Building a Global Infrastructure

Building a Global Infrastructure
August 30, 2017
Today's episode features an eye-opening discussion with Tableau Software's Daniel Sullivan, who is Director of Global Content Readiness at Tableau Software. Learn more about the needs, opportunities and pitfalls localization professionals must face to understand the growing impact of more targeted, relevant and insightful data. 
Download episode
Show transcript

Transcript

Speaker Transcript
Michael I am Michael Stevens, and today on Globally Speaking, we’ve started with a special guest as our announcer. Would you like to introduce yourself?
Aidyn Sure. Hi, I’m Aidyn Stevens.
Michael Aidyn Stevens is daughter of myself, Michael Stevens. And Aidyn, what do you know about starting a global program, getting something off the ground internationally at a company?
Aidyn I know very little to none.
Michael Very little to none. So today’s conversation could benefit you a great deal, huh?
Aidyn Yep.
Michael Yep. Well, hopefully it will benefit you too. And we will let our guest introduce himself.
Daniel My name is Daniel Sullivan. I’m the director of localization at Tableau Software. I’ve been here for a little over three years. I came here at a time when Tableau was really just starting to make their big push globally and started out being in charge of building out an infrastructure to help them do that, and then build out a team to help execute on that. Then, since then my role has expanded a bit very recently to include things like website content, infrastructure, search engine optimization, and even systems and analytics and things like that. So, it’s broadening as the program and the team matures, the responsibility levels that we’re covering is broadening. The real core of the thing that falls under my team is providing languages services for Tableau and helping us localize the best, highest quality, most efficiently deliverable content to help us go global.
Renato In how many markets are you present, Daniel, these days?
Daniel Well, I mean, we’re all over the world. I mean, we’ve got a presence just pretty much everywhere. Currently, our product and our website really focuses around eight languages. That includes English, Japanese, Chinese, and Korean, Brazilian Portuguese, Spanish, German and French. But, really, that is not the limit or the extent of Tableau’s sales efforts around the world. We really are marketing all round the world.
Renato And do you have a sense of what markets of those eight that you address are more sensitive to the localization?
Daniel That’s something that is a little bit more recent because, like I said, I’ve been here for just a little over three years, and when I say a little over three years I mean just a couple of months over three years, so it’s not that long. When I got here the localization infrastructure, and when I say that I don’t just mean our ability to provide translated content, I mean just the infrastructure as a whole and that includes even our insights into our markets and how well our content is performing.
Michael Can you give the categories those fall into for people who may not have read the Common Sense Advisory…
Michael You’ve matured three levels, you would say. So, you’re getting close to that, like, optimizer role.
Daniel Yes, exactly. I mean, when we first started this all off it was literally… and everyone starts here, even small groups within large enterprises, as I’ve experience with, where when you’re first making your steps in this you have no infrastructure and you probably don’t even have anyone on your team who knows how to do this. Someone usually gets sort of the localization component tacked onto their regular job description, and that’s where Tableau was. They were copying and pasting content from their website, putting it in spreadsheets or Word docs and then sending it by email. And so the first thing we had to do was, you know, address that and put some sort of technology around that, not just to automate the process but really to control and stabilize it in other ways. So, we made really early investments in TMS and things like that.

None of this necessarily has anything to do with the gathering of data and the analysis of data, but everything we built from the very beginning was always built with the principle of data first. When we invested in the TMS one of the questions as we were evaluating all the systems was “and will I have access to that data?” Because, I knew just from my experience of working with various TMS’s that while they all provide some various level of reporting—and some of them were good—it was never going to equate what I could do with Tableau, not just my ability to do drag and drop analysis and ask questions and just start building a dashboard how I wanted to see it, but also, then, the ability to connect that data with other repositories of data, where information and content that was very relevant to what I was doing was going to be stored.

For example, our translation management system is really great at capturing data about the translation workflow itself, where content is in that process, how much it’s costing us, how are we delivering on that content, are we consistently late or consistently early or on time, that kind of stuff. But, project management and overall budget might be captured in a completely different system, and those are just two examples. And our infrastructure right now is 20+ pieces of technology for our localization platform.

