Mike Vetter:

Alright. We're gonna get started here, this afternoon. Thank you for joining our webinar on how to integrate AI into your software startup product strategy. I'll be your event host. My name is Mike Vetter.

I'll be introducing today's topic and speakers. I'm a cofounder of Wildfire Labs, who is putting on this webinar. We are an accelerator that helps founders take an idea, shape that idea into a software product, and then a software product to revenue. We specialize in helping companies build AI into their product strategy and helping founders make that transition. So today, we'll be covering a practical framework for aligning your product with AI, And we'll also be talking with, a real world example and Chris, who is leading a company that is, going through this transition right now.

So, before we get into today's topic and presenters, I just wanted a couple, cover a couple of webinar items. So first, you'll be on mute for the duration of the presentation, but we will be opening up your mics to ask questions at the end of the presentation. However, if you do have questions, there's

Todd Gagne:

a little q and a button

Mike Vetter:

at the top of today's webinar, and you can ask a question. I'll be monitoring questions. And then at the end of our webinar, we will be answering those questions, with our presenters. So, the another thing, just housekeeping item is if you, are unable to make the entire presentation today, don't worry. It will be recorded and you can find, this, event in our Startup Tales from the Treehouse podcast.

And you can subscribe to our podcast on, Spotify and Apple Podcasts. So without any further to do, I'm going to turn it over to, my event co host today, Todd Gagne. So, Todd, thank you for leading us today. I'll let you take it away from here.

Todd Gagne:

Thanks, Mike. I really appreciate it. And like to welcome Chris Grevis, to the podcast. Chris is the CIO of a company called We Localize. Chris and I have known each other a long time.

Chris Grebisz:

I don't know. 27, 28 years. 8 or something around there.

Todd Gagne:

Somewhere in there. Yeah. So it's it's been a bit. So, and honestly, Chris, I guess, a welcome, and I appreciate you taking the time. I guess, one of the things, I guess I would say that kind of precipitated this was you and I have had a number of conversations over the last 2 years in particular around AI and just basically how it's, you know, shaping not only your business, about what we're doing as well.

And so I've always really valued your opinion and how you think about things. And so I think hopefully this is a good dialogue on some of the journey that you and I have been on together, as well as kind of what you've been doing with We Localize. So welcome to the podcast. I appreciate.Chris Grebisz:

Glad to be here. Thanks.

Todd Gagne:

So why don't we start with, your background? You know, you've done, you've been in kind of the localization space for most of your career. So maybe why don't you talk a little bit about just, your journey through that, either, you know, GSSI to Welocalize and just the scale that you've ended up having to go from something that was a pretty small startup that you started yourself to kind of where you're at today.

Chris Grebisz:

Sure. Yeah. I think it was 97 or 98, so dating myself that started GSSI. And at that time, there was, a and I I've been in localization for a year prior to that, so I was still quite young. And, and but I saw an opportunity starting to develop around the Internet and ecommerce in particular.

And so I founded a company that really focused on globalizing, ecommerce environments back in the day with no the b to b environments, metal sites, you know, companies that are starting to transact. And and that was really great. I mean, we we grew like a weed until we didn't. And and it kind of imploded in the early 2000. We scrambled around.

We ended up merging with Welocalize, which was a primarily a not a tech services global tech services translation services company. And we put 2 small struggling companies together to make a small company that struggled a little bit less. But today, it's, 1500 employees, over 300,000,000 in top line revenue, and it has seen very good periods of growth riding sort of the the wave of, global digital content. And and we are we are a language services company, but we process, I think, 4,000,000,000 words a year, and so it's heavily, heavily tech enabled. And, and that's it's been interesting.

Todd Gagne:

Yeah. And maybe the backdrop to that, Chris, is a little bit just I mean, you and I are about the same age, but we've kinda gone through a series of just different technology epochs. Right? I mean, it's like when we were younger, you know, I mean, I think we both can remember when we first got our hands on computers and kind of what we were doing. And then as you're talking about professionally, the Internet came along and really changed our perspective and the opportunities that we had.

And then, again, mobile. Right? Mobile changed it. And now I feel like we're not to another epoch where it's like, AI is really kinda starting to stress a lot of the business models and yours in particular too, on how we're changing and thinking about it. And so to me, that's pretty interesting for the time that we've been alive.

That's a lot of technology change. And, you know, digesting each one of those has met pretty seismic changes in business.

Chris Grebisz:

Absolutely. I mean, it's like I I definitely, you know, have thought about that a lot because I remember getting the first email in the early nineties. I remember seeing the first h t z page. And I like, at the time, you're like, wow. This is super cool.

Don't know what it's going to be, but it's gonna you know, it's going to change things. And and that's exactly, you know, November of of, you know, 18 months ago seeing the 1st chat CPT and starting to engage in it. And that November, I had that same sort of revelation as when I was 20 years old. It was like, this is and in our business, you'd it it was in the 1st 12 months when ChatCPT came out and, you know, I was doing a certain amount of, clanging pots to get people's attention on this because it was looked at a little bit like as a circus act. But it's a large language model, and we're in the language services industry.

So there's gonna be an intersection at some point. And the and the language and it wasn't entirely new to language services. The language services has been using, machine translation, which is a AI technology, neural machine translation for a long time. This is different though. And it's different because it's accessible.

It's because CEOs are using it on a daily basis. They're talking to, you know, their their Spanish relatives. And and and now it's like communication is is not protected. It's not in a dark room somewhere. It's like I can communicate with anybody in the world in a reasonably good way.

That's gonna affect our business and is affecting our business.

Todd Gagne:

Well, maybe before we get to the business side of it, I'm curious. I mean, even you and I just talking about new revelation, new applications to just chat. I think, you know, early on, you're playing with it. You're trying to find all these different avenues. And you had some kind of creative ones that maybe are kind of unique.

So maybe walk through 2 or 3 of your, areas that you found some value in this that maybe is more personal to you than the business side of it?

Chris Grebisz:

So, like, when 4 o came out, it's a really good language, taking language lessons. So I do a morning walk pretty religiously, and I'll do a few Spanish lessons. And I'll ask, Juniper. And I'll say, you know, I wanna learn some new verbs today, and we'll go through. And and just as I'm walking along our neighborhood, which is, like, pretty cool.

