#280 Walmart Cuts Translation Costs, 10 LTP Growth Hacks

13:36Esther Bond:Okay, but somebody needs to approve this at some point. I mean, I don't think they have hundreds or thousands of these sub-use case pages. So it is interesting. Maybe it's like even an SEO game or something that there is something called Chachapiti Translate now People don't wander off and use other solutions, but hey, okay, we went through this a couple of weeks ago. Let's go through an announcement by Levin Labs, which, you know, speaking of fast growing companies, they're definitely growing faster than 7%. I think they announced that they're at like 300 plus million ARR and they struck an agreement with Deutsche Telekom.

14:15Esther Bond:and it really, I want to flag this because it's around live translation. So Deutsche Telekom, probably one of the biggest telecoms in the world. So this seems like a big deal for kind of AI interpreting or AI live speech translation, very much kind of deeply embeds 11 Labs in Deutsche Telekom for, yeah, live translation. They had an announcement out where they say AI support in the call once activated during a call, the assistant, so the assistant that's probably powered by 11labs provides real-time support and initial features will include live translation across languages, conversation summaries, contextual assistance, blah, blah, blah, but live translation across languages.

15:01Esther Bond:So 11labs, Deutsche Telekom, and 11labs powers the live translation component. Well, to me, this seems like a big deal. So Esther, what are your thoughts on this kind of initial reaction on that?

15:13Florian Faes:I mean, I think it does seem like a fairly big deal. I'm personally pretty interested in seeing where these kinds of solutions are being embedded. So, you know, AI interpreting within kind of a very, you know, established and serious business, something like Deutsche Telkom, obviously. I mean, having this kind of solution, I think, will be quite impactful potentially. I mean, I don't know quite how many sort of conversations are happening through AI interpreting. Yeah, millions, right? Probably millions.

15:51Esther Bond:I mean, this is the biggest channel you can get if you're an AI interpreting or AI live speech company.

15:59Florian Faes:Yeah, this doesn't sort of strike me as like a pilot. But, you know, like this is this is pretty serious level of integration. So, yeah, I mean, it'll be interesting to see where they go with it as well. I think they said the capabilities arrive first for customers in Germany this year, and then they're going to be providing support for up to 50 languages. So it'll be interesting to see if we get any level of analysis around how it's working, how it's scaling, what kind of challenges they run into as well.

16:32Daniel Sebesta:My first question when I saw this piece of news was, has AI now come for voice calls like AI came for our emails a few years ago? And I'm thinking years ago, really, not just with the advent of generative AI as we know it. There have been basic suggested replies on Gmail for longer than that, I believe. The contextual assistance is interesting. I think that goes beyond translation. That might be features that supply information to you to continue the conversation, right? Not just understanding what your partner is saying, but supplying additional information that you might not think of in the moment.

17:24Daniel Sebesta:So I think this is a broader use case than just translation across languages. Same thing with conversation summaries. That sounds like what Zoom and all the likes of that have been doing for a while, something you get sent to your email after the fact. And one thing that really stood out to me was where this is happening. Germany is known for being friendly to any kind of privacy invasion and having AI implanted into your voice conversations on the phone sounds pretty invasive to me. I would have a harder time with that than seeing the same kind of technology work in the background in my email, for instance.

18:21Florian Faes:It's another sort of customer support use case, isn't it? I think customer support on the written side was a very big sort of use case for machine translation early on because of the volumes. So a lot of companies were very keen to, as far as they could, bring automation to translation or chatbots within the chat and things like that. This seems like an extension into a different mode, which is voice and telephony, which I think is, you know, in the context of customer support, very, yeah, again, I mean, there are thousands and hundreds of millions of whatever scale of customer support calls going on each day.

19:03Florian Faes:So big area.