So, as we’ve added components to this, it’s always been with the assumption and always with the question, if we invest in this technology, how are we going to be able to interconnect the various databases that we’re creating in this process and do really accurate reporting.

This is the long way of saying, because, Renato, I think you were asking me specifically questions like what kind of data are you capturing, what kind of insight is this giving you to your markets, etc. Those are only things we’re really starting to do recently, just because we’ve been setting up the infrastructure that’s allowed us to do that.

But, now, we’ve got really good insight on this. It’s getting to be more and more scientific as we go; lots of really interesting questions; lots of really interesting insights are coming up. And I feel literally like I’ve been here for three years but I feel like this is the beginning of the next story.
Renato One of the interesting points, Daniel, and it’s very interesting because when we talk about maturity, many people associate maturity with technology and with the size of the organization. You can look at Walmart, which is one of the biggest companies in the world, and they’re very immature when it comes to globalization; it has nothing to do with size. You can have a very small company that is mature because they understand the role of localization in their processes; and that’s the journey that you have described. You’ve gone from very manual, you might have been technologically immature but from a business perspective, the management of Tableau understood that localization was important, they gave you that task, and that is the first step in developing the maturity of the organization, is recognizing that you have a business problem, or a business opportunity—this put it this way—to develop your business by localizing a product.

What I’m curious about is what kind of data you’ve collected that is surprising, or unique, or something that is unusual because you mention things like you actually know the cost per unit that you are engaging there; but you know it as a relationship. Is it a profitable word, is it a money-losing word; what kinds of insights are you getting?
Daniel Okay. Well, first of all I want to answer the first part, it wasn’t a question, you were just making a very interesting statement, though, about how maturity is not tied, necessarily, back to the size or even the age of the company, and you mentioned Walmart. I’ve been at other enterprises where very mature, huge global presence, but within the infrastructure you will have a variety of levels of maturity for localization.

I feel very lucky because when I did come to Tableau I was not presented, immediately, with questions like “you need to prove to us that it’s worth it for us to invest in this content and worth it for us to invest in these countries, why are we doing this?” That was never a problem. They were already ready to support me in any way that I needed. And so, I feel very, very lucky that I was brought in, I was hired and I had the support of people like our CMO, Elissa Fink, and my vice president, Wade Tibke, in marketing operations. They saw the value of this immediately, and so I did not need to spend a lot of time trying to prove that.

And so, we’ve been able to reinvest that time, then in doing other things, and that’s allowed us to also focus on building out an infrastructure first, and now we’re starting to get to the point where the data that all this is capturing is actually finally lowing us to ask these challenging questions and finally start to get answers for them.

Some of the things that I have found the most interesting is when I was trying to see how Tableau was doing globally with sales and where that aligned with the languages that we were currently invested in, and that has been some of the most interesting things to see. But, it’s never been a question that I could answer with just the simple data we were capturing in our various systems. I have actually had to go out and create some custom databases about country and language use in various countries and official languages around the world and things like that, and try to find really creative ways to reconcile that back to the revenue we were generating around the world.
Michael So, you’re using internal data sources of which you guys are pooling, but you’re going out to public data as well to get that.
Daniel Right. But, it starts to get really fuzzy, and I’m starting to see that you cannot just make these, I guess, data-driven decisions. It’s more data-inspired decisions because while you might have the data that’s telling you one thing, you also need to be able to contextualize some of that data. You also need to be able to know that, “yes, the data is telling me this but I know why, so I’m going to reinterpret those numbers that it’s giving me.”
Michael Would it be fair to say it helps you increase your odds for success?
Daniel Yes, but here’s a perfect example where things can get a little fuzzy. If you are looking at, say, website traffic, there are a number of ways that you can sort of look at that to try to determine what language people are consuming and maybe what language they actually speak.

For example, Germans. Our traffic from Germany, if I look at geo-traffic or something like that, and I’m looking at something like their browser setting, their browser language setting, a lot of Germans never change their browser setting from the native default English; and so browser setting is not really a good indication of what the person visiting your site, what language they might speak, what is their first language.

