One of the many disorienting things about this AI revolution is that we are collectively inventing a new reality, which, I probably don't have to tell you, is much less utopian than it sounds.
But what's also interesting is how we're watching in real time, as both leaders and ICs are, I dare say, struggling to navigate the messiness that comes along with forging a new path.
The questions around things like workplace ethics, output quality and impact on future outcomes are on the minds of everyone in the org chart, but that also means we have a great opportunity to learn from each other's breakthroughs and design our own decisions around the reality we want to end up with.
My guest today is Andrew Sachs, VP of Product at Smarling.
Smarling falls into the translation and language services industry, which is practically a poster child for industries impacted by AI.
When Andrew started at Smarling, their team was performing translation services manually, and AI was just an old sci -fi movie featuring Haley Joel Osment and robot Pinocchio.
Needless to say, things have changed.
In this episode, you'll hear Andrew's take on separating genuine AI value from inflated hype, ethical considerations leaders need to be thinking about when deploying AI -powered workflows, and why the skills in highest demand have shifted faster than anyone expected.
Let's jump in. Oh, by the way, we hold conversations like this every week, so if this sounds interesting to you, why not subscribe?
Okay, now let's jump in.
Welcome back to the Product Manager Podcast. I'm here today with Andrew Sachs.
He's the VP Product at Smartling.
Andrew, thank you so much for making time.
You're busy schedule to be with us today.
Absolutely great to be here.
Can you share a little bit about your background and how you arrived at where you are today?
Yeah. Do you know what?
I've always had an interest in like, probably the mid 90s.
I had an interest in internet, and it was kind of new, and I've always kind of been in the world of technology and studied a lot of technology.
All of my jobs have kind of been in development and product, and I was kind of a fake web developer for a while, making not the best websites or code.
I eventually moved into product, and I've been at SmartLang about 14 years, I think 15 years in July.
So quite a while. But yeah, always heading up products at SmartLang.
Yeah, cool. Yeah. 15 years is no, nothing to sneeze at.
It's not a little bit of time, that is for sure.
Especially in the tech world, it's like, you know, dog years.
it. Yeah, it's like my my longest relationship, I think.
So today we're going to be focusing on the evolving ethics around AI from a leadership lens.
And so to kick us off, you want to frame the conversation through your observations at SmartLink over the past 15 years of your 10 year.
So during your time, obviously, you've seen that SmartLink is in the translation industry.
So obviously there's been a lot of big technological transitions that have happened since you started the company.
So there's cloud adoption, transparency, and now we're in the of AI.
So what is different about this shift?
Yeah, absolutely. So like, yeah, so I'm in the translation business, like return English into French and French into Spanish, and that type of thing.
And like, when you think about that process, it's really been a lot of people and like, translating things as centuries old, it's always kind of been there, turning languages into different languages.
And it's always been really a people focused endeavor.
And thinking about the other technological advances over the last 15 years and obviously cloud was probably the biggest one, they were really kind of like business and organizational impacts, I guess.
And this shift to AI really has already had a profound impact on the translation world, and probably elsewhere, really is going to start impacting humans more and more.
And that like, the, although cloud was a huge impact and allowed to scale businesses and not have to set up server farms and all this type of stuff, it was largely kind of like the end people maybe didn't feel that except they got more apps and businesses and things like that.
But it's the day -to -day workers that I think are most impacted by today's AI shift. We certainly see that in the translation world.
Yeah, I can imagine.
It must be really amplified in that specific space.
So when we talk about implementing AI in translation services, like you mentioned, there's a big impact on the workforce of translators and users.
How do you balance the efficiency gains with the impact that this has on your workforce?
And, you know, how has their role changed?
Yeah, so when we look at, as long as both the kind of a software, both a SAS software company and provide tools to manage translations and the whole process, and if you can imagine those thousands of people or hundreds of thousands of people typing in translations, provide tools to manage that, and then there's, of course, all the people who are actually doing the translations, and we offer those translation services.
And largely people have been typing in those translations but reviewing translations for quality and selling mistakes and all these types of things or brand voice for an organization.
And a few years ago, like 10 years ago, like neural machine translation kind of came out, which is a pretty big step, maybe five or six or so, I can't remember the timeframe, but Google translate has been around a long time and they kind of switched to neural machine translation, which may have had a pretty big impact on quality but it wasn't the impact that allowed you to remove humans from that process.
