Beyond the Tech Stack: Building Data-First Culture at Scale

Summary

Yanolja, Korea's leading travel-tech unicorn, made a C-suite commitment to train more than 1,000 of its roughly 2,000 employees on Databricks across 10 global subsidiaries, creating a cultural shift that drove 80% daily platform adoption.
The assess-and-architect methodology identified skill gaps across four employee personas and introduced an enterprise-wide data capability framework that gave the entire organization a common language for working with data.
The transformation produced measurable business results: a 1.4% data team ratio supporting 1,000 employees, 50% tooling cost savings, and a 30% reduction in manual reporting workloads.

Beyond the Tech Stack: Building Data-First Culture at Scale

A modern data platform without data-ready people is a stranded asset. Yanolja, Korea's leading travel-tech unicorn, learned this when deploying Databricks across 10 global subsidiaries. The breakthrough was not faster infrastructure but C-suite commitment to train more than 1,000 employees out of a workforce of roughly 2,000, creating a cultural shift that changed how leadership, engineering, and business teams collaborated around data.
Learn the assess-and-architect methodology Yanolja used to identify skill gaps across four personas, the enterprise-wide data capability framework that created shared language across the organization, and the measurable platform ROI that followed: 80% daily adoption, 1.4% data team ratio supporting 1,000 employees, 50% tooling cost savings, and 30% reduction in manual reporting. A practical playbook for CHROs, CTOs, and CDOs who understand the real bottleneck is human, not technical.
🤝

Chapters

FAQs

What did Yanolja do to build a data-first culture at scale?

Yanolja secured C-suite commitment to train more than 1,000 of its roughly 2,000 employees on Databricks, using an assess-and-architect methodology to identify skill gaps across four employee personas. The team also introduced an enterprise-wide data capability framework that created a shared language across leadership, engineering, and business teams at all 10 global subsidiaries.

What results did Yanolja achieve from its data culture transformation?

Yanolja reached 80% daily platform adoption and maintained a 1.4% data team ratio while supporting over 1,000 active users. The company also achieved 50% tooling cost savings and a 30% reduction in manual reporting workloads, demonstrating that human enablement drives platform ROI as much as the technology itself.

What is the assess-and-architect methodology for data capability?

The methodology involves first assessing skill gaps across different employee personas—such as engineers, analysts, and business users—then architecting training and enablement programs tailored to each group. Yanolja applied this approach across 10 global subsidiaries to build consistent data literacy, with the capability framework providing a shared vocabulary that allowed all functions to collaborate around data.

Why is upskilling the middle group the key to data platform adoption?

According to this video, employees in the middle tier—not advanced data practitioners but not complete beginners either—represent the largest leverage point for driving adoption across the organization. Upskilling this group, such as training a marketer to write SQL, multiplies the impact of the data platform without requiring a proportional increase in the size of the data team.

