For this episode, @GuyM and I invited @AaronK and @RogerSramkoski to join us to discuss why SnapLogic's GenAI App Builder is the key to success with AI projects. Aaron is the Senior Product Manager for all things AI at SnapLogic, and Roger is a Senior Technical Product Marketing Manager focused on AI. We kept things concrete, discussing real-world results that early adopters have already been able to deliver by using SnapLogic's integration capabilities to power their new AI-driven experiences.
Subscribe to the podcast to receive this and all future episodes directly in your podcast player, or find the Enterprise Alchemists wherever good podcasts are downloaded.
Chapters
- Integration and Application of Generative AI
- Automating Processes With Gen AI
- Evolution of AI Integration Strategies
Transcript
Dominic Wellington:
Welcome to the Enterprise Alchemists. My name is Dominic Wellington, I'm an Enterprise Architect at SnapLogic and I'm joined by my colleague, guy. Hey, guy, hello again. We are here today with a couple of our key people in the Gen AI, generative AI side of things, who have all sorts of interesting things that we would like to chat about for the next few minutes. So, aaron, roger, thank you for joining us. Why don't the two of you introduce yourselves in alphabetical order?
Aaron Kesler:
Yeah, thank you for having me. I'm Aaron Kessler. I'm a senior product manager here. I manage all of our AI products.
Roger Sramkoski:
My name is Roger Strzokowski, senior technical product marketing manager, so my role is to work closely with Aaron and the field team to help basically paint the picture of generative AI at SnapLogic.
Dominic Wellington:
Excellent, so thanks for coming once again. The topic of today's conversation is the GenAI app builder, and perhaps we should start with that name, because people might have heard a different name. And just what is it? What's it all about?
Roger Sramkoski:
It's an interesting question. I was actually at a conference last week and the last day was sort of like a speed dating, and so, to really quickly paint the picture, snaplogic essentially helps you break down the individual pieces of integration into snaps Very simple way to do it, and we've just taken what we've been really good at for a long time, and now we're applying that to generative integration or generative AI. So what that means is thinking about all the individual steps in terms of connecting to LLMs, connecting to vector databases, tying that all together with whatever your front end may be, whether that's Salesforce or Streamlit or something similar.
Dominic Wellington:
Okay, so how does that fit into people's existing architectures? I think that's the key question here. People are trying to do generative AI, often kind of in isolation, and I was also at a conference a couple of weeks ago. We do a lot of conferences around here and the Gartner analysts were sharing a terrifying statistic that 90% of generative AI projects fail and they directly linked it to the fact that people are trying to stand these things up in isolation and they're not really tied into the rest of their business systems.
Aaron Kesler:
Stand these things up in isolation and they're not really tied into the rest of their business systems. Yeah, I think the reason why a lot of these projects fail is because you know a lot of our customers. They've started by getting these sort of top-down directives from their CIOs or CTOs of we need Gen AI in everything, without actually quantifying the value or direction of why you need gen AI. And so what we've sort of tried to walk back is, you know, that sort of discovery process of uncovering what is actual, the human problems that you are trying to solve, and that can be automated by machines.
Roger Sramkoski:
My take on the failure rate, too, is that there's a lot of people just experimenting with this so far.
Dominic Wellington:
Yeah, a high failure rate is normal. With a new technology, early days, people don't know what it's good for, right.
Roger Sramkoski:
So I think, to a large degree like I believe this is normal we are seeing we're just starting to see, with our early adopters, the cases that get into production and they do take a little bit more focus. I think we're seeing the most success with teams that are dedicated to AI. In many ways, it's their focal area, so they're not trying to mix.
Guy Murphy:
We're working from SnapLogic, so we actually have a lot of people going to be listening to this podcast, hopefully with actually a traditional data background, application integration background, api development background, and also hopefully having some of the new AI style and the new people building these things. So could you quickly describe the intersection between these? So could you quickly sort of describe the intersection between these? Because it's probably not. If you were to look at our profile, historically we've been a market leader in modern enterprise integration.