Right? It's like That is pretty cool. And, you know, as soon as when I started playing with that also, I owned a little bit of Duolingo stock and I sold it really fast. It's like the other thing actually, this is a little geeky, but, like, food recipes. Yeah.

Not like you know, and I want, you know, healthy recipes that I could do, I can do fairly quick. I actually made a little GPT for this. And yeah. And so that I could identify, you know, inside of certain criteria, certain recipes, and and and and menus for the week. Those are interesting in business, you know, OCR and receipts, and you could save it out in a in a in a CSV that, that concludes all the data that I can import into Concur.

Yeah. You know, so there's it it's like the interesting change management in societal is like like I've I've embraced it. So I think if I anything I'm doing, any type of task I'm doing, I think, okay. How can I use Chargebee to help you? That hasn't set in.

You know? And it's even like you have bantering conversations and and with amongst, you know, people and and groups and over drinks. And it's always like, let me Google that for you. Even search sometimes doesn't set in with certain people, and this is an even a bigger step. And so I any anytime I'm looking at a laborious task or just wanna have fun, I'll go to chat kpt and start screwing around with stuff.

Todd Gagne:

I one that I that we that's kinda happened just recently was, we're planning on going to Germany for vacation. And, one of our partners, the one you was introduced to, he's him and his wife are originally from Germany. And so they gave us a whole bunch of, things to do in the region. And so we were like, hey. This is great.

I have no idea exactly where they are in in kind of context. And, you know, is this a museum? Is this a town? Is there something else? And so I just threw it all in and said, here's what I'm trying to do from a timing standpoint.

Here's the types of things we like. Build me an itinerary. And it it the first draft of it was very good. It was better than it would have done. But I told it where I was gonna be staying, where I was gonna be ending, how long I wanted to drive.

I'd like to hike, all these types of things. And it was able to kind of grok that information and put it into an itinerary, you know, here's morning, noon, and evening, what you should be looking at doing. And I was like, that's pretty cool.

Chris Grebisz:

It it's like now it's getting to a point of maturity where I can act I didn't actually fully believe in digital assistants. You know, because, like, Siri and Alexa, they just weren't that good, and they were very, very awkward. But now with the the more recent evolutions, I can actually see how I would really want I could I could utilize additional assistant because it's just so natural.

Todd Gagne:

Yep. It is. So let's pivot a little bit to kind of relocalize and go back to the 18 months ago where basically you saw this. You you started to talk to some of your customers, and you started to maybe the maybe the sphincter got a little tighter maybe. I don't know.

Chris Grebisz:

Yeah. It it, you know, it was one of those well, the other thing that happened, 18 months ago is is I I I reread Clay Christensen because, you know, incumbents and and, and and and it it you know, we were an incumbent inside of an industry, and it's, like, needed to reread in terms of what the the dangers were in front of us, which they still are. It's it's we are language services company, and these models, even though there's NMT out there and there has been machine translation, it was more of, the accessibility of this technology is going to accelerate the application and and utilization of MT. And in it's particularly companies that are building global content. And what I think is going to happen and is happening and this has led to some, other business decisions is that a marketer can produce their own content in comfortably in maybe 10 or 15 different languages or, a product manager can produce UI.

They don't they don't rely upon necessarily, you know, a whole big bureaucracy to do that that exists in bigger companies. So we really looked at that in the strategy with the business and where we take the business. We embraced it. We absolutely said to ourselves from our investors down, we have to embrace this. We we we have to be a disrupting incumbent.

And, and so we also then took hard looks at where can we use this internally as well, you know, trying to check of our staff, the more that our team around the world embraces this, that's gonna that's gonna drive innovation inside of the company, both for services and and solutions that we're delivering as well as for internal applications. And, you know, we we had corporate we had meetings. We set up, you know, some idea labs of trying to generate ideas from different people and and and different use cases. And in the last 18 months, I mean, I think we've done a good job at embracing the technology. We are altering our services.

We are building our own models for doing quality evaluations and doing post editing and and content creation. We're using it internally robustly. You know, great example couple of great examples. Our, dev organization is using LM for 90% of their code reviews. And and that's that's awesome.

Like, I mean, developers hate doing code reviews and

Todd Gagne:

Code reviews. Yeah.

Chris Grebisz:

What Satya said in the very early days about, the release of Copilot is taking the drudgery out of work. And I I actually really believe that's that's important. It's help desk and our IT organization. 35% of our l one tickets are being resolved through an agent. And, again, you know, that's that's that's and that allows the rest of the, text team to be able to focus on problems and put more time into the the hard things.

We're using it. Product organization is using it to build epics. It actually does a really good job at epics and stories, because you get some standardizations. Like, it's really easy to forget about logging events when you're building stories, and it it'll and you Reminds you of that? Yeah.

It'll remind you of that so you get it in there. So it's it's, we're using it inside of our production environment where we're using AI to do resource selection. We have a database that we send work to that includes over 200,000 resources. And instead of ATM having to do that, so it's it's pervasive throughout our company. And then, you know, even also one of the theories we had is, like, things are gonna get more and more narrow.

The text allows us to gets become more and more specialized. And so we bought a Martech company that we can integrate into other Martech stacks like AEM and building you know, it it basically charging up the problem and saying, okay, marketing operations manager. We're gonna put the ability for you to create global campaigns on your desktop without you ever having to leave Marketo or ever having to leave Eloqua. And then we wire in our services to for mostly quality evaluation on the backside so that they can do that in a confident in a comfortable and confident way. By doing that, they can probably produce more personalized campaigns.

They could possibly enter more markets, and they can do it faster. And so and faster and more leads to potentially better outcomes for them.

Todd Gagne:

So maybe you talked to me a little bit about, you know, it sounds like you got good adoption internally. I mean, it it's you created a environment of innovation. You tried to take away some of the fear of new technology and change. Talk to me about the other side of it. Like, do you have, like, a traditional, customer early adoption curve, and then basically people asking a lot of questions, and then people starting to create initiatives?

