Smartling CEO Bryan Murphy on Seizing the Language AI Moment

SlatorPod #188 - Smartling CEO Bryan Murphy on Seizing Language AI

Bryan Murphy, CEO of Smartling, joins SlatorPod to share insights into the language technology leader’s position as a turnkey cloud-based LanguageAI™.

Bryan talks about his background in software and e-commerce and how he joined Smartling to leverage technology in the translation and localization business.

Bryan expands on how Smartling has built a sophisticated translation and localization management platform and expert-in-the-loop productivity technology. The CEO explains Smartling’s decision to leverage its technology to gain an edge in the language services market.

He highlights the impact of large language models and generative AI in enhancing productivity, improving translation quality, and reducing costs. Bryan also touches on recent product launches like Smartling Translate, which targets the individual contributor within the enterprise.

Subscribe on YoutubeApple PodcastsSpotifyGoogle Podcasts, and elsewhere

Bryan discusses the impact of AI in the industry and its transformative potential, likening it to the advent of the commercial internet. He elaborates on the role of human language experts in the industry, emphasizing that AI enhances their productivity rather than replacing them.

Looking ahead to 2024, Smartling aims to improve the quality for customers, simplify processes, and further boost translator productivity.


Florian: Today on the podcast is Bryan Murphy, the CEO of Smartling, the language tech company everybody knows. Hi Bryan, and thanks for joining. So I said, everybody knows Smartling, but tell us a bit more for those very few that don’t know Smartling or those that want to know a bit more. Just a quick recap for the audience about Smartling’s tech, services, clients, kind of key milestones over the past 10, 15 years.

Bryan: Smartling is a translation, localization company, and we help companies like Apple, Tesla, Disney create multilingual experiences that their customers love. We’re a little bit unique in that we are a turnkey cloud-based LanguageAI™. So that means that we provide TMS functionality, CAT technology, as well as a complete solution of language services, which makes us a little bit different. And the reason we do that is because that’s what enables us to deliver to our customers a turnkey and guaranteed human quality translation at typically 40% less cost and up to three times faster than a traditional LSP. And I think in this day and age, that’s what our customers are really valuing.

Florian: Now, with your background, tell us a bit more about your background because you’re not one of those like, oh, I’ve been in the language services, language tech industry for three decades type of people. So you joined about one and a half years ago. So rather than me trying, give us a bit more of an overview of your background and how you’ve basically joined Smartling as CEO.

Bryan: I’ve not been in the translation, localization business, but I’ve been creating multilingual websites and applications for 25 years and so my background is predominantly software, predominantly e-commerce. And when we came into Battery, into Smartling alongside Battery, one of the things that got us really excited about this business was we found this 30 billion-dollar, pretty fragmented, global translation, localization business that looked like it could really benefit from infusing really great technology. And when we looked at all the companies out there, Smartling really came to the top in terms of having had a significant experience building cloud-based solutions, building AI, machine learning and AI solutions. And if you look at my background, what I love to do is come into industries that sort of look like that and transform them with technology. Going back to my days being in the automotive business, we built a B2B and B2C solution for auto parts sellers. And we were able to significantly improve the productivity of what was called at the time jobbers. Those were like the equivalent of translators, but they were the folks that actually you called on the phone and they figured out what part you needed and they were technical and all this sort of thing and it used to be like this. You faxed an order, you had to have a phone conversation, it was really antiquated. We implemented technology in that space that dramatically improved their productivity, and as such, we ended up capturing about three billion dollars of market share. I’ve done that with a couple of other companies since then, obviously, eBay was a company where I spent a few years. Where we were really transforming and created this whole new ecosystem of small consumer sellers, once again by creating technology and creating a platform for them to sell on. And so coming into Smartling, that’s sort of the way I view it. I view this as an industry that has a tremendous opportunity to benefit from technology, to benefit from AI, and to help improve productivity, help improve quality and help improve or reduce the cost within the industry.

Florian: Now you had a very broad experience again from, sure, you kind of were saying as a user of language services, language tech. But now that you’re in this particular part of the industry, compared to all of the other software, businesses you worked for and advised for and had directorships, is there something specifically different about language, tech and services compared to all these other tech verticals or not really? Like, do we just kind of imagine there’s this unique thing?