So if you think about the process, it's someone typed in a translation, someone reviews it for quality, and then it goes back to the content owner.
And it's kind of always been that way, and even with machine translation, it was still that way.
So with AI, we're really looking at taking out the humans in that workflow, in that process, or at least changing their role.
So humans are, as the AI gets better and the tooling gets better around it, the humans come out of maybe typing in the translation.
Then, we're able to generate the best first translation.
We're able to use AI and LLMs to assess the quality of it, and then decide if it should go to a human or not.
So the human role in it more goes into a validation type of rules.
Say when we've assessed the quality needs to be checked out, the human comes in and they're like, oh yes, this is maybe not great, and maybe they find all those edge cases that the LLM isn't able to check or find correctly.
Then you can feed it back into the machine and hopefully it improves overall, but the role definitely shifts from being in the beginning, in the middle of the workflow to be at the end to do random validation or sampling of the content for quality.
It's definitely a pretty big shift in how people are working.
Who knows, eventually it might get to point where you don't even need some of the validation components.
But certainly that's the move that is happening right now.
Yeah, I find that so interesting, especially when you bring up something like brand voice, like something that's so nuanced that it can be really difficult to pick up on and really master, even if it's in your own language.
And then to kind of find out, like, you know, the translation might be, I can see that being a lot more complex for something like A .I.
to be able to adequately gauge and be able to kind of translate.
So I can see how that layer is still very much relevant and it'd be very difficult to replace.
Yeah, the brand voice is like a huge component or special terminology.
Like people have brand words and glossary terms they want to use or words they don't want to stay away from and you sort of that the translations are respecting all of that.
AI does a pretty good job of it.
They can also have some missteps.
You have to build kind of the guardrails and sometimes the human aspect of that is the guardrail.
Oh, yeah, absolutely.
even just like having used AI for content ideation and that kind of thing, there's a whole lot.
Yeah, I think that it's interesting to think about what the technology could be capable of now.
But yeah, it's interesting to think about, like, there were different roles of interference that it's not just a cut and dry, you know, direct translation gig a lot of the time.
So as far as how it's transforming from a human typing task, you know, for getting a little bit further into this validation role.
So what does the workflow look like now versus how it looked, you know, five years ago, before we kind of saw this whole age of AI transformation.
Yeah, I mean, honestly, it was like, and like we've seen even from the smart landscape, like translations, use technology for decades, desktop tools and server tools and all these types of things to manage it.
But there's always been the human typing and reviewing and assessing quality and things like that.
And it really is when we think about the cost that goes into translation and organization spend hundreds of millions and billions of dollars on translation and it can take time to turn around translations if humans are doing it.
So the workflow really changes in that it's instant, cost goes down like drastically if you're paying people and you're no longer paying those people.
So like the workflow is incredibly more efficient.
And it really is just kind of dealing with all of the edge cases and things like that.
And like, there are cases still, and even though the AI and things are improving, Sometimes if you have legal documents, you may still want humans to look at all of it or all information that really needs to be like pretty exact.
So there's still some roles for certain types of content that means humans, but evidence that that would change over time or the next several years.
Yeah, I guess as the as the trust in the technology grows and it gets to be more effective.
Yeah, it's interesting, like when we're talking about ethics, that's a space where there is a little bit of internal debate where, you know, like, ethically where where do we stand on getting to AGI and those kinds of advancements, you know?
So what ethical considerations have come up in your leadership discussions around AI implementation internally or within the translation service industry at large?
Yeah, so I would say that like, for us, when we think about, like the, at least the linguists that we work with and things all the time, and like, we've been able to, I think, pretty successfully shift people into different roles and make sure they're still making the same amount of money and things that they were before.
Just maybe they're able to do, review more content because it requires less editing and less review or things like that.
So like, their velocity has improved.
They're at just a different end of the workflow.
So like, like it's something that we're, and we're at the beginning of really using these tools, but it's something we're going to have to really consider, consider a lot more.
And there's certainly in the industry thinking about like, where are all of these people going to end up?
And like translation is a, it's a very nuanced business.