You know, you hear a lot of this where I think there's a lot of tinkering going on in the enterprise, but I'm not sure that, like, wholesale changes are there yet. And so I'm curious on, you're building all these solutions, you've got your, you know, your team aligned. I'm sure there's some early adopters, but what's the rest of the market kinda telling you?

Chris Grebisz:

A good portion of our business is servicing tech companies, and so that's gonna be biased towards early adoption.

Todd Gagne:

Early adopters anyway. Yeah.

Chris Grebisz:

Yeah. And so our business we have, you know, tech organ tech companies that we service, we have a legal services company. I'm seeing less adoption there, but there is awareness that's that's developing. And then we have a a part of our business that services life sciences, med device and things and things like that. And that's heavily under regulation that is going to probably slow the adoption of the tech inside of tech services and and manufacturing type customers.

Absolutely. I mean, they are looking at implementing, you know, going, you know, moving away from a a a, using MT in a in in very selective areas to moving towards a world where we wanna empty all of our content. But we need to have the tooling in place to give us assurance that it's not gonna you know, if you're using LMs or if you're using MTs that, we aren't creating egregious content. And so like a lot of our development has been more on the risk assurance side. You know, we there's a great MT development tech out there, and the LMs are the LMs.

We are we're using both models and technologies. But we're we're focusing in our the problem we're solving is, okay, companies, you're gonna embrace this. We need to give you risk assurance that you can expand and, your global footprint, your global content without creating any additional risk. So in our experience, you know, like I said, we ran to it. I think our customers are running to it, and it's pervasive.

And, so far, I think, you know, I I do think, you know, I think early days we were definitely in front of it toward compared to our competitors and our peers. I think there's it's a little bit more level now. And, you know, and and I see this in all content services out there that anybody who's in a content service environment has to be embracing this and developing business strategy around

Todd Gagne:

1. So what about the people? I mean, you know, I guess what all the examples you gave me, whether it's like code reviews or help desk tickets or, you know, other functionality and initiatives you're doing, that requires, definitely a Copilot. It's not a standalone application at all. And then it really is starting to take away some of the drudgery that you talked about from the employee, but there's a kind of a level up kind of a scenario to go do.

And so is there is there any concern with the employees or, like, the way that you're thinking about, like, saying, okay. The bottom 10% is what you're doing. I need another 10% that's above what you're doing, and how do you kinda train to adapt to that?

Chris Grebisz:

Yeah. It's delicate. Right? Yeah. I imagine it's

Todd Gagne:

hard change management.

Chris Grebisz:

It's, you know, I think in, you know, even with my own kids who are newly into the workforce is understanding in terms of the topical areas that they're working in, you know, understanding how to use this technology, in innovative ways to to create differentiation, to create productivity. And I don't think full functions are going to be removed, but I think that, you know, a developer the developer who is resisting using it for code review, well, they're 10% less efficient. You know? They could be using this. And and so, and that's change management.

I can't say that, you know, there's we're a 100% change management, but it is just constantly, working with teams and functions out there of how to leverage the technology. And and then I think you have to to I think there's it has to be positioned as from that drudgery standpoint. We want you to focus your time on things that you're skilled for, solving whatever problem it is, whether it's developing, whether it's accounting, whether it's project management and customer engagement, customer solutions, and get away. Let the tools get rid of all the admin work.

Todd Gagne:

Do you find that personally for you? Like, I mean, do you find, like, you can get rid of 10% of the drudgery work of the day to day job for you, or do you think, like, it's not the tech's not there to actually do most of that yet? Or that does that go back to your virtual assistant concept that's coming?

Chris Grebisz:

So this is terrible. Like, if I'm I mean, just from a research standpoint, or let's say there's a long email thread, I might cut and paste the whole thing into our internal version of GPP. You can give me a summary. I'm really lazy on that stuff. If there's articles and papers, I use it a lot for getting summarizations.

If if I'm, you know, having to write content, I will often start with it, to to give me some starter stuff. And so I don't know if it's 10%, but I do use it for my own productivity. It's I've really tried to use the different toolings for PowerPoints, and those don't seem to be very good yet. But that would actually I would I would kill it if there was a good PowerPoint tool.

Todd Gagne:

Yeah. No. Yeah. No question from that standpoint. Yeah.

Oh, good. So maybe, like, the the the next piece of this we can kind of pivot to is maybe some of the framework that we have been talking about. I think, one of the needs, I guess, we've heard from our kind of entrepreneurs that we've been working with is that there's so much noise about, technology and how big these models are and what they can and can't do. But But I think what what's lacking a little bit is maybe just more of a functional product strategy approach to entrepreneurs and business leaders trying to figure out, like, how to apply this. And I I I feel like it's overwhelming, especially if you're not a technical founder.

You're kinda sifting through a lot of this. It's changing quickly. And so maybe I can just kinda walk through kind of, each one of these that we kinda think about and see what your comments are. Because I think, this is going to change over time, and we're kind of in what we're calling right now kind of the AI co copilot model. Right?

Where it is kind of what Microsoft is doing a little bit. And it's it's basically 80% human, and then you're basically leveraging the best of AI. And, you know, I think, like, Word's a great example of it. You know, some of the the applications there where it's it's looking and trying to make you better. I think the, development and code review is a great example.

They can't replace anybody, but it seems like that's, where most of the high end use cases or high value use cases are today. Yep. Any feedback or comments on, like, that piece of and and, you know, basically, from your perspective, you think most of the application for you is in that that bucket first bucket?

Chris Grebisz:

Right now, I think it is. But we're I think we're moving fairly fast to, you know, a a, a pivot to, you know, being, you know, having to be able to deliver certain services that are gonna be more AI based, and and it's the human that's in doing the eval on it. In in terms of the tech itself also as we're staring down, you know, our own tech strategies, it's one thing I have realized and I've been working with our own dev org is we can't be too tied to any one of these technologies out there. Like, we have a problem that we're trying to solve. Let's develop it, design something to solve that problem.

