Why Cost Per Word in Translation is Outdated With GetGloby CEO Diego Antista

SlatorPod #192 - GetGloby CEO Diego Antista on Cost Per Word in Translation

Diego Antista, Co-Founder and CEO of GetGloby, joins SlatorPod to talk about the complexity of translating and transcreating digital ad campaigns.

Diego shares a pivotal moment when he discovered that the cost of translating campaigns for multicultural audiences was a significant barrier for advertisers, leading to the creation of GetGloby with fellow former Googler, Juan Fusoni.

Diego shares insights into the advertising industry, such as the role of agencies in global campaigns and the metrics advertisers prioritize to measure campaign success.

Diego sheds light on the challenges of localizing keywords, particularly in languages with character limitations, and discusses how GetGloby addresses the need for consistent brand voice across various languages.

Subscribe on YoutubeApple PodcastsSpotifyGoogle Podcasts, and elsewhere

Diego discusses the platform’s current target segment, primarily large clients and global agencies, and outlines their plans to launch an online product for smaller advertisers. 

Diego explores the tech stack behind GetGloby, focusing on the use of AI, machine translation, and the development of their proprietary model called Jaga. He also highlights the importance of incorporating native reviewers for certain clients, especially in the luxury brand sector.

Looking ahead, Diego shares how clients will soon be able to activate a new campaign for a flat fee of 10 dollars per month, with a credit-based model replacing the traditional cost-per-word approach.

Transcript

Florian: Today on the podcast, we have Diego Antista, the Co-founder and CEO of GetGloby. GetGloby is an AI-powered platform that enables businesses to transcreate their ad campaigns into 100 plus languages. So you had an interesting career, more than a decade with Google before you co-founded GetGloby. Tell us a bit more about your professional background, what part of Google were you working at, and, yeah, just a bit of the backstory here?

Diego: I started working at Google in 2007 in the Latin America operations, opening business for agencies, for holding agencies, basically, in Argentina, then Mexico, Chile, Colombia, and Peru. And in 2014, I had the opportunity to open the team of US multicultural. The US multicultural team was dedicated to activating campaigns in any language in the US that wasn’t English. In the early days, it was primarily Spanish, but over the years, more languages were activated. After that, I worked on the global team in Europe and, believe it or not, in Russia, at Google Russia for a while. Then I returned to New York before deciding to create GetGloby with another former Googler and my friend, Juan Fusoni.

Florian: You mentioned activating language. What does that mean in the context of Google on the ad side?

Diego: Every time a campaign aims to address a new language, and it’s not always about activating a new country. In many situations, for example, in the US, Canada, or Europe, there are many new minorities living in those countries. Brands see an opportunity to activate a language when running global campaigns targeting different regions and countries. When I say “activate,” it means figuring out how to transcreate the original campaign into the new region and language.

Florian: Were you the liaison between Google and big accounts struggling with this process?

Diego: I started working in the US multicultural team in 2013 when Google was not making much money. The US multicultural market is a $7 billion market, and Google was getting a 30% cut of every media budget. I discovered that clients were reluctant to pay for translations. To overcome this roadblock, I offered translations for certain big clients for free. This led to significant growth, about 100 times, in sales within four years. Advertisers realized the process was easier, not just in terms of cost, but also in managing the complexities of transcreation or translation, especially in digital advertising with multiple stakeholders like media agencies, translation agencies, and reviewers.

Florian: So, that’s a significant pain point you identified. Is that what triggered you to start GetGloby? Can you share the origin story and the trajectory of the company since then?

Diego: During my time at Google, I saw an opportunity to solve the issue for US multicultural campaigns, primarily focused on US Hispanic campaigns. I thought it should be relatively easy to address this. I discussed it with my co-founder, Juan, who was the director at Google in Latin America for travel companies like American Airlines. He faced similar challenges with global campaigns, especially when advertisers didn’t want to pay for translations of multiple languages. We realized the need was not limited to US multicultural campaigns, so we decided to create GetGloby for more than 130 languages.

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: Who is your current key target segment for users on the platform?

