AI to really generate the code. that we can't really go and review and then we ran the code to test the application. And that resulted very well. So by doing those kind of tricks, then you can really save the time that not every single time you have to review the result. You have used and leveraged AI to really generate something that is so deterministic that you can use it without checking it every time. And now that's the impact of AI, right?

So Mm-hmm. Yeah, I mean this entire year, you know, you can just tell with the time and they've spent on experimentation and trial and error. It's not always that things aren't going to work right away. It's how do we get to know this technology and develop it and work with it and tell it oh how it should be delivering what we want. So great advice here on the what works. Let's jump over to the breaks. So where does that speed of AI assisted development really begin to break down? What kind of experiences do you have to share? Who'd like to jump in first here? I I can Kerry. I think Bill touched upon it uh briefly about uh prototype production.

So what I see is where things break down is in the transition from a compelling prototype to software customers can depend on at scale in production, right? AI makes it much easier to demonstrate functionality from a UX perspective, from functionality perspective. But once that functionality enters production, We as a software company have an ongoing commitment to the customer. Can it scale with transaction volumes and customer complexity? Can we secure it? Can we monitor it? Support it? Can we add features without disrupting existing workflows? Can we build and launch uh upsell modules that fit into coherent production experience? So those responsibilities continue throughout the life cycle of the product.

Architecture and engineering discipline. Now determines how effectively we can meet them. So for us at Cleo, faster coding made those broader requirements more visible. That's why we're investing across the entire product development lifecycle, the whole, the AI, PDLC, including how engineers learn and retain and understand the system they own. So my CEO test is very simple. Can we ship it, run it, evolve it, and grow the business around it, right? So for us, production readiness includes readiness for the next change. We need to make the next 10 enhancements safely, right? And therefore, our engineers approve implementation and testing plans, automate the security checks throughout the process, and do the final verification. So when something fails, the RD team must be able to explain what happened and restore service.

And that accountability stays with us as a software company. There's a huge TCO issue here. So we have it as a software company. You're gonna build a new area in your application. You're really making a commitment for something that you're gonna that you're gonna support long term. And so just because you can build it doesn't mean you should. And then we very much see this within our clients, which who have the same problem. So now all of a sudden you have this ability to solve all kinds of problems that weren't really realistic before. And and What you see over and over again is you have somebody who is who is you know technically sophisticated, isn't really in IT, and goes around and solves a bunch of problems in a in a department.

And then that person moves to a different department or leaves the company. And it turns out that you become very dependent on this set of of uh unofficial technical uh underpinnings that nobody understands, nobody knows how to support, and you've created this long term TCO issue that no one is managing. And so there's going to be a backlash over time to all of this citizen developer work because of the problems that it's that it's going to that it's going to create. And the companies that are smart are will are getting ahead of this today and understand we've got to put structure around this and understand what's going on because everybody just uh You know, cloud gone wild inside your company is not necessarily a great thing.

Yeah, and and simply uh, you know, we need accountability, right? And we need ownership, right? So you cannot bring AI agent to the call when the customer tells you, I need to better understand how this works or why this works this way. Or even worse, there is a problem. Somebody needs to explain to me how can you fix it. So s someone needs to in the organization take the ownership and say, I understand everything. And I have the ownership and I take, you know, I I'm accountable for this. Now that means that as you generate code with AI, as you automate everything with AI during the development process as a SaaS company, you also need to be in control, review everything to make sure everything goes well.

And the person in charge says, I understand everything, I can take responsibility. What it means, it means as you are closer to the concept that The number of hours for that review, for approval, making judgment, making decisions, the number of hours goes down, the value of AI moves up. And as you get closer to the customer and to the production, and you get something that you want to really give your customer and customer uses, those number of hours, if you are not really doing it in the right way and you are just putting it in a kind of autopilot and say AI does it, everything, then it gets to the point that it breaks. Because then the number of hours of review and everything, either if you do it right, it just doesn't make sense because now you are spending more time, or if you are not really doing it, then you are going to really create a you know bad situation.