But we have to exist in a reality that this tech will turn will change could change significantly in in 12 months or 6 months. Yep. And, and it's like it's like the AI is kind of like the little utility company that we need to plug into to get to get to get fuel. But ours is that we still have to think about, you know, what's the problem we're solving, and we don't wanna be too too, roped in too tightly with any one of the AIs out there. And so in our own modeling, we build model gardens, and we're using open AI.

We're using anthropic. We're using, different different models. And so we're there's a lot of dirt diversity there.

Todd Gagne:

So what does that mean practically, though? Is that just making sure that you have, an API structure? You're not doing anything that's basically tied to one of these, you know, LLMs. And so, you know, you're you're basically having to design for, what the lowest common denominator from an integration and feature set is?

Chris Grebisz:

Ideally. Ideally. And also to keep, things very, very simple. And so instead instead of build building really big stacks, you know, think very much in terms of, more simplified components and that then we can rearrange and reorganize as as the technology is evolving.

Todd Gagne:

Okay. So maybe the next component of this that we are kinda thinking about is is more of the stand alone, AI, you know, but basically that's single threaded. It could go do something functionally for you, and then return something. And so, again, it's probably 80%, the AI in that case, and 20% you. And so do you have any use cases, or do you see that coming relatively soon?

Or do you feel like that's gonna take some time?

Chris Grebisz:

Well, I mean, in our in our very, very specific use case like language and language translation and anything content wise, I think that that that path is happening. And in some cases today, we are seeing where, you know, we will produce content, different types of content, and 80% of that content is produced through AI. And then we have an evaluation that is somewhere between 10 20% of the app the com the combined effort. So I think as, you know, companies are building, models to perform specific tasks and, you know, there's there's still also, you know, how do you best train the models? Is either are you going to pick something up like llama and fine tune it?

Are you going to apply RAG strategies, which seems to be getting more and more popular? And and I think now I've seen more recently where companies are thinking, you know, how can I build my own models and and have some sort of efficient training mechanism? And now I can apply it to very specific tasks inside my business or very specific service for for someone. And, and then so I think that the whole aspect of and and training is like, we struggle with this as well. Like, there's a training methodology and that seems to be getting, better and and and more sophisticated or more mature.

But you're still sort of also dependent upon the quality of your data that you're gonna use to trade, and and and that's no joke. And so, like, I think they're getting to a world where you need less and less data, but, but you still have to have high quality data and that has to that data pipe has to be something that's is is is consistent and efficient in order to get most out of those models you're using, whether that's for an internal application or a product.

Todd Gagne:

So maybe 2 questions on that, like and maybe this is the wrong analogy, but it it feels like, you know, if you look at a large LLM like chat gbt, it's a Swiss army knife. It's bet to do a lot of different things. It's not optimized for any. And then I think what you're talking about is as you have more and more specific use cases, can you either build your own or optimize your own specifically for that? And it's not gonna do all the Swiss army knife type things that you need to go do, but it'll do that specifically.

Okay. So that's one. Seems like we're on board on board, same page on that one. The second one is you talked about, kind of an 80 20 type of philosophy with, today having these kind of, bots basically do something that's very specific. What is the, you know, like, we talk we hear a fair amount about hallucination.

I think we've all seen examples of that. How real in in, like, a business application standpoint is it? Is that really what you're really looking for in this 10 to 20%, overview to make sure that you're that that's not sneaking through and and really contaminating the final product?

Chris Grebisz:

Yeah. And it is very real. You know, we have developed, some customer service bots, for for specific engagement models inside of our portal, and I resisted actually releasing those to production because the hallucination scares me. And and I think with you know, over time, I think elucidation will be offset by by better training and tuning, or at least reduce the risk. And you can also use things like our own quality evaluation tooling, which is using an AI model.

We you know, and we could apply a couple of different ones that's gonna diminish and decrease the risk to a point that I think if you that that cue quality evaluation model of, output content, we're training that inside of, you know, the basic topical area of the input content, you get pretty close. You get pretty comfortable. We aren't we haven't we aren't removing the human. We still want a human evaluation because this is commercial content that's getting exposed. But I think over time, it's gonna get better and better and and derisk the the, the hallucination because that those don't seem like they're gonna go away by the probabilistic nature of how the technology works.

Todd Gagne:

So there's no relationship to the amount of hallucination or the variety of hallucination if it's a full blown language model versus a stand alone specific model that's very task specific?

Chris Grebisz:

If you can specific model, you have less risk because it's it's designated to perform. Yeah. Yep. I don't know the science behind it, but, I would have to ask our own ML team. But that's I've that would be task specific because you are tuning it and training it to perform a specific outcome.

Todd Gagne:

Yeah.

Chris Grebisz:

And and same with, like, we're doing multilingual models. We tune it and train it for marketing content for wildfire labs, and we'll have all of wildfires corpus in there. It's gonna produce a better outcome.

Todd Gagne:

Right. Yeah. That makes sense. Okay. Well, the 3rd tier that we're looking at that I think lots of people are interested and excited about is what if there's more of these independent AI agents that are running around and they start to mesh.

Right? And so they're starting to communicate on their behalf. And so, I mean, one of the real world, you know, examples that I have is just travel, business travel in particular. Right? It's like I had a flight a little while ago that basically, I was on the East Coast coming back, and I the flight was canceled.

And so, basically, I missed, like, I missed a lot of things. Right? So, basically, my hotel needed to be rescheduled. I had an Uber scheduled. I had, dinner with a with a partner scheduled.

I had a whole bunch of things, that I had to basically reschedule. And so if you could basically start with the genesis of that that said, alright, I saw your Delta flight got canceled. And then I look at all the dependencies that you had beyond that, and I can change all those things. And you can give me a little bit of prompts to say, hey, call that partner and see if dinner is available for tomorrow night. Here's the new flight information.

But, like, the integration, the trust, the access to my data to do Marriott app, the, you know, OpenTable app, the Delta app, the Uber app. Like, it's gotta be all connected. And so, I don't know if you feel like over time, we're gonna get to more of these things where you've got independent agents that are doing things, but then have the ability to talk to each other and share information.

