Transforming Field Service at Scale: TK Elevator's Data, AI, and Operations Architecture

Summary

TK Elevator, which maintains 1.4 million elevators and escalators serving 1.5 billion people daily through 25,000 technicians worldwide, has partnered with Databricks for a decade to transform service delivery from manual operations to GenAI-driven field insights.
A Medallion architecture and Unity Catalog consolidated fragmented data across 400 branches from IoT telemetry, enterprise systems, and unstructured documents into an AI-ready foundation powering Digital Operations Centers that deliver real-time actionable guidance to technicians.
TK Elevator applied GenAI to service brief generation, documentation automation, and continuous learning from technician interactions, while maintaining human-in-the-loop validation and enterprise governance across decentralized global operations.

Transforming Field Service at Scale: TK Elevator's Data, AI, and Operations Architecture

TK Elevator operates 1.4 million elevators and escalators across 400 branches serving 1.5 billion people daily. A decade-long journey with Databricks has transformed service delivery from manual operations to GenAI-driven insights. By consolidating fragmented data through a Medallion architecture and Unity Catalog, TK Elevator created an AI-ready foundation connecting IoT telemetry, enterprise systems, and unstructured documents. Digital Operations Centers now turn real-time data into actionable guidance for 25,000 service technicians worldwide, reducing callbacks and increasing equipment uptime.
this video reveals how to bridge hardware and software worlds through standardized KPIs, semantic layers, and agentic AI. Learn how TK Elevator applied GenAI to service briefs, documentation automation, and continuous learning from technician interactions, all while maintaining human-in-the-loop validation and enterprise governance across decentralized operations.
🤝

Chapters

FAQs

What is TK Elevator's Digital Operations Center and how does it work?

TK Elevator's Digital Operations Center is a centralized capability that transforms real-time IoT and operational data into actionable guidance delivered to service technicians in the field. Built on a Databricks Medallion architecture with Unity Catalog, it processes data from 1.4 million elevators and escalators across 400 branches to reduce equipment callbacks and increase uptime.

How does TK Elevator use GenAI for field service technicians?

TK Elevator applies GenAI to automatically generate service briefs and documentation for technicians before they arrive on site, drawing on IoT telemetry, enterprise system records, and unstructured historical documents. The system includes continuous learning from technician interactions, with human-in-the-loop validation to ensure accuracy and safety in a mission-critical environment.

What is the AOX platform and why is it relevant to AI at TK Elevator?

AOX refers to TK Elevator's digital-native, AI-ready elevator platform that generates richer embedded software data and connectivity compared to older elevator models. This video describes how AOX elevators provide better data signals for the AI and IoT systems underpinning the Digital Operations Centers and predictive maintenance capabilities.

How did TK Elevator address data fragmentation across 400 branches?

TK Elevator faced fragmented data across hundreds of decentralized branches, each with different systems and local practices. They resolved this by building a standardized Medallion architecture on Databricks with Unity Catalog, establishing unified KPIs and a semantic layer that aggregated branch-level data into a consistent foundation for analytics and agentic AI.