Full transcript

[00:08] Hi everyone. Thanks for joining us this afternoon for our next session as part of the data strategy track to explore talent as a platform and how Yanolja built a data culture across its workforce. So, a little bit about our speakers today. We've got Young Ho Song and Soo Young Lee from Yanolja. So,
[00:27] Young Ho Song is the leader of data engineering at Yanolja Next with over 15 years in data data engineering and building the data foundations for global organizations. Soo Young Lee is the head of data at Nolle Universe with over a decade of experience overseeing data engineering
[00:44] and analytics across several multinational organizations. So, a little bit about what you'll be hearing today with the team. So, a modern data platform without data ready people is a stranded asset. Young Ho and Soo Young will share how Yanolja made a C-suite commitment to
[01:01] train more than a thousand of its roughly 2,000 employees on Databricks and the cultural shift that followed. They'll cover the assess and architect method they used to find skill gaps, the capability framework that gave the organization a shared language, and the platform ROI that came with it.
[01:18] Following the session today, we'll also have an opportunity for Q&A. So, when we do get to the Q&A, um we'll have a microphone just on the left-hand side here. Um so, if there are any questions uh from the audience at the end of the session, we ask that you please just line up um at the microphone and we will
[01:33] step through those questions uh in all the in the time left that we have. Uh but please welcome to the stage Young Ho and Soo Young.
[01:59] Hi. Uh a little of us. Okay, hi. Wow, thank you.
[02:14] Sorry. Hello everyone. Thank you for being here. Our talk today is called uh beyond the tech stack uh architecting data first culture SK. It started with a simple belief. Uh putting one platform in place would be enough.
[02:29] But it wasn't. And that is what today is really about. Before you go further, uh let me show you who you are.
[03:11] We are short. Three promises, one group. Uh to keep them, we knew we needed a strong tech stack. So we built one. But one year in, we learned the tech stack was only the half of the answer.
[03:28] Uh that is what beyond the tech stack means. Yanolja is Korea's number one techno uh travel tech company. We've raised over a billion dollars and our app serves over 5 million monthly users.
[03:46] Three times our nearest competitor in Korea. But we are not just on app. We operate across 200 plus countries with global offices on several continents.
[04:04] We cover the whole journey of travel. It begins inside the hotel where our software runs the property. Those rooms uh then need to be sold. So, we handle distribution at scale.
[04:20] On top, a smart layer uh turns that into better decisions. Uh and at the very end is the traveler. So, one traveler can search, book, check in, and get recommendations, and every single step happens inside our group.
[04:35] The business is fully connected. So, uh that's the full value chain we built from property operations all the way to the customer. But, here's the catch. It all started with one question.
[04:52] How do we truly become one travel solution? Because as we grew through so many mergers and acquisitions, our data became completely fragmented. And that created four real roadblocks.
[05:08] First, inconsistency. Data in so many places and formats made it a constant struggle to keep accurate. And that eroded trust in the numbers behind the our decisions. Second, uh low discoverability.
[05:25] Our analysts spent more time hunting the data than analyzing it. Finding the right data felt like a needles uh in a haystack. Third, expensive analysis. Running questions across all these
[05:41] separately systems was slow and costly just to get basic insights. And pros, complexity. The The architecture had grown too complex
[05:58] to scale. So, our engineers were stuck uh fixing infrastructures instead of creating value. So, it came down to one practical questions. How do we collect data across our 10 plus global subsidiaries to actually
[06:15] drive business impact? And we gave ourselves an answer. The same one uh many companies gives. We believe that if we just built the perfect platform, the innovation
[06:30] would follow on its own. But the reality, the car wasn't moving. And that led us to a discovery we never expected. That great platform alone wasn't enough. Something was still missing.
[06:48] So, let me take you through that journey. The first thing we built was the car itself. And honestly, the picture in our heads was very simple.
[07:03] Uh put good good data on top of good platform, and the car called good synergy would just drive itself. So, why wasn't the car moving? This is what you were actually looking
[07:19] at. Every company run on a completely different stack. Different BI tools, different analytic engines, different clouds. The data looked different, and the way to reach it was different everywhere.
[07:38] And the real pain wasn't just different tools. Every single time one company wanted data to another, even once, we had to open our account, set up access, run secret check, and build a brand new connections from
[07:53] scratch. Every time. With many companies, they worked multiplied fast. And each one took weeks or months. None of it created new value. It was just the cost of staying
[08:09] connected. These were our data silos made real. So, we set three simple rules. One, one consistent way to share, no multiple ways,
[08:25] one shared channel for every exchange. Two, safe and easy to find, governed by default, so anyone who needs data can find it themselves and use it safely. And three, the most important for me,
[08:42] share the work, don't carry it alone. If every company had to build and run this on its own, the cost would be impossible. We wanted our companies spending time working together,
[08:58] not just keeping systems alive. These three principles pointed us to one answer, Data Bricks, a shared, common foundation none of us had to build or run alone.
[09:18] So, the technology question was settled, but settled platform isn't an adopted one. And getting every company actually on to it together was the next challenge.
[09:36] And that turned out to be the hard part. So, before any real work, we done three steps online to bring everyone on the together. First, one shared picture. Because everyone was thinking from the frame of their own existing system.