Aaron Kesler:
What's the sort of value of combining other approaches that we've had with the platform and what you're seeing with this new style of solution. So what I think is a beautiful, almost case study of one of our customers is you know, with SnapLogic, you actually GenAI App Builder is an integration problem is fundamentally what we're trying to solve, and so one of our customers has a fraud detection algorithm that they actually have built within SnapLogic in a series of pipelines, and what they actually want to do is just incorporate GenAI Builder in the next step of that process of now that I have a detected fraud, a fraudulent activity on this account. I have a human who reviews all of that and goes in and looks for summaries and looks for the past transactions on the account and then creates the exact summary of that. What they want to use Gen AI Builder to do is actually just create that last piece of that pipeline of generating that summary of all the account data, and so that's why I think it's such a beautiful integration problem that we're solving 100% agree with that.
Roger Sramkoski:
It's definitely an integration problem, because another customer study that we're sorry. Another customer use case that we're looking at is a sales team that exists in Salesforce like so many others, and their sales team is now able to leverage SnapLogic integrations behind the scene for Gen AI without ever leaving Salesforce, and so I think that empowers their team significantly to use the tools they already have. But it's still an integration problem in the sense that we already have connection to Salesforce and so now we're just augmenting that.
Dominic Wellington:
Yeah, and I love that these use cases are not ones that start with AI. They start with I have a business question, I have a data question and then someone thinks oh, yeah generative AI could really help me with this business problem, as I have Exactly right, which is also some of the work that we ourselves at SnapLogic are doing. I don't like dogfooding. Let's talk about drinking our own champagne. You guys are closer to that project as well. Do you want to perhaps explain a little bit for the listeners?
Guy Murphy:
And just to translate for possibly the less European-orientated, my Italian colleague is very cool In Europe we prefer the French version, the whole "ou must drink your own champagne using it yourself right.
Roger Sramkoski:
So how are we using SnapL ogic GenAI App Builder to help SnapL ogic? Without getting into too much detail, I'm giving enough that we can understand it. So, like many organizations and when I was at the conference last week, this came up with a lot of people who came by our booth One of the challenges, whether you're in finance or recruiting or an IT department is at some point you have some document that you need to summarize or extract data from, or both, and anybody who's tried to parse a PDF is going to run into the situation where it's text, it's an image or it's some combination of both, or it's got table data, and that's where the pain point is, and that's what I heard a lot last week was exactly that. So what we did was the beauty of SnapLogic that I've always loved is you can run things in parallel and we can be multi-point to multi-point, and so what we did for this was we actually run and parse it out three different ways, and then we use an LLM to essentially grade and then we choose the best route and if that was below a certain score or below a certain grade, you alert the human in the loop, and so we're taking this process and others have a lot of interest in this as well and just being able to automate something that is consumed.
Roger Sramkoski:
I believe our estimate was something like 40 hours per month per person. Looking at this internally, just to share the stat, that's a lot of hours, so that's a lot of hours, and so we just went into production with our own. It was last week, yep, so yeah, so this is us. You know it's putting it into action, right, yeah, and again because it's a real business program.
Dominic Wellington:
Our salespeople, they're busy people, they have other things that we would much rather they be doing with their time and not messing around with PDFs.
Guy Murphy:
So I've got to say and obviously everyone in this room knows me, my background's both in data and heavy process I think this is a really interesting point maybe to touch upon, which is and maybe this is linked to this perception of failure, where the classic testing of an interesting technology in isolation, everything we've actually been talking about is actually optimization of process, yeah, that the technology facilitates rather than and I know this is going to be very untrainable for people who love hype the idea that gen ai is the thing. Actually the thing itself is an accelerate capability of better process, or better process to human as an extension. Yeah, and yet I think this is maybe where some of the rumors of Gartner's truck of disillusionment is not actually the technology solution. It's just actually a prep for the accelerating, the framing to actually this is a whole new dimension to processing yeah, I mean that's a fair thing from what you've been seeing, both internally and with our early customers.