Diego: Currently, we focus on large clients. We’re set to launch an online product in the second week of December. The development and AI teams say it’ll be ready by December 1st, but I don’t want to push too hard, so let’s say December 15th. The product is designed for the long tail, catering to small advertisers looking to activate campaigns locally. However, the platform’s architecture was built from scratch for large global clients and agencies. We started there because addressing the needs of major clients with top global brands in terms of localization would allow us to do a great job for the long tail. Along the way, we addressed challenges with AI translation, leading to new features like brand voice, native reviewers, and integration with platforms beyond Google.

Florian: Can you tell us more about the platform? If I’m a big client with, say, 25 or 30 languages, how does the platform work? How much is AI, and where does human involvement come in?

Diego: Today, we offer clients the option of full AI translation. When they create a new project with GetGloby, they can connect their Google account, and all existing campaigns in the master account are linked. They can select a campaign, choose the languages and regions they want to activate, and our AI machine translation, called Jaga, immediately considers project-specific information and pulls data from various sources. In less than 30 minutes, the campaign is ready for multiple languages and regions. Clients can decide if they want a native reviewer for spelling or other details. About 10% of our clients opt for native reviewers, especially those focused on branding, like Alexander McQueen.

Florian: Interesting. For high-end brands where perfection is crucial, they might opt for native reviewers. I’d like to come back to the tech stack, but first, let’s dive into the world of Google ads. Who is responsible for running regional or global Google ad campaigns at large enterprises or e-commerce companies? Who’s the key contact?

Diego: For large corporate America companies, often the responsibility lies with agencies that have a special division for global accounts. Positions like Global Performance Lead or Global Search Lead handle global search campaigns for these agencies, which are part of the six holding agencies like WPP and Publicis. These agencies work for big corporations with a global presence. Then there’s another category of companies that have a presence in one country but operate fully online, such as in Asia or Eastern Europe. They have dedicated teams, usually in content or localization, facing resource challenges while dealing with the complexities of global campaigns.

Florian: Now, you mentioned results. What are some key metrics these companies look for to know they’re getting results? What are the top three metrics?

Diego: For global accounts already running campaigns in major countries, they focus on similar KPIs like CTR (Click Through Rate) or quality score for important keywords when activated in different countries. However, for fully online companies, the cost per acquisition (CPA) becomes critical. They operate globally, and if they find a better CPA in a specific country, they quickly move the budget there. These companies are highly results-oriented, with centralized teams making swift decisions.

Florian: And a detailed question: all this data comes from Google Ads, right? You’re connecting to Google Ads, or are you integrating with other platforms like analytics?

Diego: Currently, the primary pain point for advertisers is translating campaigns. We’re solving this for Google Ads, but we’ve also incorporated other platforms. Today, we can handle translations for various platforms, not just Google Ads. We initially prioritized connecting with Google, but we realized the importance of platforms like Meta, especially for type two clients, as Meta often offers better CPA in many global regions compared to Google.

Florian: With Google’s algorithms changing frequently, how do you stay relevant and maintain quality?

Diego: I believe Google makes small changes to its algorithms, especially in Google Search, as they don’t want to disrupt a multi-billion-dollar revenue stream. We stay updated with these changes and use AI to enhance the transcreation process for Google Ads campaigns. The basics, such as finding traffic with keywords and maintaining concept consistency during translation, remain the same. We incorporate over 30 rules used by Google Ads specialists for single-language campaigns, making it easy for our AI to handle translations for more than 130 languages, simplifying the process for everyone involved.

Florian: You mentioned keywords, so just tell us a bit more about keywords. Obviously super essential, like one of the secrets of e-commerce, of everything, of lead generation for the world. Yeah, as you said, easy. Or already, I guess for me, already difficult in one language. I can only imagine how difficult it is in 30 languages. And then you’re solving keywords for not just one company, but hundreds of companies at scale on a tech platform. Where do you even begin to look at that challenge and solve it?