But, even the content that they’re consuming, maybe they’re in Germany, they’ve got their browser set to English, but they’re consuming German content; does that mean they are native German speakers, can I attribute any of that traffic back to our investment in German? That’s where things start to get really fuzzy and a little bit challenging. But, it’s having access to all this data and at least knowing that it’s fuzzy; that’s where the real human decision-making is going to start coming into play. That’s where the person’s knowledge about countries and languages, and things like that, that’s where that comes into play and sort of compensates what the data is telling you. And you really can’t have one without the other and then expect to make really big decisions.

So, it’s not just having access to this data, it’s also being able to understand how to interpret it and actually use it.
Renato It’s looking at data as a decision support system and not as a decision-making system.
Daniel Exactly, exactly.
Michael Oh, so you can’t just rely on it to make decisions for you? Oh! Well, that’s good. So, one of the things I took away from you summarizing the environment at Tableau is you have multiple data sources, that means multiple tools that you’re working with. So, you didn’t really go out there and try to find one tool to rule them all.
Daniel No. No, no. And that was, again, that was very much a guiding light when I started this process because I did come here with something of a bias for a TMS, and I just mentioned the TMS because that’s the very first big purchase we made. But, I knew when I was investing in it, it had all these other little bells and whistles, but I knew I was never going to really rely on them because what I could do was going to be a lot better.
Michael And so then, there is one piece of the system that is consistent and that is Tableau Software, that’s your product. There may be some people listening who aren’t familiar with what Tableau does. Can you give us this?
Daniel It’s a really intuitive tool that is designed to help people see and understand data. It’s very much about the citizen data scientist and, basically, empowering not just data analysts; I mean, there’s always going to be a place for data analysts; and there’s always going to be things that the citizen data analyst cannot do. But, what Tableau does is it enables just literally anyone to connect to any kind of data, and that’s just not one data source but it’s also you can connect to Excel here and data extract over here in server, and then maybe MySQL data server there, just bringing this data together in a way that helps you, then, start asking questions and getting in the flow of questioning that data. What the tool does, it just makes it very simple to just adjust your questions as you go.

So, you might just start out with a very basic question, like, “how much revenue are we getting in Japan?” That’s a very easy process of dropping out country and then pulling out revenue, whatever that is. But then you might start asking other questions like “what about revenue from Eastern Japan within this sector of industry?” Whatever.
Michael Tying it into your location, that’s one option.
Daniel And it allows you to start just asking these questions and playing around with the data and starting to get answers very, very quickly.
Michael So, you are visualizing data. Tableau has been a leader in the Big Data conversation, that was kind of a hot word years ago. How would you define Big Data?
Daniel I don’t know if I’ve ever really liked the word Big Data. When it first came out, what they were talking about was just the fact that there is all this data, but now it’s sort of become this discursive word that is just more representing the usage of data in general. It’s kind of become a parody too because even we’ve parodied it.

A few years ago we did an April Fool’s about Medium Data. We did a whole campaign about Medium Data. And I remember one of my jaunts, walking through an airport and one of the best sellers in the airport was Small Data, or something like that. Like, it’s not just Big Data; it’s just Big Data is synonymous with access to all this stuff; it’s not necessarily the size of the tables and the millions of rows that you do have because, I think, Big Data can just start with a spreadsheet. That’s where we started.

We had nothing. We had no infrastructure, and we did have these multiple spreadsheets that were capturing localization-related data in various forms. So, one of the first things I did was I looked at all that and said “okay, this is what they’ve been capturing; this is where it is; we’re going to pull this into a single source; we’re going to structure this and we’re going to start using this for reporting and the thing we’re going to start monitoring with this single spreadsheet is all the projects, the dates…” because we had nothing, no infrastructure at all.

So, we just built something that was structured. We started capturing this data. We started standardizing the way we were reporting and capturing this, and then we threw Tableau on it. Now, I’m able to finally, for the first time, reconcile costs down to the penny month to month. I’m able to now use Tableau to see where all these projects are because our translation management system was Outlook. But, now, I can start seeing this.