Aaron Kesler:
Totally I mean, there's a brilliant book called automate the boring stuff with Python, right, and that's exactly what we're doing, right, we're automating the boring stuff so that the humans can be freed up to do more important, more complex stuff that only a human can do, and then they become the approver or the auditor of what the Gen AI produces.
Dominic Wellington:
Excellent. Yeah, I like the analogy. I think it was Benedict Evans came up with the analogy of these models as being like having infinite interns the intern. They're brand new to the organization, kind of by definition. They don't have a huge amount of background. But if you have a lot of them, you can throw them at the simpler sorts of filing and sorting problems and they'll get it done. And then someone at the far end has to take a look at the outputs and see was it any good? So how are we working to get this into the hands of customers? Are there any early learnings of approaches that work well, that are successful? Are there any particular requirements that customers have to engage in this sort of project?
Aaron Kesler:
So what we've found from this is that the sales process has become a lot more consultative, where we're actually doing a lot more discovery up front on figuring out what these use cases are and then, once we actually closed the deal, we for the first you know, I suppose six months since our release of Gen AI App Builder we incorporated both a PS package as well as a white glove package for these customers, and that means, you know, actively sitting with our customers, going through the motions with them of uncovering the most valuable use cases and then, as quickly as possible, trying to get them into production so that they become a reference for us and also deliver a massive amount of value to their business that fraud detection use case that I mentioned.
Aaron Kesler:
They estimated that it would be saving about 25 hours per person per month on doing that, and so those are the types of use cases that we uncover through this discovery process and then through the white glove approach, we build out that pipeline with them and then teach them those practices so that they can build out the next X use cases that they uncover after they're done.
Roger Sramkoski:
I think one other thing I would add to that too, from the sort of the marketing partner side was we also created a two hour workshop that we made available in, I think, a total of 18 cities over several months, and this was open to customers, prospects, anybody really who wanted to come in, and we took that two-hour time to help teach them. So it's also, you know, we're trying to put our best foot forward on educating as much as possible, because it's new, it's moving fast and it can be intimidating.
Dominic Wellington:
These workshops were run with some of our key partners AWS and Google, and people at that level Correct. This is not something that we're doing on our own, which is another important point. It's Gen AI app builder, but it's SnapLogic, helping people work with data that they already have in their organizations, with large language models that either they're the standard ones from OpenAI or whatever, or they're offline ones that people are running themselves. This is, you know, us facilitating. You know what SnapLogic does.
Dominic Wellington:
We help snap all of this stuff together correct so I'm sure you're why we've already had a session with Greg Benson yes, all right, and if you, if you haven't listened to episode, go back in the feed and you'll find it. Greg is our chief scientist. He's a great lesson.
Guy Murphy:
But I thought, especially when you're very close to the sales and also the trends from our business point of view, where do you see what's going on? So the trends beyond? Obviously you're working with Greg and that, but what do you see when you're out there talking, working with clients, watching the market, listening to partners, the next wave of this and the opportunity and possibly the challenges?
Aaron Kesler:
Yeah, I mean, one of the great things about the White Glove program is being so closely tied to these customers. When we first launched Gen AI App Builder, our roadmap was essentially just a laundry list of things we wish we had known when we developed SnapGPT, and so all of our roadmap was sort of what we thought was cool of our roadmap was sort of what we thought was cool, and, after working with these customers, about 80% of our roadmap turned into stuff that was directly coming from them and so but that leaves that other 20% to what do we think is next up on the horizon and Greg and I were just talking about this earlier this concept of agents, where you're actually using the LLMs to essentially make decisions on which LLM to call next in the pipeline, and so you can envision having this series of LLMs snap together that are calling different databases or different data sources, depending on what the prompt is and what gets retrieved from the answer of the LLM.
Roger Sramkoski:
You mentioned something that, if we don't cover it in some other episodes, I'll just take a minute to plug it right. But, like, a lot of our learning came from our initial attempt to make a copilot inside of SnapLogic yes, came from our initial attempt to make a co-pilot inside of SnapLogic. Yes, and you know, from the PM side, you saw that part of it. And then from the marketing side, when I was meeting with initial customers, one of the things that they immediately said when they asked us they saw it was this is so cool, can you help me build it? Yes, and we had a lot of learning in building our own co-pilot and I think what you're getting at right is we. We wanted to share that.