Chris Grebisz:

I think it's frightening for a lot of people. Right? Because because they also lose control, you have some third party tech developer out there that it's not just the app itself or maybe you have a localized version, that you can control. But I I would love that world. I was thinking, you know Yeah.

Todd Gagne:

How many times this has happened to us?

Chris Grebisz:

Yeah. Give y'all I'll give access to to, you know, those different, services as well as my calendar and my email and, you know, and tell me when to go ahead and schedule the servicing on my car. Remind me to go do a dental dentist appointment. You can see my calendar what to put in there. I mean, the travel example is great.

I would I I would embrace that, but, I don't have a whole lot of, exciting stuff to hide either. So it's pretty good that I'm okay with it.

Todd Gagne:

Yeah. Yeah. I mean, it's how far do you think that is away? Like, how far do you think, like, not super complex, like, what we just talked about for, you know, maybe the travel scenario. But, like, your dentist one should be relatively easy.

Right? It's a standalone bot. You know, it's too you know, you know that, like, every 6 months you need your teeth cleaned. I've got access to your calendar. If there's if the dentist office and the scheduling app that they're using has an API, you'd think that that's relatively easy.

Chris Grebisz:

Hey. I think there's a couple I've seen a couple of startups, like, going down this road and, I would bet there's something sort of useful, you know, in the next year, because it's there. Right? It's like and particularly with Poro and and the and the multimodality of it, I can talk to it in a very natural way. And so I don't have any barrier to entry.

Yep. It's just, and so it's it's it's all plausible. It'll just be a matter of, you know, the API connections and access to your data.

Todd Gagne:

So let's maybe talk about some real world examples. I mean, we kinda talked about this framework of going from something that's super simple Copilot to something that could be way more complex with a, like an, like an agent AI agent mesh. You know, if you kind of apply some of the companies that are that we know, the big tech companies, Microsoft is an example, where it seems like they're all in on this Copilot model. Right? They they were pretty early.

They partnered with OpenAI, you know, and now they're trying to integrate it basically with Office, with, you know, their developer tools, enterprise applications. And so it seems like they're seeing that as an opportunity. You're a user of of some of their services, and you were just talking about even before we got on that basically the version control and the and the up to dateness, or the newest versions are lagging. And so you've got kind of ones that are in Azure, and in that environment, and then you've got API ones that you're running independently that has the latest and greatest. And so I'm just kinda curious about, like, how do you guys think about that from a tech stack development standpoint?

And do you feel like the safety that Microsoft provides by maybe not being on the bleeding edge is worth the consistency, and productivity that you're you get versus maybe the new features that you're doing directly API.

Chris Grebisz:

Danny, contractually, the enterprise licensing for OpenAI also, you know, has protections in place, in terms of the privatization of of your data. So we use, OpenAI and APIs for certain modeling and certain tasks. Our corporate applications of AI. And so that's everything from, the product organization is using or project managers are using. That's all using inside of the umbrella of Microsoft and the OpenAI instance inside of inside of our, Microsoft enterprise environment.

So my my assumption is that OpenAI is probably okay, But from a usability, a use case standpoint, and accessibility from a company standpoint, Microsoft is packaging it in such a way that it's very easy to deploy across the enterprise. And so and to get to the worker change management and opportunity with that, you know, whatever version we're using inside internally, I think it's 35. It's it's, it's it's good enough, for those applications.

Todd Gagne:

But I heard maybe I'll just read between the lines a little bit. It just seems like the data sensitivity is driving a lot of the decision making. So privacy of your data, so I don't want it back in the model. I don't want it being, you know, fed back in. Basically, that's driving kind of the business decision about where and what I use when.

And so you can be on the cutting edge, but then you don't have control of your data. If it's internal stuff, I'm gonna go through the environment with the protections that they provide.

Chris Grebisz:

Yeah. Absolutely. And we're and also we're processing a lot of data and content downstream from our customers. So we have to do, you know, everything possible to maintain assurances of what we're doing with that. And so it's always inside of our Microsoft environment.

Todd Gagne:

Yep. Makes sense. So let's maybe talk about Apple next. I mean, you know, Apple, came out, you know, fairly recently with kind of Apple Intelligence, and so that's their AI component of it. They are building their own small language models that will be on the phone, as well as partnering, at least to start with, with OpenAI.

But it seems like they're kind of making that an open marketplace where just like with search, you can plug in whatever search engine you want. I think this is probably, from their standpoint, pretty smart, where it's like saying, I know the individual tasks that you can do on your iPhone, and so I'm gonna localize those, and I'm gonna really focus on your data protection and privacy, and that data is not going to leave the device. And when you do decide to punch out to any one of these third parties, we're gonna be super clear about you you're leaving the Apple ecosystem, you're going to somebody else's, and your data privacy is not protected. And so I'm curious. I mean, I think, it's just kind of interesting how they're doing it, and it seems like as they see more and more of these punchouts and more and more of these user scenarios, the next version of it's gonna have more localized models that take more and more of that off and basically build it locally.

I don't know if you see it the same way or not.

Chris Grebisz:

It's the digital assistant. Right? It is. I mean It's it's like they they they are the best positioned to, you know, how being able to introduce a a daily production grade, digital assistant. I mean, it's even the little things now.

It's like it knows every day I go to the gym at 3:45, and it pops up and says it's about approximately 13 minutes. So it's already has all my data. And so Yep. Just just unleash it and, make it even more usable. But I think I think absolutely.

And and inside of the localized model on my phone, the the the scariness of releasing my personal data is kind of removed because it's, it's already in an in an environment that I've I've become comfortable with.

Todd Gagne:

What'll be interesting there too is, like, how quickly do they open some of those smaller language models to developers? Right? Where it's like the same constraints, but now, basically, everybody else that's got an app on it can leverage some of those. And, you know, how much of that is really relevant? It'll be interesting to see.

I have no idea. I mean, you can think about all the apps that are on your phone, and are there genericized small language models that are local that are are available to help you help those, apps to be enriched or not. I don't know. It'll be interesting to see.

Chris Grebisz:

Or do you have, like, you know, you do have your local model on your phone, but you have some sort of TPT like, tooling and instrumentation Yeah.

Todd Gagne:

Where Yep. Toolkit essentially.