And, as we started adding tools onto this, eventually that spreadsheet was retired, and now it’s in a structured database, and the database is being populated by the tools, and we’ve got processes whereby we are ETLing some of this data, meaning we’re pulling it out and loading it into something like Amazon Red Shift or other repositories.

And, as we’ve gone, as we add more tools this just gets more complicated and, again, like I said earlier, it’s always been with the data first principle. So, as we add these it’s always “what is the key between these two systems; what can I use as a key between these various databases?”
Renato But this is a very good point for you to tell us, what is it? What are the core metrics that everyone in localization should pay attention to?
Daniel Wonderful question. I don’t have the absolute answer for that. I’m actually looking at one of my visualizations I have for… I actually threw all this in the database, this exact answer. And I’m looking at it right now, and I see I’ve got 148 different rows of data. And each of those represents a single metric that I want to be monitoring or I already am monitoring. So, I’m also tracking this and “have we answered this question or are we still working on it?”

The first place that I started was cost—“how much are we spending”—and then project management. The second place I went was our website. Tableau on top of our website data and the first questions I had to start answering were “how much of this website is localized and what’s available in each language?” Because, when I got here, and this is again part of the story coming to Tableau and its journey into becoming a global company. When I got here, the way that they were deciding on what to translate because it was very early in the process was basically whatever the regions needed.

So, if we published a white paper in English and someone in Germany thought “oh, I need that, can you translate that for me?” Then they would translate the page and the white paper into German. Then, maybe two weeks later the French would catch wind of this. They’d say “oh, hey, can I get that too?” Yes, sure, we can do that. And then they’d run it for French.

What happened was over the course of a year or two is that the website was just partially localized. But, the biggest question I had was “how much of our site is localized?” because all the regions saw it as a problem, and it was a problem.

So, we put Tableau on top of the website data, and we just started answering this question. And that was our primary data source and those were the primary questions we were asked for the first year was “how much of it is localized; how many white papers did we publish in German, in Q4 of 2015?”, for example, because we need that information for our QBR, for the regional QBR. Things like that.

So, we started at that level. Now, where we’re going, however, is “okay, I want to be able to drill down to a single piece of content in a very particular language”. So, this white paper in Japanese, and I want to know all sorts of information about that one piece of content. What is the organic traffic that’s coming to that page, what is the bounce rate; how many localization bugs have been filed against that content; how many of them were valid; how many of them were resolved; how much, how many times have we run that piece of content this year, and what is the ROI on it with respect to the amount of traffic?

So, if you’ve got a page that has 10,000 organic views in a month and you’ve only spent $600 to translate that page and you’ve only run it three times in the year, you can get down to the point where you can say something like “every single unique visit through organic traffic costs 0.000237 cents. You can get down to that now.
Renato So, let’s stop here. What is a profitable type of content, and what is a waste of money, in your practice? This is valuable information for me as a marketing guy.
Daniel I feel like we’re getting to a point where we are able to start answering those questions now. There is something that, actually, Michael had mentioned to me once that was really a great piece of inspiration for this because he put a nice acronym around it “Bobwow”—best of best and worst of worst”—and knowing what is the content that’s really performing; so, what is performing content.

That’s another really great thing because that definition is going to change from content type to content type. And it’s also going to… performant from one group is going to be defined in one way versus another.

In marketing we would define performant content by its ability to lead generation, its low bounce rate. It’s something that people are consuming that’s helping them make a decision to buy Tableau.
Renato What about the negatives, what is the thing that you found out that was a waste of your money and you stopped localizing because it didn’t really help? Was there anything like that?
Daniel I think most of the content for us that has not really what I would define as being really performant has mainly been because there wasn’t a marketing effort around it. And I think this was something that we may have had very early on when I was describing that the question was how much of the website is localized? This is what was on the top of everyone’s mind, basically, the state, health and overall coverage that our website provided was very weak.