It's just a matter of when, not if. So that's really where I see, you know, something can break as soon as you lose control and just put it on autopilot and not really lose that ownership and lose that kind of uh you know uh w I know everything I can be on the call and explain to customer what happened and what how can I fix Yeah, yeah, I think beyond the prototype, I think the four foundational aspects for for companies to look at is can I secure it, which is really important, right? Data can can can be a massive uh leak. Then can you scale? Because the prototype has a scale. Third, can I manage the life cycle? Once you show something, people are gonna ask more stuff. Can you can you keep on managing the life cycle?

And ultimately the governance aspect, that do you have the right e-vals and governance? So that it's it's doing things that is supposed to do and it's not hallucinating. And you avoid model drift. So those four when you look at it, uh back to the TCO topic, you want to do it right in in production. Yeah, absolutely. Let's jump into now how do you decide what to build with AI and what not to build? I'm sure these conversations are happening constantly, not just at your companies, right? Everyone's having them. Uh so love to hear a little bit more about how you you make those decisions. Well, I think one the big one of the big pieces is that you you want to have like the more you give AI to work with, the better the results are.

And so it doesn't really make sense to so yes, you can literally go build everything with AI, but why why do that? Um like you're not it doesn't make sense to go build your own agent orchestration system. There's a ton of them out there. Why go build yet another one? It doesn't make sense to we're we're uh we're here with them on. Like why would you build yet another data warehouse visualization BI uh stack? Doesn't it doesn't make sense. You want to take all of those core pieces that exist that are well tested, where someone is putting all of their effort and all of their time with AI into building out those pieces. Because when you put all those elements together and they make up the foundation that you're building on, your efforts with AI suddenly become vastly better. Because AI is a is a is a pattern matching extending system.

So the better the foundation is and the better the set of patterns you give it to work with, the better the results are that you that you get. Yeah, I mentioned the kind of saving time and also I mentioned the kind of, you know, you wanted to create a better product, so you use AI to do that. But there is also another aspect of AI that is about really, okay, uh, I need to save time and I need to create kind of better product on my case. But also I need to make sure that whenever I use AI, it's not just adding too much complexity to what I have. So I can really say you know, in control and just build something that I can really further manage it myself and go with it.

Uh the focus of the business is also another important point. Just because I can really go and just prototype something doesn't mean that I can finish it. Uh Bill Hostman uh very uh you know in a very uh intuitive way Showed us the framework that we can see for any application, those layers exist. Now, some of those layers might be hidden at the beginning, but it's never hidden when you want to put the product in front of the customer, governance and security and other things that you have to do. And you may not see it day one, you may just see a few charts. But what if that is bigger than just adding a few visualizations? And that requires the good foundation with all of these layers.

Are you? really willing to build the whole layer, even if it's outside your core focus of your business, I wouldn't do that. So I I would say what should we not build is as important and more important than what we should build because the end goal is to really solve the customer problem. So with that clear NordStar, we should absolutely avoid building AI features simply because they're cool and fashionable, right? A chatbot earns its place when it helps the customer achieve a clear outcome, not because the chatbot is pretty s pretty cool to demo. So we also don't need to build every agent or every supporting capability ourselves. Where it makes sense, we should incorporate custom-built or third-party agents or use established agentic frameworks and take advantage of embedded analytics.

So at Cleo, our supply chain use cases require orchestration across companies and systems in a multi-enterprise supply chain. And so we're investing in the platform capabilities that connect. Business context, the semantic models with the intelligence and the execution needed. And we also use frameworks such as MASTRA and infrastructure such as AWS Bedrock underneath of it. But the right balance depends on your product and where you create differentiated value. Every component you build uh should create an ongoing responsibility. And it creates an ongoing responsibility to maintain and improve. And so the question I asked is: what do we uniquely need to own? To deliver the customer outcome and what can we incorporate to get there faster.