Chris Grebisz:

Exactly. And Yeah.

Todd Gagne:

It's an SDK.

Chris Grebisz:

And I could you know, as a developer, I had I'm not gonna put apply or I'm not gonna put a a new model on your phone, but I'll sell you a GPT that will can perform certain actions or activities on on the phone.

Todd Gagne:

And so kinda getting back to, like, how they make their money. Right? They're a hardware company, and, really, they're stressing privacy, and you can see that their AI strategy is leaning into that hard. The flip side of that one is probably Android. Right?

And so Google. Right? Where it's like saying their primary way to make money is search, and I need your data. I need you to participate this. And so, I would mess I would guess that they're gonna have some element that's similar where it's, like, from a latency standpoint, it's probably more of it, where it's like performance wise, I'll do small language models, but your data could leave the phone in any case because we're gonna use it for search.

And then there's probably gonna be Gemini punchouts for it, just like they have today. But, basically, privacy is not your issue, and it's going to be really the whole business model is geared towards the search component of it.

Chris Grebisz:

Yeah. I think because that's, yeah, that's that's what they that's what their revenue is. And so so then it's that's, you know, how well, it'd be interesting to think about is, like, how because right now on our iPhones, we use Chrome and search all the time, and so they're integrated well. And there's usability and feature sets to support that. And so now if I have a digital agent, my Apple digital agent, I'm still gonna use search because it's still it's still is best of breed.

It'll be interesting to see how they combine or modify that that new sort of ecosystem out there, how they're gonna interoperate with each other, because I I want the advantages of search. Right? And I want Yep. The I want I I like where Gemini is going in terms of giving me all the reference content. It's giving me a pre prepared thing.

It's not perfect, but I like the concepts that are being presented. And so and now if my digital agent can actually execute those searches and just give me the summarization or something, I mean, there's potentially interesting UX out there. Do you

Todd Gagne:

think this drives a wedge in this in in this kind of, you know, iPhone versus, Android model? Or you think alright. The line's already been drawn. You've picked your camp, and you're you're you're good to go. I I get the reason I ask is I think that the model that that, Google is having with Android and the assets that they have, they probably have way more flexibility to solve interesting problems if data privacy isn't your major concern.

Whereas, Apple, because of their privacy and what they've done, they're going to have some constraints on all the different problems they can go solve.

Chris Grebisz:

Well, I think also that, inside of the Android environment, there's greater opportunity for innovation. Where inside of a walled garden like Apple, you're you're you'd the innovation is only from Apple or largely limited to Apple where Google can sit there and it's released it out into the wild and it can be observing all the different innovations that are occurring. And so that might accelerate interesting applications.

Todd Gagne:

Yeah. And so it'll be interesting to see if they ended up getting any sort of market share, which maybe to pivot from a market share standpoint of just talking about Europe. Right? So, maybe the the tagline on this one is, you know, Apple, with all their Apple intelligence and their AI features, has basically said we're not going to release this in Europe. And there's a lot of regulation, GDPR and others, that basically have restricted, their ability to innovate in that space because of the legal registration regulatory issues there.

And so, you know, do you see over time a less, innovative, less functioning iPhone that basically is popular in the European market?

Chris Grebisz:

I I think, you know, with GDPR and the digital marketing act and the sort of continuation of legislative privacy, and data protections. It's not just Europe also. It's like India has has created some laws and different markets around the world. But I I I do think it's going to, slow down the pace of innovation or at least the adopted innovation by the by the consumer out there or the business out there. And, and so yeah.

It but at the same time, it's such a we're so global. Like like, we have a large presence in China and where there's restrictions on a lot of softwares like Google and and Facebooks. Our teams see it every day. I mean or they, you know, they know what's out there. They're still using their the the the the, legal softwares inside of China, but I don't know how long you can sort of holster that innovation and accessibility in certain countries.

It's gonna be more successful, but in Europe, I just don't know how long how long they can prevent access to certain tech interesting tech. And so

Todd Gagne:

I mean, for Apple, it's 10% of their total revenue. And I, you know, I don't know, you know, Google and their strategies probably doesn't really fit very well for them either. So I don't know. I I think it's definitely gonna put them behind from just an entrepreneurial innovation standpoint in this space just because the regulatory environment is much more difficult.

Chris Grebisz:

Yeah. Yeah. Without doubt. This is interesting also. We, we've been investing heavily into our AI ML development team.

And, we have, without a doubt, found more talent in the United States than we have in other parts of the world where we have offices that we have a large dev organization in in Barcelona and a growing dev organization in India. But I mean, to your point where, the United States does seem to be in front of the curve in terms of of building these building this innovation as well as this the the talent associated. Now it doesn't mean that there isn't great innovation like Mistral taking place in Europe. But, I think, you know, the interference of government is going to not be to the benefit of of those countries trying to develop innovate innovative centers.

Todd Gagne:

Yeah. I agree. I think it's just an interesting topic, and it'll be interesting to see how it continues to play out. But it it'll be it'll be an interesting time. So maybe pivoting a little bit back to, like, the entrepreneur.

Like, how like, if you were starting a company today, and you knew that you had a good use case, and it could involve, API, or it can include AI, How do you think about it? Right? You you can't, are and maybe this goes back to the way you're thinking about your tech stack today, where you're basically doing to standards, you're doing to the you're not marrying yourself to one, application, really API first type of mentality. Are there other things that you think, entrepreneurs should really think about as they're trying to build their 1st MVP? Right?

We're not trying to build the Taj Mahal here. We're trying to basically solve a real world use case and see if there's a market for it.

Chris Grebisz:

Yeah. I think, 1, trying to build in as little tech as possible and use what's out there. Because you I think inside of some of these these evolving technologies, you'll run into tech deck fairly fast. I think it is really important to understand deeply understand the problem you're trying to solve and then look for a solution to solve that. And that should be a composition of roll your own tech, other services out there, and and and just make sure it's done in in in an agile enough way that the next moment we have 6 months from now, we can you can quickly pivot.