Bryan: I tell you what, there’s a couple of things that are similar. Actually, this is going to sound strange, but in many ways it reminds me of the automotive business. There’s a lot of very passionate people here. People really have a passion for translation, for language. We’re all a little bit nerdy about it in the same way that auto people are nerdy about cars, which I love and I do think that… It’s interesting, one of the things that is different is that there’s just sort of like this disconnect from the C-suite in this business that I’ve noticed, which is really different. I’m used to having conversations with C-suite folks and I kind of step back and that doesn’t really happen here as much. That’s beginning to change and I’ll talk about that in a second. But it wasn’t so dissimilar for my service. I said earlier, I’ve been creating multilingual experiences for 25 years, but I was one of those people that had this global strategy they had to build out, and so we’d do that and then I’d say, hey, who are those people that make this the other language, right? That’s sort of the way that it works a little bit. But two things that I do know, having built and operated multibillion dollar e-commerce platforms and software companies is that things like digital footprint and conversion rate really matter. And people in the C-suite generally understand those concepts. And I think that one of the things that we’re working on doing is helping link and educate the folks in the C-suite about the potential of using language and localization to dramatically improve those numbers for them, which really affect the top and bottom line. So I think to me that’s one of the biggest differences is that this has been a function that sort of really needs to be elevated into those types of strategic conversations.

Florian: Now that’s gotten probably both easier and harder over the past 12 months, but with all the kind of LanguageAI boom, hype, whatever you want to call it. I mean, how do you position Smartling in this really fast-changing, emerging language tech and LanguageAI ecosystem?

Bryan: We talk about Smartling as being a LanguageAI platform and kind of going back to this idea of providing this turnkey solution to our customers, right? So if you think about traditionally, we’ve provided the automation layer, we’ve provided the workflow management layer, all of the translator tools like CAT, et cetera, and also the quality management tools which are really essential. The introduction of large language models has really been an unlock for us because we’ve always been really good at leveraging neural machine translation engines, right? So we’ve got this neural machine translation engine hub where we kind of pool them all together and use that to produce much higher quality translations than you would with a single engine alone, right? So that’s something we’re very used to doing and have a lot of experience in, but it sort of results in… There’s sort of like a cap in how much quality you can generate out of that, which most people on this call would be familiar with. The advent of large language models and introducing that and combining those two functions has really been the unlock for us that’s been able to take us past that cap and deliver exceptionally high-quality translation at a significantly reduced cost. And so for me, that’s really the opportunity where we can go into these customers and say, hey, listen, we can now deliver to you, guaranteed human translation, in about a third of the time with an MQM score of 98, right, so human quality and for about 40% less. And that becomes a very interesting conversation because now you’re in this business of, and this is what I refer to as rationing, right? So most of our customers, most companies, they ration the amount of things they translate because it’s expensive, right? But given the opportunity to use it more broadly in their organization, they would, because we know that it increases digital footprint, which we talked about, and that increases conversion rate. We’ve seen that time and time again. So our job, the way I see it, is to help improve the productivity of the linguist to bring down that cost, improve the speed and quality to the point where we can give our customers strategic advantage by helping them unlock as they go global.

SlatorPod – News, Analysis, Guests

The weekly language industry podcast. On Youtube, Apple Podcasts, Spotify, Google Podcasts, and all other major platforms.

SlatorPod – News, Analysis, Guests

Florian: Now, these conversations are, as I said, do you feel they’ve gotten a little easier now that we have the C-suite’s attention? And how do you retain kind of ownership of that multilingual component within the enterprise with the entire environment being so buzzy and hypey and so many big companies now also kind of discovering LanguageAI something they’re interested in. I mean, first Smartling specifically, but maybe also the industry at large, how do we make sure we don’t lose the narrative here and then eventually, of course, customers?

Bryan: Yeah, that’s a great question. So I would say that we’re in the very, very early stages of having that C-suite sort of narrative, right? So I would say that folks like myself, CEOs of other companies, they see it all over the news, right? So the first thing they ask is they say across the organization, what are we doing with GPT? How do we use it, right? So let’s make sure we’re using large language models or generative AI in our business. And that comes naturally to the folks in the translation, localization department. And I think one of the challenges that we get here is that a lot of CEOs are saying, this is great, let’s use it to create content and to translate content, that should be it, should take our cost to zero, right? And I think that this is where the education process comes and say, well, no, not so fast. The reality of the matter is that large language models and we’re benchmarking all of the major ones now just the same way that we benchmark all of the major machine translation engines. It’s good. It’s not, frankly, from a quality perspective, as good as the NMT engines today. And it’s got a similar cost profile, so costs roughly 18, 20 bucks per million words for machine translation engine and it’s not a dissimilar cost for GPT and some of the others. So it’s not quite ready for primetime yet. However, it is getting much better and it is combined with these other solutions, resulting in much higher quality translation. So I think that the challenge is helping them understand that, yes, this is a huge opportunity for you to reduce your cost of translation and improve your digital footprint and conversion rates, et cetera, but it’s not a standalone solution. And I’ll give you just a couple of examples of that. Let’s take first and foremost the top of mind for us and for all of our customers is quality. Great, so I’m going to use the GPT prompt and it’s going to create a whole bunch of content for me and I’m going to say, okay, now translate this into 20 different languages. Great. Is it any good? I have no idea. Whatever source language I used to create it in, I don’t speak those other languages, so you do need a tool. You need a platform to be able to measure, to do LQE and auditing, and that’s really what a solution, a platform like Smartling does, right? So we’re able to provide that, you create the content and then you do your language quality estimation during the process, and then you do your MQM auditing to be able to say, yes, this is actually very high-quality content, or no, this needs to be adjusted in some way.

