The road from AI prototype to product

Hey everyone, welcome. My name is David Abramson. I'm the CTO here at Qrvey and I'm pleased to be sharing some thoughts with you on AI prototypes to product. I'm going to jump right in and share my screen and we can get started talking about uh this very sort of interesting and preient topic. Um, I think one of the things that we're seeing is how AI tools can be an amazing prototype accelerator, but where we're going to think and focus on is how um we can get those AI built features um to close the gap between what makes a successful prototype and getting your product production ready. Um, just real quick, I wanted to give a little bit of background about Qrvey. Um we are um a a specifically built um embedded analytics tool for SaaS companies who want to provide a rich set of analytic capabilities to their tenants and users.

Uh and this is 100% of what we do. So all of our customers are SaaS companies, SaaS product teams like yourselves who are looking to provide rich set of of embedded analytic capabilities to their customers, tenants and users. Now, um, AI coding tools have absolutely fundamentally changed how product teams build. Um, I'm sure many of you have seen, um, working features come together in just an afternoon or maybe even a couple of hours. Um, so the the speed is definitely real. Um, and the question is, uh, what do we do with it? Um but this is this is kind of the story that's generally the the really good news. Um for sort of idea and feature validation um AI tools um can be some of the best options available. Um lowcost uh and the ability to show stakeholders something real instead of you know just specifications or mockups um as well.

Now, um the proof of concept, however, is is really designed to answer the one fundamental question. Could the feature that I'm building work? Um a product has to survive the prototype. Um and the jump between sort of these two boxes here, um is really where a lot of teams and product development can stall. Um and this is not just me or Qrvey speaking here. Um, I've actually been posting and talking a lot about this, not only in my, you know, LinkedIn, uh, but also to other SaaS leaders about this topic for the last several months. And everyone's sort of saying the same things. Um, figuring out how to close that gap between kind of the AI accelerated prototype and getting into a full production ready solutions. Um, so I think that there's a trap here that we don't want to fall into and that is um sort of that AI built prototype um trying to convince us that we're already

there. Um, and you know, I think there are a lot of things that we want to make sure that we're not overlooking when it comes to, you know, how do we secure the solution, how do we scale it, how do we support it, and all the uh the customerf facing requirements to get to production ready, which is incredibly important when launching new features. Now we're in the analytics space, customerf facing analytics space. So this is actually a perfect example for using AI to get to sort of that proof of concept prototype stage because um I think one of the easiest things to do is to start creating these outcomes. Like a chart can just appear the sample data that you're working with looks correct and it sort of feels like it's ready. Um but you know every um concept really needs then to sort of um solve for the main challenges that those AI prototype tools aren't really going to get you um into

that sort of production mode. So I think what we need to think about is what are those challenges? What are the key challenges that we want to not ignore when we're going from sort of that prototype phase whether it's with customerf facing analytics or any of the other tools that we're working with or any of the other feature sets that we're working with within our products uh that we want to get into production um as well. So um let's talk about um some of the challenges that we see and that we've heard from other SaaS leaders that sort of present themselves when you're starting from that AI powered prototype and trying to get into production. Um probably the first major one um that we're hearing often is all about handling security. Um security is generally the first thing that the prototype skips out on and and the last thing that you can really afford to get wrong, right? Especially in something like a customer analyt customerf facing analytics solution. Concepts like record level security, tenant level security

have to be and must be enforced um for everything. So every report, every dashboard, every drill down, every chart that you build has to enforce that security logic and making sure that when you do actually go into production, uh concepts like that multi-tenant security um is is fully there is fully baked and is not overlooked from that sort of AI generated code or the prototype that you started with. Um challenge two that we also hear quite often um from the folks that we're talking to uh is all about how do we get it to scale. Um I think one of the things that is appealing about the sort of AI generated code especially for a proof of concept or prototype is that um it's obviously going to be um written and using sort of the data that was made available to it.

Um, so the queries that you're running, they look great when it's working off of a sample file or a small subset of the data just to get that proof of concept um in front of the stakeholders or in front of um the customers who need to validate it or you know these types of things. But actually going up against your production data systems and production table tables and all of the different tenant environments, tenant workspaces um is can present a whole different challenge. Um so scaling to performance scaling all of the to getting all the data that you want that's where kind of that real engineering effort is going to happen and where again often times the AI generated prototypes are going to sort of fall down uh in that process. So um performance at scale is going to be an incredibly important and um this one certainly goes very much handinhand with the previous one which is all about the security. Again um often times when you