Right? A useful chatbot can help someone understand an exception and resolve it. The outcome should justify the features. Or, for example, supporting agents from different sources is valuable. At the same time, it requires clear permissions, shared context, and accountability. So buying a component still leaves us responsible as a software company. You know, we have the responsibility of how the complete product pl platform works for our customers and take the total ownership of end to end functionality, the security, the scalability, and not blame an embedded component. And so we still own the responsibility. Um that's very critical. Mm-hmm.

Yeah, for sure. So, you know, Mahesh, you really talked about if it's something unique to your product and you're the only one that can provide the most value, then you know you should build it. You know, Bill also made a point of look, why and reinvent the wheel? There's a ton of really great products out there that can just speed you up, give you a great foundation to build on. So I kind of want to go back to um, you know, if you decide, you know what? We don't want to build this in-house. There are other good products out there. How do you make a decision on which vendor you choose? Are there specific questions that you're going to ask that vendor about how that product will integrate into the evolving AI uh landscape and how your product will evolve with AI as well?

Any questions that come to mind? Well the obvious thing is you you just want a a hundred percent AI focused, API driven, MCP oriented platform, whatever whatever it is that we are we are talking about. And so if it's a hundred percent API driven and uh and and all of that is exposed via MCP, then uh you're gonna be able to to leverage that platform via AI in a in a very straightforward and direct way. Yeah. Yeah, I I would I would add to to what Bill said, which is which is um absolutely correct. More in terms of longer term viability of the vendor.

Are they, you know, uh is it something that a hyperscaler can actually provide you? Is it something you want to go to a a vertical AI company which which is going to be there for the long haul? So uh and then do you have alignment on on the long-term value creation? And then of course the technology fits. So I think uh these are important choices. Uh and if you do them uh without a careful set of um understanding and filter and prioritization, um you might end up in the long haul through the life cycle, a lot of mismatches and and and changing vendors, right? You you want people who are providing embedded capabilities to you to understand your north star and work with you. And that's important.

Absolutely. All right, great. Well, I do have a a closing question for everyone. And then we do have some some questions in the chat too. so first uh we wanna close out. What are you building next with AI? Who'd like to go first? I I can go. we're we're super excited about um where we headed with um with supply chain orchestration. The next big frontier for Clio is order orchestration focused on perfect order delivery, ext extending all the way through to on-time payment. So we're extending on our chargeback prevention capabilities to connect supply chain execution with financial outcomes. That means looking across orders, shipments, delivery, and invoice to identify issues.

That could lead to reductions, disputes, or delayed payment and orchestrating action before they affect your cash flow. And we think that's extremely important from a customer perspective: the opportunities to protect revenue and improve working capital by closing the gap between fulfilling an order and getting paid. So to us, it's a very natural next step towards our vision of autonomous supply chain orchestration. But we we we work in in finance in law firms from mid-size up to the to the world's largest firms. And so it's really important when you're working with financial numbers that when you when you put stuff in on one side, you get the same results on the other side. Which means that that machine learning is really the right uh form of AI for a ton of the work that we do.

And the thing that is uh there's so much fun in our development process is that Gen AI is perfect for really speeding up the process of doing ML work. So the ability to ah to to build out ML driven algorithms, to test them at scale, to run it against the the data for you know, the uh a number of of of clients were previously just with the level of compute and the time that it would take, like you can do it at a scale today. And and some of this is really compute as opposed to just AI, because the the two really go together well. Like a lot of the advances we have in AI have really just been driven by the fact that we have so much more compute available. And you put those together, that means that what we can create in terms of the results from an ML perspective, they're they're really impressive.

So I'm just loving the progress that we're getting on uh on that front. Yeah, on our side, on the curve side, uh it's interesting to see, for example, there is a fine line to walk between making the product or keeping the product simple yet powerful. So when you add too many, for example, when it comes to the simplest example I can provide, just changing the look and feeling style of your dashboard, for example. If you make make it extremely flexible, that means there are too many attributes available and it makes it very flexible, but then very time consuming for someone to go there and just modify.