Florian: Tell me a bit more about Smartling Translate. I think it’s something you’ve launched. I hadn’t seen it like a year ago. I think this is somewhat new and you say on the website it’s GPT-powered, can you just tell a bit more about that one?

Bryan: Sure, so Smartling Translate is a product. It’s basically an interface. It’s a very simple interface where you can drag and drop files, you can cut and paste and select the language that you want to translate to, and out comes your output. And this really came to light. We have thousands of enterprise customers, and one of the things as being kind of like a big enterprise solution is that we didn’t have a great interface for individual contributors that wanted the availability of that automation, dynamic workflow, quality management, et cetera. So we’ve created Smartling Translate, which is a very simple browser solution where you log in and you just do your translation and out it comes. And the advantage of this solution for the individual contributors within these enterprises is that it gives you safety and security, privacy, confidentiality on content that you may be putting in there, as opposed to other tools that they might use that are available on the web. It’s integrated to workflows, it’s integrated to linguistic assets like translation memory, glossary, style guide, et cetera, right? So it’s like an enterprise simple solution for individual contributors and we make GPT available in the platform as a translation engine. So if you want to use GPT as a translation engine, it’s available within the Smartling platform, which our customers really love, once again, because once again, using GPT as a standalone is like, okay, great. It’s not integrated to any of my systems, it doesn’t take advantage of any of my linguistic assets, and I can’t check quality. So is it really that helpful? I would argue no, but as a part of a platform like Smartling, where I have all of those things, it is actually incredibly useful.

Florian: Also you can rephrase text. You have all kind of options right on the platform. Hard to show here unless we share a screen. But it’s this new generative ability of GPT type tech to rephrase and maybe shorten lengths and all of these things, right?

Bryan: Yeah, that’s exactly right. So, for example, we can deliver a translation and then we can prompt some questions and we can ask you, hey, do you like this translation? And we give them prompts that say, no, make it shorter, or change the style of it. I want this to be, for example, I want this to be a business style, or I want it to be more of a casual, fun style, and we can fix gender sort of issues, right, the machine translation tends to break, all kinds of things. And this is really kind of the beauty of large language models, of generative AI is that with building out these prompts, and we have now several patent pending prompts, we’re able to significantly improve quality of standard machine translation in a very meaningful way.

Florian: At Slator our position is that TMS’ or TMS providers, historically, I don’t know now, probably there will be a different term going forward, but let’s just still refer to them as TMS providers are kind of ideally positioned because you can add features such like you just described very, very quickly to the tech and so it’s very integrated. Now, where do you see kind of the most impactful and useful, maybe emerging use cases, some of these LLMs in the next two to three years? A big question, admittedly, but just two or three things that go beyond, like to shorten or expand going forward.