19:05Esther Bond:What's remarkable is the choice of vendor. I mean, they didn't choose anybody that, I mean, like 11 Labs wasn't around in 2022, or it was founded in 2022, right? That's four years ago. And this arguably is one of the biggest live rollouts of like AI interpreting. I mean, if it's Deutsche Telekom, we don't know how broad they're going to roll it out. Maybe it is a pilot and they're announcing it as something that seems much bigger than it actually is in reality. But if they're really planning to roll out the next 12 months for up to 50 languages and they're doing it with 11 labs, okay. Like, you know, there's a few companies that have been doing this type of speech interpreting for a long time.

19:44Esther Bond:So why didn't they get the nod? So yeah, I think the choice of vendor is interesting and the scale of the potential scale of this is interesting. And you're right, Daniel. I mean, like privacy. I mean, is this going to be run on servers in Germany or is this all going to get processed in some like AWS something somewhere, right? I mean, not that AWS doesn't have data centers in Germany, but still.

20:05Daniel Sebesta:And you know, there are always two parties to a call, right? Even if the one party actively chooses to use this, the other party should have the opportunity to consent or not.

20:17Esther Bond:They'll maybe make you jump through hoops. But yeah, I mean, just, I mean, I will be calling that maybe in like six months. I mean, live translation, like, you know, the conversation summaries and all of that, that's sure you're going to get a summary and it may be right or may not be right. It's not going to impact your actual, like the thing you want to achieve with a call. But if I'm on a call and I speak, you know, German and the other person at the other end speaks English and the translation breaks down somewhere. I'm not going to get to the thing that I'm actually trying to achieve with this call.

20:48Esther Bond:Yeah, I think big piece of news, staying with Eleven Labs because they do all kinds of other stuff. They also have, it seems like some type of deal with Lex Friedman, the podcaster, who had the inventor or the, I don't know, the, I guess let's call him inventor of Claudebot. Now it's called Moldbot or, you know, Google it. It's gone through a bunch of variations, some AI framework. Peter Steinberger on his podcast, three hours, and it was dubbed into German by 11 Labs. And Lex Freedom has done this before a couple of times. Now, though, it's much more voice cloned and the quality is pretty good.

21:36Esther Bond:I think it's probably still human in the loop. I listened to it, I think I could probably listen to this for like 30 minutes. Before that, it was more like five minutes. I think now we're at the 20 to 30 minutes before I start zoning out. Yeah. Have you had a chance to look at it?

21:52Florian Faes:I think it's only in German, isn't it? Is it only German? I mean, I opened it. I thought it said it only in German.

21:59Esther Bond:You're right. It's only German. Oh yeah, because of course he's Austrian, right? So it's so funny. Like he speaks English and then he gets kind of voice cloned back into German, which is what we did once, Esther, and then it gave me a Bavarian accent.

22:11Florian Faes:It's long enough, though. I mean, 20 to 30 minutes is not a short amount of time in podcasting world, you know? If somebody stays tuned for that long, it's better than them switching off in the first 20 seconds when they realize they're not going to be able to understand anything.

22:25Daniel Sebesta:I haven't listened to it, but Florian, do you think it's your ever-increasing tolerance to suboptimal audio? No. Or is it actually getting better?

22:36Esther Bond:No, it is getting better. And there is human in the loop. This is not raw AI. The voice, obviously voice cloning is AI and all of that, but I think the translation, there's somebody tweaked this a few bits here and there. All right, look, one more, a couple more out of the world of TAF. No, translation is a feature that one is from Flexport, somebody who's very active, the founder is very active on X and they launched, I mean, Flexport's kind of a logistics platform, trade platform, you know, digitally bringing trade into the digital age and yada yada, that kind of a spiel. But they launched something called Flexboard Translations and we don't have to go into the details here, but the only remarkable thing is that they chose GroK.

23:27Esther Bond:So kind of XAI for the translation bit. Now, maybe this is because the founder is so like, you know, addicted to X because he's posting like once a minute or something, but yeah, it's remarkable that for a product translation, a translation product by a trade facilitation company, they're using now like live Grok translations. So yeah, for, hang on, they're saying we've launched this week by translating the entire platform, including live message translations with support for English, Mandarin, Vietnamese, Thai, Korean, Italian and Spanish. And yeah, Grok's powering this. So maybe, you know, we hope it's not too rude, some of the translations, because Grok's trained on X.