[09:52] So, even simple conversation got lost. When we set the platform, each company imagined something different. So, we started by agreeing on the same word and the same mental image. Second, a plan for each company.
[10:09] Each company had its own dedicated pipeline, and they need to see exactly how those would carry over. So, we surveyed each company, what environment they were on, what
[10:26] their main data sources and pipelines looked like, and there was a separate plan for each each case. Third, let people try it with their own hands. When you introduce the all of these through documents and presentations
[10:43] online, it just didn't land. It stayed abstract. So, we built small working samples to show what it actually looked like in practice. And let people use the key features themselves.
[10:59] That's what really raised their understanding. That got us a long way, but the limits became clear, too. And the biggest one wasn't technical. Online, every decisions and every round of feedback and simply took too long.
[11:16] Across time zones and busy schedules, even small decisions dragged on for days. So, we did something different. We flew everyone to Dubai for 1 week, 1 room, 1 goal.
[11:32] Decisions that used to take a days started happening in minutes. We discussed each company's real data flow together and on the last day we finally connected every company's data to each other. It was just a one focused topic
[11:49] but work that used to take a months is we finished in a week. And everyone felt the changes with their own hands. That one week made the next 12 months possible.
[12:08] The platform was ready. Now we had to put the data into it. We made the pri- priority simple. The data tied to Google level KPIs goes first because that's the fastest path to real business value. Now this is what the work used to look
[12:25] like. Every month getting one number meant a manual download, rolling it up in Excel, and then chasing down format and value errors by hands. The same painful loop over and over and over.
[12:41] A full migration would take months, even years per company. And we couldn't wait that long. So our principle uh for the short term was simple. Connect first, integrate later.
[12:56] Almost any platform uh can export its data. So instead of migrating everything, we took each company's raw data, their business ready KPI grade tables, and built automated pipeline on top of it.
[13:11] Now the full migration was enough uh to get the number flowing right away. And as we built this together uh our engineers and analysts, for the first time, started the end, started
[13:27] looking in the same direction. And here's the result. Data that used to live separately inside each company now flows automatically into one shared view.
[13:44] This is one constant way to share made real. Data sharing let us connect every company the same way regardless of what environment it's run on. And it's safe and easy to find, too.
[14:00] One secure place to find everything instead of opening firewalls wide between companies the way we used to. So, it's not just less manual work, it's far more secure.
[14:15] And remember the four problems we started with? This is where two of them are quietly went away. The inconsistency because everyone now read from one constant source. And the complexity
[14:30] because there's one shared way to connect instead of a different setup per company. We'll bring all of our company's data together into one unified data platform. There are many different way to get
[14:46] there. But the point isn't just to connect. It's to link our systems in a way our engineers can actually work on together. So, instead of each company solving it in each own one-off way,
[15:02] we are adopting methods that work the same in any company. That's the third rule, share the work, don't carry it alone, becoming a real architecture. Engineers from different companies building side by side on common ground.
[15:22] Okay, the body of the car was the fuel tank was full. We had the platform and we had the data flowing into it. But the car still had one part missing. The one part that actually make it move.
[15:39] What the missing part was and how we built it, Soo Young has been right at the center of that work. So, I'll hand it over to her. Thank you, Emma.
[15:56] Well, and nice to meet you. And I start my turn. Uh so, as Young Woo just shared, we built a fantastic car. A powerful unified data platform, data bricks. We expected to start it moving right
[16:14] away, but it didn't. The car was already, but it was not going anywhere. So, I still remember the first week after launching the data bricks,
[16:30] Young Woo and I was so excited. We wait for everyone logging and start doing everything like input the secret statement and make that super dramatically, but nothing happened.
[16:48] Totally complete silence. Oh, yeah. So, people still using their Excel or their Google Sheet. So, but it looks me uh they just looks me so
[17:04] scared of new system data bricks. Is there any data bricks step because, to be honest, my close friend in core work send me, "It's look pretty ugly."
[17:20] Yeah, they said "It looks geeky." So, I don't want you see I like Tableau like that. So, this is a critical critical moment us. We were missing
[17:36] the most important part, the driver. A car cannot move without skilled and good drivers. So, we shifted our focus from architecting the platform to
[17:53] architecting our people. It's most important part. To create good drivers, we first need to understand deeply why they were not ready.
[18:09] When we looked closely, we found three main problems. First, structure. Our company, I'm from the university in Yanolja Group. It also merged of three big OTA platform
[18:25] in Korea, South Korea, in the end of 2024. So, more 50% of our company people expected a centralized data team. So, they expected
[18:42] and our mighty our team, data team, do all work about data. But, as you know, Databricks um require the people do work with data directly themselves.
[18:58] This was a huge mismatch. And second, as I mentioned before, we as we merged company. So, we use same word, but the mean is very
[19:13] very different. For example, revenue and MAU. So, revenue for flight industry revenue mean is margin. But for accommodation that means
[19:30] total booking amount because Yanolja uh pre-purchased inventory. And MAU except for Yanolja platform, well, we we merged three big OTA platforms, Yanolja,
[19:48] Interpark, and Triple. Except for Yanolja uh other platform required uh register before buying something. But Yanolja does not require