Full transcript

[00:09] Thanks a lot for inviting us. Thanks a lot for hosting us and it was a pleasure to listen to you how you were able to combine the digital data AI world with a elevator speech which is callbacks, uptime. And that's basically what Christian and me have been doing since a couple of years. Christian even
[00:25] longer than me at TK Elevator. So, we are working hard on understanding the latest and greatest what's happening here and in other parts of the world. And all of you who just listened to Ali this morning how technology is advancing
[00:40] how we are seeing AI ready data, agentic AI ready data and not only 10% improvements in performance, but step change performance in improvement. Well, in my previous company we would call that disruptive innovation. It's just crazy how the
[00:56] speed of this technology is advancing and what we can do with it. And bridging that that development with the other part of our job which is making sure elevators work 24/7. Elevators are and escalators and you see
[01:12] some outside. All these elevators and escalators are safety critical and mission critical infrastructure. So, bridging these two elements of the real hardware world with embedded software and with wonderful service
[01:28] technicians to the world of data, AI, large language models and how these two can combine to bring us to a state where we provide even better service, that's what we do. And that's our homework every day, every night and actually most
[01:43] of the time it's really fun. Just a thought experiment up front to get you all awake in the afternoon here. Maybe some people jet lagged. Some of you probably come from Europe. So, plus minus 9 hours. Some of you from Asia, some of you from
[01:59] North America, some of you from South America. And when you came here, right, some of you probably took a plane, went to the airport. Just imagine this travel without elevators,
[02:15] without escalators, without passenger boarding bridges. That would have been a cumbersome, right? So, that's what we do. We provide these elevators, escalators, and the service for these elevators, escalators, passenger boarding bridges
[02:30] 24/7. And we provide that for 1.5 billion people every day. 1.5 billion people every day we are moving with our elevators, escalators, and we provide the service.
[02:48] And actually wonderful service in California because Twinkie's taking care of that. And we have 1.4 million units under maintenance under service. So, if I go back here. 1.5 million, 1.4 million elevators and
[03:04] escalators under service. And the real business is done in the branches. 400 branches globally with 25,000 service technicians. So, 50% of my colleagues, 50% of my colleagues are service technicians. They're well trained, they're highly motivated to keep elevators, escalators
[03:21] running 24/7. So, if you look at this, this is a global business model and very decentered business model. So, whatever we do on the digital team, we need to service this business model. And we call it rigidity at the core.
[03:37] So, one global company, but flexibility at the edge. So, we need to make sure that our branches, which are best athletes, get the support from us. And there, Christian will talk about it, Trusted Data Foundation comes into the play. Our journey with Databricks since
[03:54] 2018-2019, and then we bring this into field operations. Field operations mean how we do service for with 25,000 service technicians 24/7. So, it's at scale. It's not a proof of concept, a test.
[04:10] So, it's really at scale. And that's a wonderful challenge to have. We want to have make sure that elevators, escalators, moving walks, passenger boarding bridges have enough time that makes our customers and passengers
[04:26] happy. And by the way, for those of you who don't know escalators and elevators, I'm I'm talking about elevators now. Elevators are the most safe model of transportation. The safest model of transportation. So, don't worry, elevators are safe.
[04:43] And what we do, we want to make sure that we even have more uptime, so that the elevators and escalators work all the time. And for that, we're using modern technology. And we have gone this journey since roughly 10 years. And that's a part of our journey. You
[04:59] see on the left-hand side already 2016. We inaugurated the MAX Box, a partnership with Microsoft back then. MAX Box is our connectivity device, connecting the elevator to the cloud. You can also call it the gateway. And that gateway sends the data to the cloud. Then 2018-2020,
[05:17] we moved to real large-scale data. And we needed a good solution, and that's where we went to Databricks, because back then Databricks was the best company. Christian picked it to help us on large-scale data. How do we work with large-scale IoT data? We moved to
[05:33] enterprise data, common KPI definition. Christian will talk about it. That leads to the fact that we can then take the IoT data and enterprise data, ERP, CRM, FSM, all these wonderful tools, bring the data together with a
[05:48] common definition and around the company at scale, but also decentrally. And then, just recently with the emergence of more AI tools, we've been able to bring a genetic AI um to first branches globally on a
[06:05] ready-to-scale platform. And some of you might have seen it online or offline at the Microsoft booth at the Hannover Fair, which is also a strong partner. And then, we've been able to do this jointly with Databricks last couple of months and actually years.
[06:25] Our technology, our solution to make this happen, to bridge the hardware world, not pressing anything here. The hardware world, connecting the hardware world, the better software world, the service technicians with the data and technology world, we're having three pillars of