Roger Sramkoski:
Essentially, like, how do we take what we've learned and allow our customers to go build their own? Exactly right, and then guy back to your question around, like some of the challenges and trends. One of the challenges that we've heard from a few folks already is the AI request backlog is getting thick fast and that's a fair challenge for them. So even customers who have coded this well before we were delivering the easier integration piece, they're already swamped too, and when we show this to them, what they see is an opportunity to move faster and I think that's a massive trend that is going to come out is how do I do this and move faster? And that's a lot of filtering it out, but it's a lot of accepting a tool that might allow you to do that, instead of trying to do it all with code.
Guy Murphy:
Yeah, and it's an interesting trend. And again, I've been around the industry for a long time, so I've gone through loop after loop after loop of emerging technologies and yet some challenges keep on recurring. So you said everyone codes first. Come back in 200 years time I guarantee they'll be coding first in prototypes it might be written by AIs, but they're still coding, but being a bit more serious, and that's perfectly acceptable. But what's very interesting in this domain from my point of view is the speed of the need to mature this.
Guy Murphy:
So if you think about it, this didn't actually really exist two years, two and a half years ago, even as a concept outside of academia and Greg's domain. And now it's pertinent with the chaining model, because if you do one, yeah, it seems acceptable to script around it. That's what you do. We saw that happen massively in the big data period when that was hot. But the fact is, yes, as soon as you're saying I'm chaining systems and I'm moving data flows between them and I'm going to be feeding these to enterprise systems because the business context is and they do something, recommend something, trigger a process, I hate to say the old dealities are going to kick in.
Guy Murphy:
Can I see what happens if it doesn't work? How do I scale these types of concepts? How do I also then plug it into my sorry people who are very excited about this stuff? How do I plug this into my ops management? How do I do the end to end management when it leaves the labs and it becomes the core and it's? But for me it's fascinating. For me, this was these sort of cycles were actually five years long. I think this is actually down to nearly 18 months or two years at most. We're seeing that evolution, so we're coming to an end. But it'd be really nice for you obviously with discretion from the customers that we're actually working with today, who actually are deep into this journey, maybe ahead of a lot of customers that we have and a lot of the listeners and the enterprises they have Can you articulate their maturity of how they've gone through this curve and actually why, if they've been doing it, they've ended up working with ourselves?
Aaron Kesler:
Yeah. So one of our customers I think they articulated it very well in that they kept seeing that their backlog of hard-coding all of these Gen AI applications that they kept seeing that their backlog of hard coding all these Gen AI applications that they were building with Python and SageMaker and all these different tools their backlog was growing way faster than they could even accomplish any of the use cases, and so one of the reasons why they're looking at and buying Gen AI App Builder is because they saw that it would allow them to go much faster and tackle more of these use cases with fewer folks and less data science expertise than if they were just hard coding everything.
Dominic Wellington:
I like that. The tech is moving so fast that the determining factor is not the tech. There's no point planning it out. For those of you who are already current SnapLogic users, maybe you were at Integrate last year Roger was presenting in San Francisco. I presented at the London running of SnapGPT and that product has just evolved by leaps and bounds. It's completely different from what we showed on those two stages, and Gen AI App Builder is obviously following that same trajectory. So it's more about the needs that people already had, and the technology catches up with those needs and suddenly bam, you have a new thing and we make it easy to get to that. Okay so, thank you.
Guy Murphy:
Great insight and I'm sure we'll be getting you back on the show, especially as we engage with more and more customers and learn more new things to share with the community Anytime.
Roger Sramkoski:
Absolutely. Look forward to it. Thank you, gentlemen, love it.
Dominic Wellington:
Thank you! For those of you listening. Please do check out the page on Integration Nation on our community site, where you will find all of the show notes for this episode and additional resources that you can go and check out if you want to find out more about the topics that we discussed today and, of course, as ever, if you want an individual one-on-one consultation, you can reach out. Thank you all. Talk to you next time.