[20:04] registration. So, without logging the buyer a product in our platform. So, that means people think monthly active user mean is very different.
[20:20] So, this create a lot of confusion in meeting. We arguing and uh always persuade others. So, it made collaboration very hard. And third, skills.
[20:36] We had a massive gap in data and AI literacy. Some people were data expert, but many others are just starting, beginner. So, we realized we had four different
[20:52] persona. Skeptics, observers, and challengers, and champions. As we had a wide range of skill, we need a plan for everyone.
[21:11] To solve these three problems, we designed three specific solution. We call them our prescription. For the structure mismatch, well, we change our team color
[21:27] from centralized data team to hybrid model. So, still our central team uh maintain pipeline and manage shared data. What but business unit also analyze
[21:44] their own domain data by themselves. And second, to fix misaligned metrics, we publish the canonical metrics first. And then, it designed for searchable and
[22:01] reusable. Easily cloned. So, actually, I hope that no more confusion, no more arguing in meeting. And last, to close skill gap, we didn't do one
[22:17] size for one size fits all. We designed personalized training to fit each persona as I mentioned before. So, but here's the key question. How did we
[22:34] actually close this massive skill gap? We did not guess. Yeah, we did not guess. And we did not do alone. We partner with Databricks and used their talent transformation program.
[22:52] With their expert super super expert consulting, we built a systematic approach to guide our transformation journey to our people in company. First, we started with a clear diagnosis.
[23:09] We ran deeply assessment across company to measure our exact starting point. No cold starting. Yeah, we try to find starting point. And based on those data,
[23:25] we ran survey and asking and have interview. Based on those of data, we derived strategy action item. And then put everything into the action item. Eventually,
[23:40] we drive we drive we delivered customized training tailored to each persona. Through this program, we successfully transformed our entire organization.
[24:01] So, what did you all this work actually give us? Let me show you the result in three numbers. First, adoption. 80% of all team in our our company use Databricks daily.
[24:17] And for individual, our company-wide adoption is at 65%. Second, efficiency. Our central data team make up only 1.4%
[24:33] of our total head count. Just 1.4% I Yeah, I think um generally, average point uh data team ratio is around over 5% to 7%
[24:50] in globally. I I I saw that. Um we we have exact nine data engineers and nine data analysts, we now cover 14 different business units. Supporting over 1,000 employees.
[25:08] And the best part, simple undefined. You know, I hate it. Undefined and repeatedly request dropped almost zero. So, that's why we
[25:24] I handle my our team data team with just 18 18 people. Yeah. And third, business impact. We saved nearly 50% on tooling cost.
[25:40] Before launching Databricks, we run Airflow, Presto, Druid, um Jupiter, too much. Yeah. So, but in this in this at the time, it
[25:56] is so hard to track the exact cost for each task. And now, we can. We can do. And now they I'm focused on the policy and governance, how much we
[26:12] can spend money for our each task. And our team now spend 30% less time making manual report. They are no longer just pulling data. They are finding insights.
[26:32] This is the result of upskilling our people. A small team create a massive impact. When people see those numbers, they ask me one question. How does a team just nine engineers and nine analysts support to
[26:50] 1,000 employees? This secret is so simple. It was our training strategy. Remember the four persona I mentioned earlier? If you look at our organization,
[27:06] you have the top talent at level three or level four. And the majority of our people at level one and two. Instead of just focusing on the experts, we put our main whole main focus on the
[27:24] middle group. Our level one and level two business user. We gave them practical and hands-on skill relevant their daily task. So, we have them step up and become
[27:41] challengers. Now, they can pull their own data and they can handle everything in their briefs. This saved our team so much time. So,
[27:59] I I'd like to share some uh story. I remember one performance marketer. She's typical of servers. So, uh I'm sorry. She always wait uh days for a simple
[28:16] report. Before our training, she every Monday morning is a nightmare for her. She had down She had to download and three different heavy CSV file
[28:34] and do complex VLOOKUP and pray, "Oh, don't crash it. Don't crash it." So, anybody that come here from marketing field? Do you agree with that? Yeah? So,
[28:50] when we first approached her, we She was terrified when she listened the word of SQL. So, she said, "Oh, I don't have a license. And I don't understand SQL. I don't want it." And at the time we sat
[29:07] down with her and we showed her how to write a simple SQL sentence, select from blah blah blah in Databricks. So, actually
[29:23] I will never forget the look on her face when she checked the data in Databricks in just 5 seconds. So, she said, "Oh, too fast. I don't have to waste time."
[29:40] And yeah. Okay. Imagine over 100 people in our universe. It's a real world. Like this. So, that's is how we scaled.
[29:57] This is how our small team create massive scalability. So, last page. Let me bring it all together. Data Databricks gave us the car.
[30:15] Fantastic car. A unified platform connecting our over 10 global subsidiaries. Then, we trained the drivers. Over 1,000 people. All leveled up and ready to
[30:31] ready to use data. And this combination gave us speed. Over 100 terabyte of data are now queried every single day from business user, not data team.
[30:50] When we start this journey, people thought data was just name of department. So, but today in Yanolja Group, data is everyone's second language. And
[31:06] maybe English is third language like me. So, the transformation did not happen overnight, but it was worth yeah, every single struggle. I want to remember one thing
[31:23] from our story. Our platform give us potential. And people give us power. But together, yeah, they give us speed. Thank you.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.