But it will give them ultimate power to really change everything. Now, in reality, we had to always make and select a very well defined set of attributes. Let's say, you know, fifty of them to really make sure that users can just, you know, just adjust it and make the look and feel. Now with AI, we have created a very much bigger list. Let's just in addition to that 50, we have maybe close to 1000 other attributes behind the scenes sitting somewhere in a JSON or something that users don't have to deal with it, humans don't have to deal with it, it's hidden from them.

for them, but it's uh it's something that the agent has been trained to really look at those and stay with them. So as soon as you ask this highly trained agent and say this is the way I want my you know dashboard to look, even just point it to a URL or give it an image, it just goes in a matter of seconds or maximum minutes, come back and say all these adjustments done and The agent is actually doing with all of those attributes, hundreds of them, that otherwise we would not have been able to provide those to humans because it would have made their life much difficult, more more difficult. So so that's a good example of, and it's very much look and feel so they can see it if they approve it. Yes, if they want to redo it, if they want to undo it.

It's under their control. So they can really look at everything and say, now I like it, change this start or go to prompt and fine-tune it. So that's A big time saving for many of customers who want to really change look and feel with every deployment they have for every client of them. And uh another quick example would be self-service. There is a rule of thumb that if you have really more than five clicks, you are going to lose your users. So the user adoption goes much lower in the self-service manner that you want to empower your end users. And if you give them something they can do on their own. But it takes more than five clicks to do it, most people would not do it. I'm not talking about power users, I'm talking about very end users. And that kind of five click is puts a very hard limitation on glass ceiling, because in many cases you say, no, that's too difficult for them.

I'm not going to give this ability to end users to do that. So you don't do it. So it's less self-service. And because it's less self-service, then in a different way, users won't like it. But leveraging AI. Now you can really pass that five-click limit. And in some cases, you can let AI to help humans to go easier, simpler, but also type something that maybe AI takes the steps for them. So from humans' perspective, is it still five steps or five clicks? But AI is putting some of those clicks for you. And it's still it's easy and quick enough to be counted as a good self-service capability. Both of these enhancements are part of Q4 for Qrvey and then you will see it at the end of Q4.

I actually have a question for both both the the two CEOs, Bill and Arman, as well as uh Bill Hostman, for your um you know, your uh perspective in the industry. one of the things I see as an opportunity and a challenge is that there is so much of AI, um, you can call it AI signals and AI noise going on, right? A lot of you know, vendor blurs, people talk about AI native, AI powered, AI-driven, AI can sing and dance. And As CEOs of companies, um, we get we are in a you know in in a position to to not only accelerate internally what we do, but also grow the business and really separate ourselves from all of the channel going on.

So what what do you do as CEOs uh in your businesses um for that? And I'll be happy to add what we are doing to you. Just to clarify your question, you mean w how do you deal with this noise or how do you really keep up with the market? How how how do you rise above the noise? Right? Because anybody can you put a chatbot, they can talk about AI, which I'm thinking the in the audience you've got lot of SaaS CEOs and CTOs. And they're building a lot of great stuff, but then they're also seeing competition throwing things which may or may not be real, but you can do a few things quickly and and put it out externally, right? So there's a lot of what I call vendor blur and website blur going on.

Um and so we have the responsibility to elevate our our companies in the in that A AI wave and security as how both of you are navigating at your companies. If I uh if I go first, um honestly the way I personally see it is that for everything, there is a hype level, there is a hype kind of cycle. Um I I was just asking uh one of these AI, I don't know which one, say I asked Chat GPT and said, can you give me the number of blogs posted uh with regard to uh you know, for example, one of these topics that is hot about AI.

And I went with a few topics and I asked different number of blogs written. And you could see the hype cycle. That, for example, when it comes to you know, some of these, the number of blogs posts goes up in February and then goes down. And you could see exactly the kind of questions and answers that you see in the sales demos. It's very much relevant with that hype cycle. Now. The personality of a company and the culture of the company, in my opinion, should dictate and should provide the kind of guideline that if you are a company that you wanted to move fast and be the first, or you wanted to you are okay not to be the first, but you want it to be the best and the most pragmatic. And that automatically, you know, you see with Apple versus you see with some other companies.