24:12Esther Bond:So what a, you know, source to be trained on.

24:18Florian Faes:And then X said, Germans next, because they did not translate it into German. They said, Germans next, we'll perfect those umlauts soon.

24:27Esther Bond:I'll check if I ever get a Flexport subscription, which I'll probably not. I guess these guys are super busy now with the world of shipping in this array. All right, let's move on to Walmart. Who wants to take this? I think we had a look at this briefly before. I think it's a super interesting story. Their CPO talking about how they moved from what did they call it? The Walmart translation platform. WTP. Yeah, WTP to full AI.

24:56Daniel Sebesta:What was remarkable to me was the 99 % cost reduction. That's like everybody's dream in any kind of localization program management, right? Bringing something in-house, putting technology to the task and achieving a 99 % cost reduction, while at the same time, I believe they said, increasing quality and customer trust. that's a remarkable achievement. I think a lot of our listeners would be interested in the fine print of what exactly it is that they measured and how exactly they reduced the cost and what the customers really thought about it.

25:40Esther Bond:Esther, you had a look at this, right? Can you give us maybe the rundown? So how did they get to this? Is this maybe a little bit more corporate PR or is there, it seems there's some meat to that bone here?

25:52Florian Faes:I think there's a bit of corporate PR as always, but I mean, there's definitely a big initiative that they have rolled out at Walmart. I mean, just some of the figures, I suppose, huge operation. I mean, Walmart itself serving 255 million customers in 18 countries each week, they said. Hugely complex operation. And they said, okay, originally they were spending something like 20 25 million dollars um in well presumably outsourcing costs for translation um that they were translating millions of catalog items each month across 22 languages uh now as daniel pointed out they said it's only one percent of the original cost which is you know 250 000 dollars rather than 25 million dollars um and they are they pointed out that every single translation is done within 20 milliseconds.

26:47Florian Faes:So I mean, for me, yeah, it's a massive sort of cost saving exercise, something that they can obviously point to, to say, we've built this, we're using AI, we're using LLMs. And we know that translation is a pretty obvious use case for AI within businesses. But they do, it's a relatively short podcast interview, but he gives them, And Tim Simmons, the CPO, gives a good amount of detail, I would say, for that in terms of, yeah, I mean, some figures that I just threw out, but a good amount of detail within quite a short space of time. He goes into somewhat a little bit of detail about the Walmart translation platform that they built.

27:31Florian Faes:So he says it uses three layers of orchestration. There are human cultural experts who write rules, they say. Then they point out that there's NMT models for first pass and LLMs for quality review. And then thirdly, what they call data pipes, ensuring signals are fed back in. I mean, I think where there's a bit more PRE is around the kind of trust aspect. I didn't see any metrics around, oh, customer acceptance or, you know, approvals or, yeah, how it's kind of impacting in real life. I think it was around something to do with, I mean, I think just making translation available to more people and more quickly.

28:14Florian Faes:On trust, he was saying AI orchestration actually makes us feel more human because we're translating intent and meaning, not just words. Maybe there's something, some metrics around that, but it seemed to be, you know, a happy sort of secondary impact.

28:32Esther Bond:So you're saying that the CPO of Walmart, an organization that employs like literally half a million people, went on a podcast, spoke about the Walmart translation platform and went as granular as kind of saying that first…

28:46Florian Faes:Localization spend, yeah. And then all the setup.

28:50Esther Bond:Okay, first we have the whole human cultural experts who, interestingly, he says, write the rules. Then the NMT. So that CPO said they use NMT for the first pass and LLMs for quality review. Yeah. So the CPO of Walmart is talking about NMT for first pass and LLMs for quality. I know.

29:06Florian Faes:I just added him on LinkedIn. Well, I tried to connect with him on LinkedIn. So let's see if he'll come on Slater.