And as you were saying like, just cause you speak Spanish or maybe you speak Spanish, like, it doesn't mean you can translate documents and specialty content and things like that, it just may not come out right.
So it is a very like, kind of a thoughtful exercise.
So I think there's still space for that.
And it may even go more into like creating content in language, rather than editing existing content for translation.
So there will certainly be roles, roles for people still in the translation space.
This echoes in many ways that content space as well, where there's now that the tools exist to make the process more efficient, the skill set that's in demand and the space has also shifted where, you know, it used to be that output and an attentive to some of the technical requirements, especially SEO and that kind of thing, are very, very high in demand.
And now, I imagine it's kind of similar in the translation space, where it kind of sounds like the ability to be perceptive to specific nuances, or linguistic nuances, or style or being, you know, very detail oriented around what a client, you know, their requirements are, rather than just being able to just translate, you know, word for word. So it's interesting how these kinds of technology shifts have like a domino effect on the whole industry and kind of what is in demand.
They definitely do.
And even as we look at, like, some of the research and things that we're doing, and we'll probably deploy this year, whatever, there's a lot of trying to capture some of nuance like being able to say how does your organization perceive itself or how do you want customers to talk about you and like coming up with some of those types of descriptions and things can really influence what the AI or LLM or whatever is providing and we've seen changes in quality based on that like as we're looking at how humans are right now even in some of the validation and review areas are still like involved in some
of that nuance like that will also shift probably over time, the way that I kind of like to think about it is, you know, wherever the market kind of creates an influx in one area.
So it could be volume of output, there's going to be, you know, kind of a sort of a pendulum swing into demand for higher quality, more, you know, authentic insights or more kind of, you know, what's gonna make this stand out versus a competitor piece when they have the same tools as you.
Such an interesting time to be alive.
Yeah, it's super interesting and like, it's like seeing the dramatic shift in such a short amount of time is.
Yeah, yeah, it's it really feels sometimes as if we're witnessing a shift that would normally be observed over multiple decades.
And it's happened over the course of a couple of months.
My head is still spinning and it's so interesting.
You know, every time I have conversations like this, I feel like we have an AI conversation on the show, but once a month and every time it's crazy.
Let's get back to workflows.
because I'm curious, you know, when you're making decisions around deployment and how you change your workflow, as you said, you want to really enable velocity.
And I think that's it's really cool, because it sounds to me like a lot of folks on this workforce are really now being empowered to do probably some of the more rewarding work rather than some of the more, you know, the menial tasks that can be outsourced.
So how do you decide, you know, what workflows or what tasks should be primarily human driven, and which should be kind of relegated to, you know, this is an AI task and there's an expectation that you're using AI for this workflow?
It really depends, and like in the translation world, there's this cost -quality -speed matrix that people always try to navigate to get the best quality at the lowest price and fastest as possible.
We really do a lot of research and studying and testing to make sure that we're offering the right tool for the right type of content.
So we really have these confidence levels that we look at, and especially in our AI tools where we're confident that a certain set of technology or tools or a workflow process will produce a set of content that we can guarantee that quality around.
If we can guarantee the quality around a certain set of process or a certain workflow or process, then we use that with the customer to pick the workflow that they should use.
Some customers really do have a requirement, they went, no, I want a human, and I want two humans involved in this.
And maybe they have certain types of content that require that.
But really, like we're seeing just being able to more broadly apply technology enabled workflows to really all types of content and with one caveat, sometimes there are like long tail languages that maybe don't have maybe enough information in the LLM or whatever it might be to like really have an understanding of what the translation should be.
So maybe there's a long tail languages that still require humans, but really, like we've seen, like being able to broadly apply these tools to really any type of content and still be able to guarantee the quality.
I'd like to zoom out a little bit also in the organization because we're focusing kind of on a specific, like a specific team, really.
When we kind of zoom out at the organization that's smart link at large, how are you guys approaching adoption?
Are you finding that I mean, at this point, everybody has a use case for AI.
Are you finding that the organization is adopting AI in a fairly uniform way, or is there some differences in terms of where it's more applicable or where there's more pushback on doing things the old fashioned way?
I mean, I guess like maybe two sides to that, one, like as a software provider with customers, or customers are like, they often have mandates internally to ask for AI.