So, our mission for about a year was to just get as much coverage and shore up things as much as possible so that we had a much broader coverage and a better base-line to start with. I think in the process, in trying to fill out as much of the website with language content as possible, the thing that we weren’t doing was then wrapping marketing efforts around a lot of this. And I think that’s where the real difference happens because…

And I can see this. I know that just making a product available in a language, just making a bunch of content available in a language, that does not mean that sales go up as a result, at all. You still need to put a massive marketing and sales effort around it.
Michael So, that’s what I’m hearing from you, that I hadn’t thought through, is that the effort to create equal websites, globally, you weren’t really saying “wow, we’re going to see a big trend upwards in sales because we’re now adding more German pages” or French pages, or whatever it may be. What you’re saying is “I now have a baseline of performance that’s equal to the English that our marketing team, I’m able to support them, so they have better efforts globally.”
Daniel Right. And this is not that we were doing anything wrong initially. We really did not have a good baseline for measuring a lot of this. It’s only because we actually have started to make this content available that you can start seeing who’s engaging with it. So, you do kind of have to start it at that point of “okay, here’s a baseline; now you can start adjusting and measuring.”
Michael And this goes a bit against the flow of what we hear from some companies who just say “we let the GOs decide. We let them request and we let them decide what to do with their budget.” This says that is one option but there needs to be a baseline that I can judge from that is equal so we can make good global marketing decisions.
Daniel And we do do this; we do have, people have the ability to request content from us. And we’ve been very good, and people have been very good; they don’t ask for stuff that they don’t need. We just do not see that. We have set it up so that when a request for a white paper does come to us, it’s already coming with that conversation with the regional folks having happened.

So, when they come to us and say “we need this for these Asian markets and none other” it’s because the conversation with the regions has happened. So, we do know that that content is going to get used.
Michael Do you find that you’re reporting into marketing, you’re reporting up there to help them improve their marketing efforts, do people on your team use data differently and Tableau, for that matter, to help inform you; are they looking at different questions? Does the use case vary by role in an organization?
Daniel Yeah. And that’s part of what I mean in answer to Renato’s question earlier about what is performant content? And how many situations that can depend on. For someone in-region, a really performant white paper might be something that was, for whatever reason, coupled with an event that they ran and, as a result, everyone was reading that white paper, and it had some impact on some training that they were doing, whatever.

So, for them, what’s performant might be very different from what I’m monitoring because I don’t run campaigns, and so I’m not really implementing this content; other people are doing that. So, that’s why for me something like organic traffic, that is something that I can have some control over, and that is sort of a really good measure from my standpoint since I am not running campaigns on content; how well that content is performing.

That’s part of the reason why SEO is trying to come underneath me as well and be one of our responsibilities because we’re actually producing this content and since we’re not running campaigns with it, our metric is organic traffic.
Renato I only have one more question. I’m always chasing new stories, and you mentioned that the way that you sell Tableau is through stories; that’s the one area where the clients come and find more information about Tableau. It doesn’t need to be a localization story, but tell us a good story about Big Data because I’m tired of telling people the story about beer and diapers!
Daniel Okay, I do have a story. It involves our amazing editor for Japanese, Akipo, in our Japan office and the way she was using data. So, I mentioned to you earlier, we had a data-first principle for a lot of the systems that we’re rolling out, but there have been some things that I knew eventually I was going to invest in a tool that allowed me to capture some type of data, but I wasn’t there yet and I needed something that was going to tide me over till we had a really robust tool for that.

So, what we did was we partnered with our vendor and set up a SharePoint site, and set up various SharePoint lists within there to capture data; some of it was just capturing data. And one of the things that we did was we knew that eventually we were going to incorporate TAUS into our translation workflow, and that we would want to use TAUS to do linguistic QA.
Renato So, you’re talking about the DQF.
Daniel Exactly. But, a year and a half ago, we did not have the infrastructure to start using DQF, and I had other priorities rather than DQF. But, I wanted to start capturing that data because I did have a very—one my most important questions is “what is the quality level of our content that we’re getting back from translation?” So, what we did was, we were already using SharePoint to help our in-country reviewers monitor their projects and record their time spent on projects and how many words they were working on, etc. All we did was we just tacked onto that five extra columns for capturing data on quality, and we made sure that was aligned with one of the DQF quality models that I was very likely going to invest in or use once I shifted over to TAUS proper.