Apple doesn't mind not to be the first, but when they provide a product. maybe a year after everybody else has done it, but it's still you may trust it and you may use it and you are happy with it. And honestly from Qrvey perspective, I would say I'm not trying to be the first. So it doesn't bother me that we are not the first on making these kind of noises, but we would like to understand it, be pragmatic and be really something that customers at the end, when they see the product, they like it, they use it. And uh we get feedback and be implemented, we might be a few months behind, but we'd rather to provide something that is meaningful. That's that's great, Arman. Bill curious about your business.

Yeah it does. definitely gotten pushback on uh on on our approach which is which has been just to focus on on building the best product and that that doesn't include slapping some AI features and and getting it out the door. So uh that that does you know the disadvantage of that is it does allow competition to get out there and say, hey, we've got the AI whatever thingy um and then the The reality is that doesn't accomplish a hell of a lot. And so I'm I'm just I'm confident in winning in the in the long run and building and building the right product. And I I think that's the only the only route that really that really works.

So we'll uh we we can do this session every year and we'll we'll see if that uh if that proves up. Yeah, th that that's remarkably consistent. you know, for us, the North SOR is always creating significant customer value. And if you look at supply chain, AI gives massive agency, right? There are things we're able to do today, which would have been very hard to do at scale in real time pre-AI. Can I would not say you cannot absolutely do, but you know, there is a big difference from automation to try trying to go towards autonomous. And so for us, it gives us agency and it gives our customers the ability to capture significant value they were leaving on the table, almost to the extent of doubling the profits.

And so when we you know, I always tell our team that when you're focused on solving the hard problems and leverage AI wherever it is needed to speed up, you're just rising yourself above the noise. And everything else, um, you know, the the companies that don't do that. will have to get disrupted, right? The AI economy is going to create significant winners and others who can't keep up. And I think it's a good thing for companies that are focused on solving hard problems and unlocking value for their customers. So it's pretty remarkable responses from everybody. Great. Yeah. Yeah. A great question. Yeah, that's a good question. Thanks, Mahesh. We did get some other questions in from the audience. so one of the questions is, you know, since everyone uh is has deployed AI in their products, this person is wondering how can you get rapid customer feedback on what's being deployed and how are you engaging customers um on those AI features to continue to iterate for a better, more valuable business outcome?

Is there anything different that you're doing from your normal? Customer feedback process. I I can jump. So there are two specific things we're doing differently now. One is our um our user boards, user advisory boards. We do it by industry. We have a manufacturing advisory board, we have logistics advisory board, um, retail advisory board, and have them uh we actually invite them uh to to to Cleo or we go to them. Um we we do a dinner the previous night, talk about topics and then the next day walk through prototypes uh and get feedback. Right. So that's That's been very, very powerful. And the second thing we've done much more intentionally is have the have our AFDE teams, the forward-deployed engineer teams, go and start mapping ASS and 2B process with customers.

And so then you realize there are things that might not be available on a platform, but those signals may be needed. Uh in our case, we might have to get into a SAP instance or a Salesforce uh instance or or some customer support product like a Zendesk. be able to get those signals and triage the the process. And over time, once we see patterns, we bring them back into the product. So it's a combination of industry-based advisory boards giving rapid response and prototypes, but also having the MD teams have a loop back into product management to get um proven capabilities into the platform. You know, I I think senior leadership in the company, exposure to customers is absolutely crucial.

I spend mo you know, I spend a ton of my time talking to uh talking to clients. I think it's really important for for the executive team to for everybody on the executive team to be doing that. And I've long found that you don't have to be expansive. Like it's better to go deep with a smaller group. If you have five representative clients and uh and you're going deep with them, you're get 95% you need to know across your entire client base from those five. And that's gonna be much more valuable than what you will get from a survey that is from 200 clients. Mm-hmm. Yeah, as as you know Kelly better than me, uh because you manage you're part of some of these groups as well. And we have these caps that customer you know advisory uh kind of uh groups that we have.