They want to use, they're not even sure how they might use it, but they have organizational mandates to make sure that they're getting efficiency, and their vendors and whoever they're using are using AI and being able to report on that efficiency.
So customers definitely want that.
And they're asking for it, and we're trying to figure out what those solutions should be to get them the efficiency or cost savings or whatever they're requiring from that organization.
And then just like even internally, just like how we use AI and AI tools, I know every team is using AI tools and certainly our engineering team is and our product team is maybe to a lesser extent, but certainly still using AI.
And I know that our customer support tools are all AI enabled and our help desk is AI enabled, it's like really everything is kind of moving in that direction.
I'm seeing more and more of that, the AI mandates, and I think right now it's interesting.
I've had a few conversations about this so far, where there's a few different approaches that organizations are using.
And I've heard it described as sort of a carrot or stick approach, where it's like you can either incentivize AI and if that doesn't work and the adoption rate is still not quite where you want it to be, then it becomes a mandate.
And I think that the purpose is to try and get people to be explorative and use the tools and get comfortable with how they can support their workflows.
But I could see that also having a bit of a...
And like, internally, we've definitely put in some policies and allowed tools and usually are whatever vendors we're using have already gone through security audits and all those of things so that we can kind of have some safety and things around that.
But like there's a new tool every day that everyone could use and I'm sure trying it out.
Yeah. So if we talk about the hype cycle, so this is, you know, the new technology, there are mandates, there's excitement around it.
There's a lot of hype.
And I think right now we're kind of coming off of a I'd say the great skepticism is what I guess I'll call it, of you know, what's hype what tools actually need AI features.
And this is, you know, this is something that happens every time there's a giant advancement.
We saw it with social media web three.
So how are you helping the organization kind of separate hype, you know, trying to ground it and where's the real value in these tools and, you know, trying to cuz I think there is kind of the other side of the coin where there's folks who are reluctant to adopt and then there might be folks who are maybe a little too eager kind of.
So how do you rein that into, you know, what's like, where's the use case for AI that is valuable and not detrimental to quality, or One of our AI researchers and engineers like he's always like, if you can, not everything has to be AI, if you can do it with like, a rule or like, if then then like, just like don't like don't spend so much time thinking that everything has to be AI enabled.
I think like when we, when you use tools like chat GPT or something, and you're just amazed that like it just wrote this huge paper for you or you typed in something that came up with all this information.
That's one aspect that creates some of the frenzy because it looks like it's just doing all this amazing stuff.
But when you actually start to apply AI to business solutions and things for customers, you need to have a different level of confidence in what it is saying.
I think that's where you start to see maybe the skepticism a little bit and you're like, oh, this isn't just like typing something into chat GPT and getting a great answer out of it.
It's like how to apply the technology to something that improves your business or process or customer's outcome.
I think there's certainly the frenzy and certainly everyone wants it, but once you get into building some of the tools, there's a bit more measured response.
You want to make sure that you're building something that is actually beneficial for people and that will be useful.
Sometimes that takes more time than just like typing something in chat, GPT.
Yeah, absolutely. I kind of feel like there's a parallel here with social media.
And when that kind of came out in mass, where there's, you know, sort of this social norm, well, every business needs to be on social media.
And some people were not taking it seriously at all or not enough and others kind of went the other way where it was like you're over indexing in your social media where to the point where it's like, where's the ROI?
So, yeah, I think there's kind of that fine tuning the balance, you know, having a playground and encouragement to explore and get people to zero in on where the good use cases are, but then into only hold on to the ones that really have a proven benefit to the business.
Yeah. When we built, we did lots of experiments and things and we've been testing one that we think will be really great for customers.
We always try to get 80 -85 percent confidence.
For a few months, it's been a coin toss of whether it works or not and AI is great, but it's still 50 -50 whether it reduces the result that we want.
So yeah, there's definitely that trust layer and confidence of, and the right use cases for the development technologies.
So looking at industries outside of translation, if you care to comment, what parallels do you see in how AI's being integrated, you know, just in the product space in general?
You know, are there lessons that you have from your own experience in your career that you think are more broadly applicable that are kind of informing how you're seeing, you know, as we move forward with this technology?
a little bit. I think like to try to keep up with the latest tools for product and product managers and things, I think that like, there's a lot of things that I haven't seen yet where it's been like, super helpful in maybe some of the product development.