[06:40] technology. First one is connectivity, connecting our elevators to the cloud. The second part you see here, secure cloud and data platform, security from the get-go, data protection from the get-go, and a genetic AI for the innovation part. So, these are the three pillars
[06:57] we've been building over the last couple of years. How does it look? Well, look at it, you see a wonderful elevator at the left-hand side. You see our MAX box that's connecting the controller, getting the controller data,
[07:12] taking it via the MAX box, doing some data intelligence, moving it to the cloud, to the Databricks platform, and then on the right-hand side you see we have predictive insights for service, for management, for sales.
[07:28] And these elements we have been driving since 2016. Nowadays, if you order an elevator, an escalator, moving walkway, passenger boarding bridge, in this case an elevator from us, and if you need an elevator at home, please go to TWINKEY. Just raise your hand.
[07:44] She can sell you elevators and service them even better. Sell elevators and do the best service to them. Keep them up all the time. So, basically Careful, it's moving all over. So, basically what we have is uh digital native and AI-ready elevator, called
[08:01] AOX. It's digital native, so it's connected to the cloud directly. AI-ready because we have also XPU or small GPU on board with camera uh in areas where it's allowed. And
[08:16] you see also And you see also other upsides here, which is 45% lower energy consumption, reduction in the body carbon use, so it's even very sustainable. Digital native, AI-ready, and we have been developing this at
[08:32] least on the tech side with very, very strong partners. Now, how do we bridge this elevator world with the data and AI world? Before Christian jumps into the platform part of the common data platform and data bricks collaboration,
[08:49] there's one element I wanted to share with you, which is our digital operation center. The digital operation centers are the transformation engine of our service transformation. The digital operation centers, we have eight Now now by now we have nine of
[09:05] them globally. The first one we had in Atlanta. So, they are a global network of digital operation centers. And they turn data, real data, IoT data, live data, plus other data,
[09:21] feedback from the customers, into insights. How do we do this? We have a couple of experts in the digital operation center. For sure, we have data experts, data engineers, data scientists. But then we add two more competencies, which is an operations expert, really
[09:37] understanding how processes are run in the country of the duration center because the country in every country is doing the service operations a bit different. North America a bit different to France, to Germany, to UK. So, we have the different duration centers.
[09:54] We have a data expert. We have a operations expert. And for sure we have elevator experts. These three colleagues sitting together looking at hundreds of thousands of connected units and turning the data points into insights. So, what then happens?
[10:10] They take what they see. Maybe they see in San Francisco Fourth Street building X 25th floor, there is a an element where we see that door is not moving as smooth as it should because it's closing slower or
[10:25] faster. And that's what we see in the data. Then what we do is we create a duration center ticket. We say DOC ticket. This ticket is not written in Christian or Matthias language. It's written in service technician language. So, the service technician next time he or she goes
[10:41] there knows what to do. So, we turn data into insights but in the language of the service technicians. And with this we have been able to really substantially reduce callbacks last year and even more this year. These duration centers built on the com
[10:58] data platform and that's I think the transfer to you, Christian. Thank you. So, Matthias talked a lot about the the digital transformation and also the digital products that we have at TKE, but what does it mean from a data perspective? There's one thing to mention, TKE as a company grew a lot
[11:16] through acquisitions which left us with a very very fragmented system landscape. What you see on the slide here is a a lot of data data from from different systems over 50 or 100 systems
[11:32] worldwide on the one hand side and on the other side we have IoT telemetry data that we collect through MAX. Yeah, and the challenge here was to bring this all together and as the company was steered like very heterogeneously in branches, um we
[11:48] originally did not have one central plane how we could connect the local steering with the uh with the management of the company. Through the introduction of our common data platform in a in another project that we also introduced called NEXUS, we
[12:05] brought this all together. So, we brought all of the data in one place. First use case was really automating reporting in the company that made our CFO happy. And this is also the basis for more data use cases and we'll see it
[12:21] also for AI use cases. So, and we we had this topic here Ali also this morning talked a lot about ontologies and semantics. We on the one hand side took IoT data, but that alone does not create value. So, we brought
[12:36] that together with the enterprise data that we saw before. But what's still missing is the meaning, the semantics that describes how TK's business is working. That's another crucial element that's important for reporting purposes, but guess what? Also
[12:52] for agents. So, we call that then AI ready data because without that there's no real intelligence. So, but how did we do that? We leveraged the classic medallion architecture as probably many of you did. Um interesting