Diego: There’s a lot of available information today about keywords and verticals. So you have some help with Google Traffic estimator and then some other tools that are out there. They are not all of them from Google that we incorporated in that technology at the time to be doing the translation. If you think about specifically for those countries where Google is number one on search, that is basically the whole world. Except in certain countries in Asia where Facebook, Meta is actually the number one browser, believe it or not. So it’s not exactly the same nowadays when we are talking about the world; we need to have a specific mindset about which country or region we are talking, because it’s not all happening exactly in the same way like in the US, Europe, or in the Americas. It’s very different in some of the regions.

But going back to Google, in those specific countries where Google is the number one, over 90% of the search, you will find out that it’s certain keywords that are very important by vertical. So you have a vertical, then you have a subvertical, and then those keywords must be included because those are actually the broad terms that are going to be bringing you traffic. In general, specifically for very developed countries like in the US, many of the brands, they are not a fan of including broad terms because that is bringing a lot of traffic, but it’s affecting the CPA. But when you are activating a campaign in a region that you don’t know how your product is going to work or you don’t have much experience, we strongly recommend including those specific broad terms because it’s going to be bringing you traffic, and you’re going to be bidding on those special categories in that specific region or country. You will see traffic on your landing page, and then with all the rest of the tools, you can optimize. You cannot be optimizing over something that you don’t have traffic. The key thing about GetGloby is trying to figure out in the process of localization how to bring you traffic for those specific languages, for those specific regions or countries.

SlatorCon London 2024 | £ 980

SlatorCon London 2024 | £ 980

A rich 1-day conference which brings together 140+ industry leaders views and thriving language technologies.

Buy Tickets

Register Now

Florian: Now, you mentioned some of these luxury brands, and I’m sure a lot of other brands have very strict brand guidelines, brand voice, making sure that this is consistent across at least like the top 10 or top 15 languages. How do you inject that in the system? How do you ensure that this is enabled for them?

Diego: Yeah, that is one of the things we figured out during the last year. When we created the platform, we said, “Okay, now it’s ready. We connected the Google ads so the brands can do it in a snap.” We figured out that in some tests, we were doing it with one of the great Holy agencies in London with Dentsu. The global team was doing some tests with what we call the elevator, what they were testing for free with brands like Gucci, working with Gucci, even for free, and those testing the campaign, let me tell you guys, was very challenging. The team, it’s amazing. They know their product and they know their audience so hard. I mean, in each one of the regions, when we were receiving the brief, and I can’t imagine how it was working before when they were doing the localizations with a human being because it was so much information. I said, we need to do something. That’s why we created last year what we call the brand voice. The brand voice is where the brand can be choosing what they think is the ideal translation. Let’s say they have a new product that needs to be launched, I don’t know, in Canada. And so what they say, “Hey, this is exactly how I would love to communicate in Canada in English,” okay? And it could be anything. It shouldn’t be a Google ad campaign; it could be any translation that they love. We incorporate in the platform. And then they go through a very easy form where they can just be pulling all the information they know about the audience, they know about how they would like to communicate their brand in that specific region. The more, the merrier. Okay, so some other translations and some translations they didn’t like, and that is basically training Jaga. Jaga, as much information is getting. What we have at Jaga is like a little lion that is going to be trying to collect all the information. I mean, that is how it works on the platform, and then it’s helping a lot at the time to improve the quality. We figured out that in the last two quarters, it was a dramatic increase after we incorporated the brand voice process.

Florian: Interesting, because you’re one of the most recent startups in the kind of automated localization space, and almost like you’ve been around just a little longer than ChatGPT has been a thing, right, and all these capabilities that these large language models bring. So for you, it seems like this kind of rewording and like training on the fly and all of these kind of things that historically were extremely advanced or even impossible are kind of something very like a natural capability. Tell us a bit more how when you started, well, okay, ChatGPT and large language models weren’t around. I mean, large languages were, but ChatGPT wasn’t. But you started and then a year ago this hits. What did you think about it and how do you think, well, this is going to help us make more advanced features for the platform? Yeah, just tell us about that thinking and then transition into maybe explain a bit more about Jaga, what it is and how you build it and how you customize it?