And the way I say that is I think like, at least we're like, we're supposedly inventing things that maybe don't exist yet, and new tools and interfaces and things that hopefully people haven't seen or new technologies or whatever applications, I think it can be useful for writing, give me the outline of a stack and it's going to be this thing.
It would be great if it could turn that into geotickets or something like that, so remove some of that complication of what PMs end up having to do.
I think there's some cool tools that I've seen for just drawing something on a piece of paper and it turns it into a nice mock up and can point out different areas that you can focus on.
And so I think there's a lot of application that we'll start to see that will improve the, kind of just some of the, I don't know, like day -to -day, in the weeds type of things that maybe require, maybe don't require as much thought as like inventing something new.
Yeah, well, and like to the point of inventing something new, there's something that I'm very excited about is this five coding trend.
There's so much potential that I feel is being unlocked now for collaboration within teams to be able to have an idea and be able to kind of generate a prototype that can be iterated on by a team that really has a skill set to take it to the next level.
Have you seen some of those tools in progress?
Yeah. Our internal team has been playing with it a lot.
And it's really cool to see what people are inventing.
But I think what it's really neat is like, no, I feel like we've all kind of had that moment as smartphone users or whatever, where you're like, oh, there should be an app to do this thing that I specifically want, probably no one else.
And now it's kind of you have the technology to kind of try it out and see if it's you know, like, oh, this is actually a stupid idea.
I have one, I can't reveal what it is.
It's a great idea. Yeah, that's a great way for like, test to see if your idea actually does pan out or not.
Yeah, exactly. Like, I think it's kinda cool to be able to instantly validate something that you know, because that's where ideas come from.
And often it's like, well, someone's probably doing it, but maybe you can be doing it better.
Exactly. As far as the relationship between AI and human workers.
So as a leader, you know, we're talking about some of the impacts that you're seeing on the organization right now.
And I think we can both kind of see where that's kind of going for end users and for folks at large.
But what would you like to see?
Like, what excites you about how this is enabling your workforce?
And what would you like to kind of see in the next three to five years?
I think from a platform perspective, there's at least in the translation world and a lot of things, there's a lot of management of things.
A lot of people are in the weeds all the time or there's a lot of platforms people spend their entire day in and that is their job.
I think it's turning into more of a managed by exception, and that if you can trust that the AI or algorithms or whatever you're using, as creating the outcome that you want, then really you should just be managing what doesn't create that, but it recognizes it maybe didn't create the outcome that you wanted.
So I think like repositioning and re -imagining a lot of tools, including platforms like SmartLink, to be more of a manage by exception type of thing rather than A.
Also, if you don't need people to do all of these things, then the technology and the tools and all these things can be in more places than just maybe inside the smart lane problem window or whatever.
What do you think of this hot take so I'm because one of the things that I've long been critical of, just in general is I feel like a lot of the time, the tradition, when it comes to your career trajectory is that people who are really good at a specific task within their, you know, individual contributors tend to rise through the ranks based on their skill set within their core function, and then become people leaders and lack the I feel like this is the first time that because we are now in a position where ICS are overseeing technology that are executing base level tasks, we are actually starting
to train management at a IC level.
This is like a skill set shift that people are starting to develop an actual ability to manage or an understanding of what it takes to oversee a team before they reach the people management layer.
that makes total sense actually like I haven't thought about that like it's something that I will certainly think about and even looking to the sparkling team yeah I kind of think it's kind of interesting because it's the nuances of managing human being issues but I think it's it's cool to see how people are starting to think like managers before they become managers so I'm excited for that shift as well totally that makes total sense yeah yeah I kind of just see like a lot of new and maybe that's that it creates a lot of new opportunities for humans to deal with things that they wouldn't have done
before. You know, so interesting to kind of see these unexpected sort of outcomes that happen when technology has completely created a paradigm shift. Yeah, sure.
This has been really fun.
I really appreciate you coming on the show.
I love an animated AI conversation.
I'm sure you'll have many more.
Probably next month too.
Where can people follow you online, Andrew?
They can find me on Linkedin.
I think it's just a -sax, a -s -a -x -y.
Wonderful. Well, thank you for coming and appreciate the time.
Thanks so much.