And so like I said, for some of our own instrumentation, we're we're we're calling it, you know, a a garden a model garden. So we're using 3 different models to perform similar tasks. That's great for research And, but it also allows us to have a little bit of agility and nimbleness as to how these and and figure out, you know, what what models are best for what types of tasks out there. And so but understanding the problem and even for us like this we've moved into this marketing tech. We found after doing customer research, you know, what's really really important to to the customers we're trying to, sell to is easy.

They they don't wanna have to learn new software. They don't have to wanna have to learn new UIs. And so we're giving them an enterprise grade software, but they'd never have to leave their home base. And so, you know, we give them a simple plug in to Chrome and it punches out and get it will start an AI workflow for them. And so That's pretty slick.

The the customer never leaves, you know, where they're comfortable. And I I see more and more stuff like that happening because, you know, we want things to be easy and and we have more and more work, so I don't wanna have to learn new things. And if I can't solve it with chat kpt, then I'm not interested.

Todd Gagne:

So would you say back to the kind of the entrepreneur piece of it, start with kind of these large language models, plug them in to solve the specific problem, even if it's overkill, right, where it's like, you don't have to build anything. And then basically, as you get more and more data, then look at localized models, that maybe are more specific tuned for your problem specifically. And the investment's there, but you've already identified there's a demand for it. And so go generic to start with, and then over time, get more specific and and a higher degree of ownership.

Chris Grebisz:

Yeah. Because I think the differentiation in the USP is gonna be inside of that specialization. And and In turn. We can yeah. And and you can do that in terms of multiple specific solutions inside of one topical area or functional area.

Or, you know, you could have a similar like, we're actually going more horizontal. So, like, we have a translation service that we'll use with AI, but we can persist that across different verticals. But each tie each vertical we go into, we want it to be very, very customized to that particular vertical. So if a marketing operations manager is doing translations, they want it. They they use certain words and they think and talk in a certain way.

If a paralegal is doing translations, they have a different user experience. But they all also have kind of their home based systems, and we wanna keep them inside of their home based system but have access to be serviced if that makes sense.

Todd Gagne:

Yeah. It does. So maybe towards the end here, Chris, like, I I'd almost take the bear case. I I shared with you, I think it was a Goldman Sachs, you know, article or, like, a PowerPoint on, you know, are is there really, a benefit to all this long term? I think a lot of it is talking about maybe it's a $1,000,000,000,000 to build out all this infrastructure.

And do we really have a $1,000,000,000 ROI that's really behind this? And so and I think, you know, and as I think part of the innovation curve has maybe slowed down a little bit in the last 18 months, like, you know, from the 18 months, it was like bam, bam, bam, maybe the last 6 months aren't as, materially changing. We're talking about 50 at some point in time and what that's gonna do. But I'm curious, like, from the bear case standpoint, do you feel like, we're in early innings and then, you know, like, how do you think about this? It sounds like in your own business, you're finding lots of low hanging fruit to go continue to do.

And, and you drew a parallel that you were talking about even just like when you first got on the Internet.

Chris Grebisz:

Yeah. I I I do think it's early innings, and I think there's actually I don't you know, I've read through those reports and, you know, they're looking for these home runs. And I think just like with email, you know, it was it wasn't necessarily email in the in the in the nineties. It made it easier to communicate with people. I didn't have to fax.

I didn't have to call someone. And I think as companies get comfortable with the tech, it's been I think one of the big things also is it's like it it's taken a while for people to have trust in it. And so once you have trust in the technology, then you can start looking for applications. And I said, I think I last I checked, I mean, we have 22 different applications in our business of using AI models to to improve. Now not one of them is leading to something that's gonna show up on our p and l, but the accumulation of them is making is I I do have deep confidence that it's making us more productive.

It's improving some work, jobs out there. And and it's only now once we start applying this tech into our day to day work, we start getting better and better ideas as to what this could do for maybe we get a we get a triple at some point next year as to where this could be improved. So I I think so. The 1,000,000,000 or 1,000,000,000,000 and a half investment, I mean, I guess that's beyond my pay grade. It's like it's a I it's like it's out there.

It's being used. I don't I don't see how it's bearish because it's just look. I mean, and even if you look at OpenAI's, licensing, what that they have, like I mean, the the pickup, they have a 100,000,000 or 200,000,000 users. I mean Yep. There's a good percentage of those people that are using it for productive ways, every day of their life.

So, and does the accumulation of that benefit equal that investment? And so I don't know.

Todd Gagne:

So far it does for you, which is what you're saying.

Chris Grebisz:

Yeah.

Todd Gagne:

And so do you think there's a 10 to 15% efficiency in your organization for those 22 different AI applications? Do you think it's more or less than that?

Chris Grebisz:

I don't know. I I bet there I bet it's close to that. And and so and I think as as it becomes more, standard procedure, part of our tech stack, part of our our, digital experience inside of the company, and we're using it in more and more areas. I think, you know, if we are in that 10 to 15% now, we will be very shortly, and then I think we'll only accelerate from there as it persists across the organization. So That's pretty cool.

Todd Gagne:

And that that's good.

Chris Grebisz:

I've talked to some of our data people. You know, I know they're writing a ton of SQL with it. I was talking to a kid who was in a data science program, and he had to take a course in r and, or or he had a couple of courses in r. He chose not to take the second one because he knew enough of of it from the first one that now he can just use OpenAI. And so that is just it's all sorts of different applications of it.

Right?

Todd Gagne:

Yeah. Yeah. Well, good, Chris. Well, I think, you know, we've used up 52 minutes. I would like to leave some time for some q and a.

I think this has been really good. It's been kind of a rump. And, you know, honestly, I guess the thing I would say thank you the most for is just being honest about kind of what you're doing in the application of your own business. I I think you saw this as a change agent. I think you embraced it.

You tried to find different areas. We're still in early innings, but I think the real practical application of this has been probably some of the most interesting part of the conversation, at least for me. So thank you for doing that.

Chris Grebisz:

Yeah. No problem.

Todd Gagne:

1. Thank you. Mike?

Mike Vetter:

Alright. Guys, great discussion. I enjoyed all of that. So thank you for sharing your experience, Chris, and and for you asking the the tough questions. But there's more tough questions, Chris.