29:13Esther Bond:We should totally get that on the podcast. All right. Cool. That's Walmart. Are there any Walmarts in the Czech Republic?

29:19Daniel Sebesta:Not to my knowledge. There are many chains that are bigger than some of the farmers would like, but not a Walmart. But many of them are international, of course. So they probably have some kind of localization program manager somewhere in their structure.

29:36Esther Bond:What about London? Walmarts in London? No Walmarts.

29:39Florian Faes:I don't think so.

29:40Esther Bond:No, right? None here.

29:42Florian Faes:Maybe like in an airport or something just for the, you know, comfort of Americans arriving. But I really would not know. I haven't seen, I don't think I've seen one. It's like when you go to Paris and there's like an M &S or something for Marks and Spencers for the UK arrivals.

29:59Daniel Sebesta:I think if Walmart wanted to be in a country, you would notice it because they work with large scales, right? They might have other brands.

30:07Florian Faes:I don't know if they have sort of like a British equivalent that they don't brand as Walmart or something. I haven't really thought about it.

30:13Esther Bond:Good point, Danny. I think we're actually too small. It's like we also don't have Amazon in Switzerland because I think they don't like the Swiss market. It's just too small. I mean, for Walmart, we had Aldi and Lidl come into Switzerland 20 years ago and gave the local retailers a run for their money. And by the way, amazing localization by the likes of Lidl and Aldi, like totally different stores, totally different kind of ads and all of that compared to Germany. So they did a really good job at localizing their presence here. So now how do you get to$330 million ARR like 11 labs? One good place to start first port of call is to go and get our Growth Hacks for Language Technology Platforms report, which let me spin up the cover page here for those of you who are watching this on YouTube.

31:07Esther Bond:All right. So this is the report, Growth Hacks for Language Technology Platforms, 10 Proven Ways to Grow ARR from Top of Funnel Strength, Sales Execution, Procurement Realities, Partner Selection, Integration Readiness to Enterprise Success. Yeah. Tell us a bit about this, Daniel. So we had a lot of discussions for this. You went into what we call hermit mode, researched, wrote, and then, yeah, we published it about three to four weeks ago. So, yeah, tell us a bit more about this, the goal, the content.

31:42Daniel Sebesta:We pivoted a few times with this report, and it was a challenge to find the niche that would end up being the most helpful for a clearly defined audience. But I think we got this piece right. And we focused on growth hacks, like you said, for LTPs, language technology platforms, language technology platforms of the sort that are external disruptors. They bring innovative tech, they're startups, scale-ups, without, typically without language industry lineage. And the downside with any such external disruption is that you can get blindsided at times. You may not be 100 % familiar with the industry you're trying to disrupt, or you might not even realize that you're trying to disrupt a legacy industry of sorts.

32:41Daniel Sebesta:You bring a fresh perspective, but some of this might be a hindrance. And we noticed with some specific LTPs, and I'm not going to name them, that they were kind of trying to reinvent the wheel in commercializing the technology, in scaling adoption. And we noticed that this applied to a bunch of different elements in how they market their products. This applies to features. Human in the loop, expert in the loop are still workflows that exist in the broader industry. They are not going anywhere fast just because of one tool that's innovative and helpful in the process. This applies to connectors, because if the promise of AI in general is hyperscale and hyper-personalization, as we hear from so many people these days, you can't copy and paste your way to that.

33:53Daniel Sebesta:It's time-consuming, it's risky, it introduces friction. And one of the interviewees that we had for this report literally said that whoever cracks connectors will own the language industry. And when we talk about connectors, it's not just APIs, but deeper, more native connectors, connectors for agentic AI.