And then we really go there and with a smaller group as uh you know Bill mentioned, four or five people, not more. And then we have a better option now with AI because we can really show them a working prototype, not just some slides. And we can bring better discussion to these and ask for feedback before we really go there and just implement things. And we can do it more often than before. So these are kind of the impacts that AI has had with regard to interactions with customers, asking them about the capabilities and incorporating their feedback into the product side. That's great.

And yes, you're right. I am very much involved in that. And it's actually the one of my favorite parts of my job to hear exactly from the customers what they're liking and what they'd like to see. Uh the other question that came in here, um, generally we're seeing the buzz that SaaS is dead. On the same note, when we see clients creating applications by vibe coding, you know, so it's hard to to balance both. So from all of your perspectives. How do you see the future shaping up for SaaS at this intersection of AI? I have strong conviction. I would say software is really becoming the chips of the AI economy, right? You've got frontier models that are getting better continuously. You've got agentique frameworks evolving, but then you need context.

You need the semantic models of the of the data and the use cases. Some of the points Bill Hosman talked about. The integration, the context, the semantic modeling. then only through that you can drive intelligence. The intelligence could be traditional machine learning, you can get, you know, Gen AI uh frontier models. And then you need a whole bunch of work around governance and eval before the agents can can pick up and and drive the agency execution for a for a workflow. Humans have to be pulled in based on the agent based on the governance and eval. And you need lifecycle management. So to make all of that, I think software is so fundamental and foundational. to bring the power of AI, to really unlock this productivity benefits that AI is is is expected to deliver, software becomes a vehicle to deliver the AI benefits. So I'm I'm more convinced than ever before about how much software is gonna significantly take off in the era and age of AI.

Love I love this question. And the real the I think the core answer is does your application have so much value that if it's used headless, the the your clients still have to have it? So is the inherent value in that intelligence layer and the workflows that are in there, if no one ever goes into the application, does it still provide so much value because it's all driven by AI that it's worth the worth your clients buying it? In which case You have absolutely no nothing, you know, you you have a you have a great future in front of you. If not, then you you should be contemplating how to how to change that. Uh it's really that CCO discussion because right now we're in this sort of uh great pause phase where companies are saying, Well, yeah, I can build that with uh with AI, and they're only starting to realize what that means from a TCO standpoint.

And so it's it's all going to come back as long as there is trem as long as there's real value in the intelligence and workflow of of your of your platform. Yeah. And I would say, you know, we uh it we used to as a we used to build uh software products using data and algorithms. So software and also, you know, using the data. These two elements were kind of the ingredients for us. Now we have AI as the third ingredients. And of course we are kind of understanding how to use AI in order to build this three entity now. kind of tree ingredient kind of, you know, food that is in harmonious and is really tastes well and the market likes it.

And if you really look at it that way, uh some some software makers may put too much AI into it and they may try to replace human with AI so that decision making is done by AI. I don't think that would work well. But if you are facilitating it and if you are saving time for humans to make decisions because AI is there, and then it will work well and it will taste well and you will get a delicious food out of it. So that's the way I see. That's great. That's great. It's great advice. I think this was a a very dynamic panel. Uh we've got a ton of takeaways that we've been taking notes on over here on the side. Um this has been fantastic.

I appreciate everyone's time and for sharing their afternoon with us. Um and at this point, we're ready to wrap up. So closing thoughts are Let's chat, right? So a lot of good discussions today. Our panelists are all on LinkedIn. You can go ahead and connect with them there. Uh they're happy to continue to talk about this. Um, so feel free to do that. And if you're interested in what we're building here at Qrvey, uh, with our focus on embedded analytics designed specifically for SaaS companies, AI, you know, gen AI and agentic experiences. If you're struggling with how to deliver those in your own product, we'd love to chat with you about that.

So uh on behalf of everyone at Qrvey, and I want to personally thank Bill Hostman, Bill Bice, Mahesh, and Arman, thank you so much for joining today. And thanks everyone for joining on the call. Thank you, Kerry. Thanks everyone. Yeah. Thank you.