Diego: I remember when we started working over GetGloby like it’s going to like take three years. Okay. I remember one of the first rounds we did with the investor, angel Investors. We would talk, hey, we are testing with OpenAI. We are one of the first beta testers about OpenAI. It was ChatGPT and the number one version that no one was even talking three years ago and an investor saying what? I mean, what are you talking about? Even some VCs during those early conversations they were saying, what? And it’s fun, but then on the run we did in 2022, they were saying, oh, you are testing with OpenAI, that’s good. I mean, okay, now you know what we are talking. Yes, I need to thank Juan, My co-founder. He will start talking to me about GPT and AI, specific new technologies more than three years ago when we started discussing about this project. How we were working and at the very beginning we were incorporating, for example, Google Translate and some other translators in the process and then we were trying to fix the translations on top of that with code. Okay. The depth in putting the hands on that. Then we figured out that it was not good. The quality we were getting specifically for translation for digital ads was very bad with any of the platforms. Okay, besides half of my heart, 50% belongs to Google. But even Azure or any other of the platform, even DeepL, that is really good, it was not good. So then we figured out that we were trying to improve how we were doing the transcreation process and that’s why we decided to create Jaga. Jaga is actually a new machine learning model that is getting all the information from different platforms. What is the secret sauce on Jaga is depending on the prompt, I mean, depending on the request. The specific request is which platform is going to be using, how it’s going to be combining and how it’s going to be fixing internally. For example, we figured out that if you are trying to create a new ad. You have an ad in English and then you need to translate to German, okay, could be very challenging. Why? Usually you have the same character restriction, length. I mean, the length, it should be 90 characters. You have an ad in English for 90 characters. When you translate it, German, Spanish, I mean, those languages is going to be an issue because you’re going to be over 90 characters and it means that you need to redo it, redo it again. So for those specific cases, we are using certain technology. If we are doing for German, that it will be absolutely different the platform, if we are doing for Russian or if we are doing for Spanish, et cetera, et cetera. That is the way that Jaga works. It has internally certain rules that are getting better day by day, learning in the process how the things are working, if they are doing a transcreation process, they are activating the campaign. The second part of what is great about Jaga is getting information from those campaigns connected. So if one campaign was working better than the other one, said why this is working better, what is it actually affecting? And that is where the AI team, they’re trying to figure out which of the rules on that specific campaign was making the things better than the other one.

Florian: Very interesting, so it’s like platform dependent, performance dependent. And then, I mean, the language kind of conversion part, translation, transcreation part is kind of the given. It’s at the bottom, it’s the bottom, and then you’re adding all these other layers on top.

Diego: Then it’s going back, and it’s going back and it’s going back. But in many of the cases, we are still learning a lot from AI. So one of the key challenges we have at the beginning was the platform, when you are not giving a lot of context, it’s having what we call hallucinations. I remember that one of our key clients, that is LATAM Airlines, they were trying to activate a campaign for flights from Europe to Latin America. And I remember an ad was saying, let’s get the chance to note that South America or something like that. And then when it was translated to German, I’m not speaking in German because I’m not good at that, he was basically saying in English, Latin America is the future of the human being. What is going on here? And it was like the AI team saying, you are not creating enough information. So the platform is just dropping and some options. Of course, that is creating internally some process that is going to be okay, this ad, it’s not good, it’s going out. But yeah, it’s still a lot of learning process. Again, when you are thinking about OpenAI and the rest of the technologies, Jasper, etc., etc., that is like the mother technology and we are connecting with that. But then it’s a lot of work how you are combining and how you are training and giving the information to get the best. Because if not, it’s not only about just dropping and getting okay. The quality is still, you need to give a lot of context to any of the technology, Jaga or OpenAI, to get a better result.

Florian: In terms of the language, the target languages, I guess it’s just the top Spanish, Chinese, Portuguese, German, like the European languages. But you mentioned you do 100, but is there a drop off where it gets really complicated once you have languages that are of lesser diffusion, that a lot less data is around, etc.?