build something from a prototype stand standpoint you're not necessarily thinking about um all of the tenants maybe just looking at a single tenant use case and then the data that goes with that and so thinking about how am I going to secure it and then how am I going to scale it. um is is kind of the the key challenges that you want to make sure you're paying attention to as you're looking to scale from that AI powered prototype into the production mode. So all of those challenges are certainly things that you want to make sure um you're looking at and paying attention to. Um and then probably the the third challenge um and this might be sort of um the least obvious uh challenge among um the ones we're identifying today and that is the long-term sort of maintenance uh customization and ownership of the solution. And we see this a lot when it comes to um the embedded analytics use cases and that is once you start putting these features in front of your customers um in inevitably they're going

to ask for more. So it's all about hey now that I see this report or this dashboard or these other features uh around my data how do I um add to that? How do I get that next chart that next dashboard? uh with tools like Qrvey uh we tend to solve that with capabilities like full selfservice where you can actually empower the customer um to create um their own content and create their own outcomes as part of that process. But again uh you want to make sure that it's not just about getting kind of the first iteration in front of the stakeholders in front of the customers. It's all about how do you then continuously maintain and grow the solution to match to sort of all of the things that the customers are going to continue to to ask for. Um and then you know one of the things that we've also seen kind of in this um embedded analytics space is this can become sort of like a full-time job right responding to the customer requests maintaining that solution um building out a full

featured you know roadmap set of uh capabilities for your customers so that you have a steady pipeline of new enhancements and new capabilities that are are coming down the way and you again um leveraging AI to potentially prototype these new features or leverage a combination of tools to prototype these teachers uh these features along with self-service capabilities baked into products. All of that can come together to create kind of a a full production system um that you have that um allows you to um present to your customers exactly what they need um and and maintain the system long term um so that you're they're getting what they need. uh overall in the in the long run.

Now I think um essentially those challenges then become sort of the three questions that we want to ask ourselves um before we ship any of these features maybe that we're we're starting with AI. So instead of just saying hey I built this in a couple of hours it looks great let's go ship it. um those three challenges. Can I handle the security? Can I handle the scaling? Can I handle the ongoing maintenance, support, and management of those features? Um instead of just saying, "Hey, let's get it out the door and and ship it." um let's uh assess whether or not we can uh handle all of these challenges um and essentially uh save us um a huge amount of headache in cleanup if we don't actually address these items up front.

So really the point of kind of all this framework is to understand um how we make this correct decision. So um really um the worst outcome would be ship something and it's not complete, it doesn't secure, it doesn't scale, and it's hard to maintain. Um and in fact, if we decide, hey, let's just throw this away, that's actually a success because the prototype validated our idea. either it killed it quickly because it wasn't the right uh feature set or outcome or um it was a a great outcome, customers loved it or stakeholders loved it and now we can actually put that into production. So I think um again all of this are are just questions that you want to ask yourselves as you're going forward with your um your products and your features and you want to be able to leverage uh AI to get there. Now, um I think sort of to recap everything here, so what do we want to take away from this?

Um you know, number one, I'm a huge fan and a huge proponent of you of of using AI to sort of aggressively validate and prototype the concepts. I think uh obviously as I said earlier, a lot of us have already seen these proof of concepts get generated maybe even in a day or an afternoon. Um, it's a great way to learn. It's a great way to validate ideas. It's a great way to um get stuff in front of stakeholders extremely quickly, but um those three costs are going to matter. The security, how does it scale, and how do we maintain? Um those are going to be really important and you want to make sure that you're addressing those questions before we actually do um get into production. And then you want to make sure that again you're you're handling all of that. so that the the demo or the PC or the prototype doesn't become this huge amount of sort of tech debt that you're having to deal with um in the long run. Um now the last thing I just want to um sort of leave you with

is at Qrvey A we've really architected our solution to help with this process. Um, we have a platform that allows you to create the production ready systems, but also incorporate AI models and AI tools, including MCP servers and AI agents that you can even customize to start creating those uh, proof of concepts, prototypes, and even just getting started uh, with delivery of um, you know, the capabilities, the customerf facing analytic capabilities that you want to provide um to your to your customers, to your end user, to your tenants within the solution. So, I'd love for you know to chat more with with all of you about this. Um you can certainly connect with me on LinkedIn.

I've actually been posting about this topic a lot and this is where we've we've actually heard a lot of this same feedback from uh colleagues and other SaaS leaders where they're saying, "Hey, yeah, we're we're doing a lot with AI to prototype, but we're also trying to take into account all of these challenges and considerations." So, um, shoot me a note, send me an email, visit us at curve.com. Uh, you can, uh, see a demo of the software. But yeah, I'd love to to chat with everybody and hopefully continue discussing this wonderful topic of of how we work with AI effectively and how we go from that PC or prototype into production. Uh, thank you so much uh, and uh, hope to see you on the next one.