[13:07] part here is that we not only used that structure data as we brought in also unstructured data like documents, we said let's not reinvent the wheel here. We have something that already works. So, we also channel documents through the same pipeline here. Um and also then
[13:24] in in the silver layer essentially have markdown of documents that we can afterwards work with in the same way as we could work with structured data. Of course, through Unity catalog, we had lineage, which brings trust and is
[13:40] important so that people actually believe what what they see there. This was all before some of the announcements came, so we had to build some of the stuff ourselves. But we already looked at like that. So, on the one hand side, we have a data
[13:56] platform, which Databricks always was, but on the on top of that, the semantic layer, we leveraged metric views already for that to describe the data better, and in the end to on top of that define business KPIs that can be used locally in the
[14:12] branches. We heard it, so our business is actually happening in the more than 400 branches, but also that that we have the same view if we look at the top management perspective. And that's what's depicted here. So, we really now align global
[14:30] steering. So, if we make strategic decisions, that we that we look at this on the on the global level in the same way as in the branches. In the past, this often deviated. So, branches had a
[14:45] completely different definition of certain certain data points like what's a cancellation, what's an active unit. So, there were different definitions in the company, and a lot of the reporting was manually. So, people would extract data from one system, key it into
[15:01] another system, and that was the basis for for for OKRs or QBR meetings. And then guess what happens if you make decisions based on that data, it does not reach the branches. The the decisions don't make sense. So, by putting this all together, we really
[15:17] solved the issue of bringing the fragmented data into one place, and connected how different people or different levels in the company look at the data, which makes the steering way better, and it also brings
[15:33] IoT data together with enterprise data. So, and that now helps us to really reach our branches. It follows the platform approach, not only from an infrastructure perspective, but also from a data modeling perspective. So, we
[15:48] can build it once, and then scale it everywhere to reach the uh more than 400 branches, and Matthias mentioned that there are over 200 and uh the over 25,000 service technicians. So, that's the let's say that the data story
[16:03] behind the digital story that that helped us to to make all of that possible. Um but now let's take a step back and look at what what is the challenge in our industry that we have today.
[16:18] We We heard already elevators are mission-critical infrastructure. There is a high expectation from customers that elevators are working. We see it here in San Francisco, if you go outside of this building, many of the other buildings here have many more than 10 stories. If elevators wouldn't work,
[16:35] we all had a problem, yeah. We would walk a lot the stairs. So, high demand to keep the elevators up and running. How do you do that best? Um by adding more context to the people that maintain these elevators. So, that's one thing. And on the other side,
[16:51] there's another challenge that we have, and that is that um a lot of the people that used to maintain our elevators are leaving the industry, the so-called baby boomers. And new people that are coming into the industry expect way better tooling, the
[17:07] so-called digital natives. So, they they expect more support. We see it here through either tablets or whatever. So, these are the things that we're now trying to address. Uh but to let's let's let's look at how you can depict the the different areas errors of
[17:24] service. So, we think of this as before we started our digital journey, we were in the manual error. Everything was dependent on individuals. So, they needed to be well trained, but if they left the company, that know-how was gone.
[17:39] Um with digital, we already improved this, yeah? So, we made already data-driven. We saw that virtual operation centers on the picture before that gave guidance to service technicians on how to fix up elevators if if there's a shutdown, so what to do, and also on the next maintenance visit,
[17:56] what to focus on. Which improved the situation and made it less dependent on the individual. Now, with what's available to us since about two or three years, um we have GenAI reasoning capabilities that and a
[18:13] GenTech AI field that helps us to go to the next level. Which means we can also look at unstructured information through document and challenges, and there's one other element that is now also available to us, and that's everything around
[18:29] conversational agents. And with more context, we can now improve the service experience even further. But how did we now use this? So, we really looked at the overall service journey. So, that's a com- com-
[18:45] complex slide here cuz we of course did not start from scratch three years ago. We looked at our overall journey and looked at what did we already cover? Uh and what this is showing is really So, it's it's what happens before a technician is on site, what happens when
[19:02] a technician is on site, and what happens when he leaves the site, it's documentation or debriefing. You see there's a lot of tools already in place that help us to do that, But there were a few missing elements that we touched on already and that is
[19:17] often the context was missing where GenAI can help um to better brief technicians. Um context from un- unstructured information. And also, we always had a little bit an issue with the documentation part afterwards. And