Diego: One of the key languages I remember, Polish is a very challenging language in Europe, okay. In terms of doing the translations and etc., etc., and trying to address, because it was always very complicated. And when you have, I think if you allow me, I like numbers, but if you think about in this way, it’s a matter of the size of a population you have and the complexity of the language. Okay, so if you have a very specific language in certain parts of the world where there’s not many people actually speaking or using the language, it’s going to be getting very complicated because then you have, for example, India. In India, it’s more than 40 different languages, not dialects, 40 different languages. You think you have one of these like Tamil, one of the most important. And I said, okay, so Tamil, actually, how many people in the world besides people who speak Tamil, they know about Tamil. And we figure out that it’s not so complicated because the size of the number of people who are actually talking that language is making richer and richer every day. So, yes, it’s depending on the number of people and the region.

Florian: You did mention that you’re launching that kind of more self-serve version in December. Anything else around the corner that you can disclose? Product roadmap? What can we expect in next few months, 2024?

Diego: Next year we’re going to have new tools that are going to be allowing to create social content in more than 130 languages. Every time that you are having a campaign, I don’t know, let’s say for e-commerce or direct to consumer, today you don’t only need to activate a campaign on Google or Meta, then you need to be creating a lot of content around the product, okay? And doing that today, it sounds very easy with AI, but incorporating the brand voice, incorporating the performance on the campaigns on the back and then finally incorporating the additional information is going to be making that content in more than 130 languages much more better. That’s what we’re going to have for next year. Besides that, by let’s say half of December, any client around the world, they could be activating a new campaign at Google or any other platform for a flat fee of $10 per month. So we are launching a very, very low cost activation because we want more people testing the tool. Of course, it’s going to have a cost for us because this is increasing all the data that we are managing. But as time that this is getting bigger, we have more campaigns, Jaga is getting better in terms of conversions, knowing more about different languages and different regions, different verticals, what should be better, which keywords, etc.

LocJobs.com 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.

LocJobs.com I Recruit Talent. Find Jobs

Florian: Very interesting. I think OpenAI and some of these APIs aren’t super cheap, but let’s see if 10 dollars covers this for you guys per month. I’m sure you have some fair use clause in this or something like that where if they start doing like tons and tons.

Diego: It’s going to be a restricted amount of credits. We are not doing a cost per word. We want to be out of that model, and we think that for digital advertising and for the new days, cost per word is outdated. So it’s going to be credits, and those credits are going to be linked with words, but also with other things like new activations, new campaigns, etc.

Florian: I love when you say cost per word is outdated. So you’ve been in this kind of, I guess, localization industry now for a couple of years. What was your journey like coming from, in a sense, the outside being aware of this big problem that there is, but then, oh, there’s this huge kind of industry with like 10,000 companies that are kind of involved in this translation problem. Can you tell us a bit more about what was the feeling coming into this and discovering all these kind of nooks and crannies of this industry?

Diego: I think it was great. I figured out that there are very old companies that are doing a great job, especially when you’re thinking about not just digital advertising or marketing assets, but also for books. I don’t think it’s going to take a long time to translate a book by AI. I don’t see that happening today, of course. There are many things that need to get better, but one of the key things I realized when I started working on this is that the cost per word is not the proper way to activate a campaign in any digital campaign. Because, for example, you could have a very small client with only a $1,000 budget for selling, let’s say, a cable on Amazon. They might need thousands of keywords for certain specific characteristics of that cable. And that small client will be strongly affected if you are using a cost per word model. On the other hand, an automaker may not need as many keywords; it’s very specific, like “Toyota Corolla.” So you have multimillion-dollar companies using fewer keywords and fewer ads, while a small client may need more. It’s not a connection in the number of keywords and the marketing asset; it’s a disconnection. So, at the time we were activating and creating GetGloby, we were offering our clients a flat fee because if they are using AI, let’s do it with a flat fee. Additionally, there’s an extra cost if they use more. That’s the way to unlock opportunities, like opening new markets, regions, and languages.