Bryan: Thank you. Well, we agree too, that TMS is ideally positioned and an important part of what we do. But if you think about translation, if you’re an individual contributor, you just got a few things to translate, then typically the web, and you don’t require privacy, security, then web based translation tools tend to work fairly well. But if you’re a business that’s doing a significant amount of translation, you really do need a platform to manage that. There’s a significant ROI. We’re typically looking at kind of like a six X ROI for our customers, what they save versus what they spend. And I think that when you think about the fundamentals of those, which is automation, dynamic workflow management, integrated translator tools, and most importantly, quality, right, so quality, LQE and MQM, auditing solutions that’s sort of like the ideal platform into which to insert large language models. So where you’ll see us going with this is kind of two directions. One is, once again, I spoke earlier about so we’ve got a lot of experiencing building AI-powered neural machine translation hubs, right? So taking all of the major machine translation engines, harnessing them together, and using that to produce a higher level quality of output, typically we’ll see up to 350% higher quality as a result of sending out to the engines at a string level, right. And then using AI to score the returns and then picking the highest quality score is typically how we do that. We’ve now begun to use LLMs in two ways. One is once again, sort of pairing that LLM with the results of that or with that process so that we can do preprocessing, in other words, correcting or improving copy before it hits the machine translation engines, translating it and then doing post, right? And this is what’s really unlocked a significant quality boost is by pairing these things, right, so there’s the pair. Additionally, we’re also beginning to use the Gen AI or the large language models to do the actual translation itself. And we’ve started with one with GPT, and now we’re beginning to add the other machine translation engines, which then our customers can either use because that’s what they prefer, or we’re also in the process of bundling those together in the same way that we did with the machine translation engines. So I think in two or three years time, I suspect that all of those will merge together into a hybrid, right? Because they’re all basically being produced by the same companies, by and large, right, so they will ultimately merge those together and you’ll end up with a very, very powerful machine translation and large language model translation engine underpinning all of this translation workflow.

SlatorCon Silicon Valley 2024 | $ 1,340

SlatorCon Silicon Valley 2024 | $ 1,340

A rich online conference which brings together our research and network of industry leaders.

Buy Tickets

Register Now

Florian: You mentioned at the very beginning what’s kind of fairly special about Smartling on the very sophisticated technology side of things is that you also offer services. And so tell me a bit more about the services component, how you think about that in terms of scaling the business, but also let’s stay with a two to three years kind of time frame. Where do you see the human language experts in all of this?

Bryan: That’s a great question. So the way I view the world, you heard me talking a little bit earlier about how we’re… I’ve always been a fan of when I like to go into a new industry and think about transforming it, right? The human is always at the heart of what we do. So I think about back several years ago, we used e-commerce technology to empower these jobbers, which are these auto parts salespeople. It took away sort of like the daily grind of their job and enabled them to focus on the really high value things that they did, right. Same thing with sellers at eBay, right? And now in this industry, I see the translators as our customer, the translators and the project managers, the localization managers, right? So what we’re doing is using large language models to help take away sort of like the daily grind of translation, right, the non-value add stuff like the workflow management assignments, et cetera, as well as sort of like the raw translation. And then enabling them to put them in a position to really use their skills and their experience in the most high value way possible. So to sort of bring that to life a little bit, in the last 18 months or so, by doing these things, applying AI, introducing large language models, we’ve been able to improve the productivity of our translators by almost three X. In other words, they’re able to produce three X as much content as they could before at higher quality, right? So that’s a significant improvement and I see that… That to us as a goal and that to us is an accelerating number as we integrate more of the capabilities and understand more of the capabilities of large language models. So I think that in two or three years time, I see the translators as being high value ad operators, managers of these large systems, and they’ll have incredible value as they’re producing more and more work. And I’ll say one more thing, because a lot of times people… Well, does that mean that we’re putting translators out of out of work? I don’t think so. In my experience, what happens is that, and let’s take this rationing thing that we talked about before, once again, our goal is not to necessarily reduce budgets, right? Our goal is to deliver more value. So I think that we know that the more translated content you have out there, the bigger digital footprint you’re going to get, the more SEO traffic you’re going to get, the more higher the conversion rate. So companies are hungry for this to unlock and to unration. So I think that there’s going to be a tremendous opportunity for skilled technical translators in the future.

Florian: What you’re describing is kind of the elasticity of demand and for localization, translation, multilingual kind of in general, which we’ve discussed with investors over the past five, six, seven years when they approached us to do some industry analysis. And we’re like, well, as things get cheaper on a unit basis, appetite for higher volumes is going up. And I think the next two years are going to prove that probably, hopefully, that indeed accessibility to this will actually drive demand for just more and more and more and more content.

Bryan: I’ll give you a great number. The amount of content that we’ve produced in Smartling has increased 700% over the last 36 months. I wish our revenue had increased 700% but, well, it’s done well so that’s huge value creation for our customers.

Florian: Now with basically not just content conversion, but content generation being something that AI can deliver, who knows what’s going to happen to the content conversion part at kind of human level. So this is a very tricky environment for a CEO to navigate and there’s also these two forces. On the one hand, we have a macro slowdown, risks for kind of the global economy increasing generally, budgets, inflation, et cetera. But then you have AI as kind of the single buzziest, boomiest topic out there. And I mean, a company like Smartling obviously is exposed to the macro. Yeah, people have budgets and they need to make sure that they meet those budgets and maybe shrink them a little bit. But on the other hand, you have this incredible AI boom, and you’re also in the middle of that. How do you navigate something like this on a CEO level?