So, for a year and a half, we had this system whereby we were capturing data across these five verticals, and so now I don’t have TAUS implemented yet, so I’m not able to have this really robust and automated and mature process for linguistic QA proper, whereby you have editors going through content and the only thing they’re doing is scoring; they’re not editing, they’re not changing anything, they’re just scoring the content.

Since I didn’t have that then we just used SharePoint in the meantime and all of a sudden I had all this data on quality for every single project that went through our workflow. And what we noticed during the editorial process, and actually not we but our Japanese editor noticed, was she felt that there was starting to be a quality issue in Japanese. And all she had to do was take Tableau, because she had this question, “I think there’s a quality problem in Japanese”.

She’s not the only one doing editing on the Japanese content. There are two other editors she’s working with. And so what she did was put Tableau on top of this data we were capturing in SharePoint, and she arranged this across these five verticals and she arranged it by project and date, and she assigned a color to the score level. And we’re doing it on a scale of 0-4. So, zero was obviously bright red and 4 was bright green, and you’d have variations of those colors in between.

What she saw for the month of August, this is last year, was a field of red. So, she was able to take that and go back to her vendor and say “look, we’ve got a problem in Japanese; I don’t know what’s going on but we’ve seen a dip in the quality and now we’re spending more time editing than we are actually… it’s becoming more of a process for us to finish our projects because we’re editing more, and we need to do something about this.”

Well, it turned out sure enough, the vendor was on-boarding at that time, a new translation vendor for Japanese, so we had that. But, now, we were monitoring and watching it and so in addition to this, in addition to isolating that there was a problem and using data to show that there was, we tacked onto this a scheduled coaching meeting, once a month, with the editorial team and the entire team of Japanese linguistics that were working on our account.

What happened was within, I’d say about two months, that sea of red started to turn very green and within two more months we had nothing but a sea of green. Like, it was every now and then there was a little drop of red here and there which there might have been some problem on one of these verticals, but it was a field of green. And it was a beautiful thing.

And it happened again, several months ago, it happened again, but it was basically the same process. We could see it immediately; we could identify it immediately; we addressed it immediately, and it turned around very quickly.
Renato This is a great story because you transform something that usually takes too long in the cycle and you start being proactive instead of reactive. That’s fantastic.
Michael So, what advice would you give the person who’s listened to this and been inspired; how do they get started?
Daniel You can start with a spreadsheet. That’s what we did. We started with a spreadsheet, and it went from there. You do need to start with a plan and you do need to do it with the principle of data-first. You do have to go into every single component, everything that you’re adding is “okay, this is going to solve this automation problem; am I going to be able to report on that as well?” So, asking that.

Then, you do not start with these really advanced questions like reconciling localization cost back to each and every organic traffic that you get to a certain piece of content. You don’t start there. You start with the money, what’s available, how much are you spending? Start monitoring those things very early on and then you can add all the more advanced questions as you go because they are going to come by nature anyway.

The more you stay engaged in the data, too; this is another thing I’ve noticed. The more I’ve worked with the data, the more I’ve played with the data, the more engaged I am with it, it’s become one of the driving factors for me now, as opposed to not just automating things but then how well are we doing in showing that performance and using data to do that.

End of conversation

Daniel Sullivan

Daniel Sullivan is Tableau Software’s Director of Global Content Readiness. He has over 15 years’ experience in the translation and localization industry, including 8+ years architecting enterprise localization platforms to improve delivery time, quality and cost. Prior to Tableau, he worked in localization at Wizards of the Coast and Amazon.

In a previous life, Daniel was an academic at Stanford University, where he earned an MA and completed candidacy requirements for a PhD in Japanese historical fiction and historiography, before shifting to work in enterprise localization.

Stay Tuned

Subscribe to receive notifications about new episodes

Play episode
0:00
0:00