34:22Daniel Sebesta:These blind spots also apply to go-to-market strategies, to distribution channels, and that's the reason why we covered partnering with LSI's language services integrators who already own client relationships and in many cases are struggling to add an AI layer of their own to their services. So new tools for established processes that help streamline them. It's a very diverse set of growth hacks because we felt like different startups, different scale-ups needed help in these different areas. Another set of growth hacks centers around content marketing, content marketing in the form of live events, training, certification, reports, and thought leadership in general, which all can add to the top of the funnel for sales conversations, which helps educate buyers about products that are, in many cases, category-defining, products that have not existed before, and there's no precedent for buying them, and content that presents these with authority and a regular cadence to smoothen the path to a sale.

35:58Daniel Sebesta:We also covered understanding procurement, understanding who wants the product in a company, who pays for it, who can block it, and who simply needs a concession. because you want to avoid losing deals for non-product reasons. You want to map budget ownership early. You want to be prepared for the kinds of committees that make the final decision jointly. You want to give your sales team a range of materials to work with for these different entities in an enterprise that contribute to the buying decision. You want to support your champions. You want to remove the blockers. We also covered partnerships, different kinds of partnerships, upstream and downstream, so to speak, partnerships with tech companies, So big tech, hyperscaler ecosystems where language technologies, startups, scale-ups can benefit from deeper integration.

37:13Daniel Sebesta:They can benefit from trust by association, so to speak, easier security reviews, and so on and so on. but also information or intelligence, access to early signals that something is becoming a commodity, early signals that ChatGPT or OpenAI will be pursuing a translation feature or not. And that's the kind of signals that allow you to pivot to a more specialized, more complementary offering than you would perhaps produce otherwise. So being part of marketplaces of these big tech companies and integrating deeply is one good area for growth for language technology platforms. And so are partnerships with go-to-market partners.

38:17Daniel Sebesta:So language services, integrators, LSIs, but also complementary tech companies. You know, a voice company does not necessarily have the IP for translation, doesn't need to reinvent it from scratch. So partnering with complementary tech, partnering with content platforms, content agencies, and that goes beyond translation agencies, LSI's, as good as your tech might be, enterprise clients typically need end-to-end services, not just AI infrastructure. And some of these partners likely have the human component, the process component, the relationship component that helps you fill that gap. Then we covered security because, you know, audited information security controls are something many enterprise buyers will ask you about sooner or later in the process.

39:31Daniel Sebesta:many tech startups have to ask themselves, do we need to build to a SOC 2 standard, for instance? It signals trust. It opens doors, but it's also extremely costly, especially for early stage startups. They need to ask themselves and answer to their clients questions about AI governance. We all know about the black box kind of behavior of LLMs and what happens to data that goes in. Is it ever used for training, etc.? And another important consideration in this industry that relies so heavily on distributed expert in the loop workforce around the world is access governance. And with that, we covered the zero trust policies that some of the TMSs or LTPs in the, I should say, conventional language industry cover.

40:45Daniel Sebesta:and then in addition or on top of partnerships, we covered cases when a simple partnership is not quite enough, when you want to buy rather than partner, when you want to fully integrate, acquire somebody that was previously your partner and the partnership proved to be a productive one, but is simply not enough because as one of our interviewees said, no partner sells your product like you do. And I would add to that no contract and no service level agreement can account for all the twists and uncertainties of the real world while ownership can or is more likely to account for all of that. and with acquisitions you want near-term revenue lift and some kind of strategic structural synergy in the longer run you want to combine companies products in a way where one plus one reliably adds to more than two and where you remove coordination friction where you can offer tighter bundling where you increase the switching cost for your buyers of each of the offerings.

42:13Daniel Sebesta:So, like I said, a very diverse set of growth hacks. It's a bit of a mix and match. I don't think any one company will necessarily need to pay the same amount of attention to all of them, but they can mix and match for the areas where they feel like they have some room for growth.

42:39Esther Bond:All right. Well, that was super, super helpful. And if you want to go get the reports on our website, available now. And we'll probably do some breakout pieces in our analyst desk in the weeks and months to come. But yeah, if you want to get all of them, All of these 10 growth hacks in one nice, what is it, 46-page PDF laid out. Go and get it now. Cool. Well, it was a real pleasure to have you join us today, Daniel. Thanks so much.