Bryan: It’s tough. I’ve been around for a while now, so I’ve been around at the beginning of the internet, not the very beginning when it was ARPANET, but since the 2000s, right? And I’ve been around through cloud, right, through the cloud and crypto. And in my opinion, I don’t think this is hyperbolistic. I think I’ve got a little bit of foundation here. But this is to me, what’s happening with AI is going to be as transformative as the sort of advent of the commercial internet. It’s really having that type of impact, so it’s really, really important that we pay attention to it and we really lean into it and make sure that we’re able to deliver value, right? So once again, there’s a ton of buzz around this, right? There’s always a ton of buzz, but where there’s smoke, there’s fire, or whatever analogy you want to use, we see the impact that AI is having for our customers daily, right? Once again, we talked about the acceleration of growth, 700% increase in content delivery in the last 36 months. Our ability now to deliver guaranteed human quality with a 98 MQM for 40% less in three times, in a third of the time. Those are all meaningful impactful things, and frankly, I think that’s one of the things that’s driving our business in a time when the macros are soft, have been soft, right? So especially since the second quarter when Silicon Valley Bank went bankrupt, they got a little chippy there. But then since that, this is the fastest our business has ever grown. And I think it’s our ability leveraging that technology to deliver essentially a solution that’s faster, better, cheaper than the traditional solution.

Florian: Talking about building, in terms of the parts that you want to build and the parts you maybe want to go for kind of a more best of breed because there’s so much, there’s so many layers to that technology onion. Sorry, terrible metaphor, but still, how do you think about that? Like, what do you want to really own, build in house? What do you want to maybe get from a third party? Obviously you’re constrained in what type of details you can share, but just your general kind of thinking about that.

Bryan: I think the things I can talk about are when I think about neural machine translation engines, right? So we’ve built an engine, we have an engine. But I think those are the kind of things where a lot of large companies are investing billions of dollars into building these types of engines. So I would say that’s probably an area that we would not get involved with in specifically building our own. And I think that that also probably applies to large language models, too. Once again, same thing, there’s probably at this point, trillions of dollars going into building those things. And I think that those are, I guess they are kind of commodities to me. They’re these big giant things that then aren’t all that particularly useful in a standalone environment. But when you combine them with a platform and the types of capabilities that we discussed in terms of automation, workflow management, quality estimation, and very specific prompting that we do here in our industry, they become, on a value add basis, incredibly valuable. So that’s the way I view our job is to create this business solution layer for our customers, for enterprises that enable them to access the benefits of these engines, whether they’re machine translation engines, whether they’re large language models or they’re the human linguists that make them come to life.

Florian: Yeah, and I guess in a scalable and secure way as well. Sometimes kind of if you’re an individual user at a company, you think this works for you now, but if you want to onboard 10,000 people at an enterprise, it might not work at all and not in a secure way, right?

Bryan: That’s a whole thing. I think there’s a lot of attention that’s being paid now to large language models too. You seem some companies that are outright banning their use because they don’t understand or they don’t want that access. So the way that we address that is two ways. One is we have optionality where if you don’t want that functionality, it’s basically an on off switch. We can eliminate that, but also we have special agreements in place where we are able to protect privacy, to provide privacy and security. I Recruit Talent. Find Jobs

LocJobs is the new language industry talent hub, where candidates connect to new opportunities and employers find the most qualified professionals in the translation and localization industry. I Recruit Talent. Find Jobs

Florian: Another area for buy and build where that term is used is M&A. Where does Smartling stand on this? If you can share a bit of light about your strategy.

Bryan: I can’t provide a lot of detail on specifics of what we’re after. I will tell you in general, we are a net buyer and we’re prominently focused on technology companies that fill a need for our customers.

Florian: In terms of the roadmap, key initiatives. We’re almost not quite there in 2024, but soon, so what are you and the team most excited about for next year?

Bryan: Yeah, we’re definitely getting close to 2024. Yes, so we’re really focused on three things, three pillars. One is quality, right? So how do we improve quality? How do we automate quality measurement at scale? Those are the things, that’s one pillar. Number two, the second pillar is improving what we call the quality of life for our customers, whether they’re once again localization managers or project managers or translators. So once again, sort of taking the daily grind away from them and automating that solution. And thirdly, to what we discussed, improving the productivity of the translators. We’ve seen a three X improvement. It’s hard for me to sort of forecast what that will be like a year from now, but I can almost virtually guarantee you it’s going to be significantly higher than that.