So you are an incumbent. You're an incumbent who's also a disruptor. So I would like to hear your advice for early stage startups that are resource poor. How can they compete with the resource rich incumbents using AI? What's your advice to them?

Chris Grebisz:

I I think about this a lot, as an incumbent because an incumbent has to think about, like, a disruptor because the disruptors can come in and just undercut you. And incumbents thought we're resource rich, but we're slow as shit. It's just like it's like it's just a fact. I mean, it's like producing you know, releasing technology to our production environment. It's in you know, it just takes a long time because of obligations and responsibilities, and so we can't be as innovative as I would like to be conceptually.

And so I think I think as a start up, going back to that problem you're solving, the the more specific that and then you can also align the solution that you're producing to an outcome even if it but a very specific outcome, you can get opportunity inside of an incumbent like us. You know, I we are looking at a variety of techs, and we use a variety of techs that come from smaller companies and and early stage companies. Because, you know, we aren't gonna have an early stage company replace Workday for us. But we're we're looking for a very specific point solution. We'll look at across the spectrum of companies out there.

So

Mike Vetter:

get focused and add a ton of value in a narrow area is what you worry about.

Chris Grebisz:

Yep. Yep. K. I like it. Our strategy as well is we're we're trying to take that similar strategy understanding that the tech is allows us to to be more specialized and verticalized.

Mike Vetter:

Alright. So, Todd, I want you to take a crack at this one. What would you tell nontechnical founders, to focus on when they think about, pushing AI into their product strategy?

Todd Gagne:

I mean, I think it's probably twofold. I mean, I think some of this is what we've already talked about, but one of them is just tinker, man. I I think, like, spend time, doing what Chris has talked about. Right? Where it's, like, everything you can think about, try it.

Right? Try conversations with it. Try, to understand the boundaries of it, where it's helpful, where it isn't helpful. And I think don't treat it like a search engine. Right?

I think that's like I think so many people get turned off where it's like, this sucks. It's not like Google. And you're like, okay, but it's not solving the same problem as Google either. And so I do think that the more I've fiddled with it, tried different things, had success, have not had success, I've been amazed at, like, just the plethora of ideas and ways you could go use it. I write every week, and so even just trying to clarify my own thinking.

And it's like having somebody like Chris to talk to and bounce ideas off of it and come back and say, is this sharp enough? Does this make sense? What would the next logical piece to you? Like, those types of conversations, I think entrepreneurs should be doing all the time. And they should first think about, like, everything from logo design to, you know, how I'm communicating.

Like, if they start thinking about that for their own business, that curiosity, I think, will carry them a long way. The second piece to it is kind of what Chris was talking about about, about like how to leverage these models is there's Swiss Army Knife and there's an API with it. And so if you start with this and you kind of understand some of the opportunities and limitations to it, and you're trying to use and find ways to API into it and solve your problem better, I think those are 2 things functionally you can really go do well. Chris, would you have anything to add to that?

Chris Grebisz:

Yeah. And I I can't emphasize the tinkering. You you kinda you can't treat it like technology, actually. You kinda gotta treat it like the digital assistant because it takes to you you you build a skill set around prompting and and the refinement of your prompts. And and the more you pursue, like, a particular thread, what I've learned, you actually, the outcome gets much better.

But now I've I've actually, you know, kind of retrained myself as to how to become a better prompter. And that's it. That's a real thing.

Mike Vetter:

We see that too in our cohorts that, when you have founders who are trying things, sharing that knowledge and figuring out what works is a really interesting way to innovate. I know you guys have done that together. So alright. Last question. This is a chicken and the egg question.

Do you start with the problem and try to find an AI to solve the problem, or do you start with what AI can do in your tinkering and then try to go solve problems knowing what AI can do? So, Chris, I'll let you start with that one and then Todd shake his head already.

Todd Gagne:

So Yeah. Yeah.

Mike Vetter:

So, Todd, Todd, you get a first crack because you're you're shaking your head. So go ahead, Todd, and then what we'll have Well,

Todd Gagne:

I always think with entrepreneurs, like, right, we say what problem are you solving? Right? It's always the problem. What problem are you trying to solve? Right?

I mean, if you're looking for AI and saying what problem to go solve, it's it's a little bit of a whack a mole. I think you're trying to go find a real world problem, understand it deeply, understand the customer scenario, understand pain. What are they willing to go through to change it? And then is there an AI component that that makes it better? Right?

It's not always the solution. I think we're in an in an area today where, like, it's not a black box. You just throw the problem at it and it solve it. But I think most of our entrepreneurs that we talk to, they don't understand the customer, and they don't understand the problem set well enough. And I think it's a very dangerous thing to say, well, AI will figure this stuff out for me, and I'll just add technology to it.

So that'd be my 2¢.

Mike Vetter:

Chris.

Chris Grebisz:

Yeah. Like, as an entrepreneur and I'm starting a business, I'm trying to make something that I can sell to someone. And they're only gonna buy it if it's if it's solving a problem or it's bringing value to them. And so I I just believe wholeheartedly that you have to start with a problem and then and then the outcome is, you know, is is what you you're particularly getting paid for. Yep.

Mike Vetter:

Great. Well, Jocelyn, thank you so much for sharing your knowledge and experience. That actually, ties in nicely to the wrap up and that if you are a founder on the webinar, today or you're listening to our podcast later, one of the key things that we dive into is how to map customer problems to, what AI can do. So we are going to send all of you, a copy of our AI product strategy white paper, and we'd love your feedback on how that maps to your challenges. Also wanted to remind you to subscribe to our Substack where we have articles on this very topic that will dive into more details on exactly how you can, turn your start up idea into an AI driven startup idea.

So thank you all for coming today. We appreciate your attendance, and thank you so much, Chris and Todd, for sharing your wisdom and experience. It's super valuable for the next generation of entrepreneurs to be thinking about this. So thank you all for attending and have a great day.

Todd Gagne:

If you enjoyed the podcast today, please just take a moment to, like, rate it or comment on it for us. This feedback really helps us, and it helps us get the word out to, like, helping other entrepreneurs and founders. Thank you.