[19:32] exactly these points, we'll see it, we addressed that afterwards. How do we do that? Um that was important to us is um the human stays in the center of all of that. So, even if we now introduce the GenTech AI and have LLMs to help us,
[19:49] the the human still owns the results for everything. That's super important because I cannot blame an LLM for doing something wrong. Um but how did we now do that uh concretely? So, if we look at this, we introduced in some five five special-
[20:07] specialized agents. On the left-hand side, you see data sources. Some of them we already connected earlier. So, IoT data we saw, also data from our operations from the enterprise like the service history, maintenance recommendations. What's new is now we also look at documents like contract
[20:24] documents, PDF documents cuz we had the challenge that not all of the information was really sitting in ERP systems or in field service management systems as humans missed to update the data points there. So, if you really wanted to know it, you really needed to
[20:39] look into the contract. Yeah, and and now we bring we read these contracts, um bring it together with all of the other information, again have humans validate it, and then aggregate the data up, and give it to the technician um through the dispatcher. We all we call this all
[20:56] data-centric agents because that's happening in the background and not in the user contracts. On the on the right-hand side, you see two user serving agents. We have the deep brief the the briefer, which really helps the service technician to find out
[21:12] what what's up what what do I need to do on the next service visit. With information on how how to fix the unit with IoT information on but also with contractual information. And then we have the deep briefer. But let's take a closer look at how this works.
[21:28] So, here we see again the agents and we see a picture of of our tablet device showing how the information is presented to the technician. Technician can read this or there's also a button and can read to
[21:44] him, which then helps him to really do that already in the car. So, you can read it out and the technician can can get prepared while he's on the way to the job site. The right-hand side we see more contractual information like what's
[21:59] the SLAs, what are the opening and closing hours of a building, things that when does the contract end, is there an obsolescence clause, stuff like that, which he in the past often did not have access to. Now, if we do that already, we we provide a lot more context to the
[22:15] service technician with help which helps him to fix the the issue faster. But we can also use that now for the documentation. Um you Everybody knows that nobody likes to document things, right? So, it's and and and in our case, technicians
[22:31] very often we we earlier started with predictive maintenance, yeah. And then we we we gave them guidance and but we never really heard back how how helpful that was or got very very limited information. In some cases in the ticket system, you basically find a
[22:48] fixed, which is nice cuz the technician thinks he fixed the problem, but he never really gave us the information what exactly he did. Now, with all of the context that we spun out through the briefer, we now have a conversational agent that is really interviewing the service technician in the context of the
[23:06] respective elevator unit and contract and gets us the information that we need that we can afterwards also provide to other service technicians. So that really solves um real problem which is about documentation and it guess what is also
[23:22] a knowledge system because with the people leaving the industry they know how normally it would be gone now we extract it from them and make it available to others. So all in all we really built a continuous learning system that learns from every
[23:37] interaction with the technician and builds up a knowledge system while while we do it in the process. So we don't do it as a as a separate task but rather when it's happening. Um
[23:53] so that's the let's say the the case and how we how we added to to what we did before. Um we built the foundational elements for that um through the the 10-year journey that we saw. Um and uh let's uh conclude with uh
[24:09] with the two things. So first of all what what did we learn here on this journey? Uh important I think is we did not just do gen AI because somebody was saying you need to do AI in a gen team. We really looked at you saw it before a business problem. What did we solve already and
[24:26] what was missing and can technology help us to also solve that? Um so that's super important. Um another topic is choose the right technology for the problem. So even though now we have LLMs in a gen team AI that's not the best solution for
[24:42] everything. Predictive maintenance we still do with old AI or machine learning and some things are just deterministic problems or rule-based problems so we use that. Um the other thing is we thought we need to break down the bigger problem into smaller pieces
[24:58] so that they're digestible and implementable and that all of these individual artifacts already create value. So, modularize it. And the other thing, we heard it already before, that we combine data and AI expertise with
[25:14] domain knowledge in the field. Because only if you bring both together, you can really improve the business. So, let's conclude with what's next?
[25:30] Just one more one more slide and then as Catherine said, there's time for questions if there are some questions and we're happy to discuss. Um what's next? Well, first one is we want to scale proven capabilities. A lot you have seen is live, but not everything is live in all
[25:46] countries. So, we want to scale as much as possible and as much as allowed. There are certain regulation in certain countries why we can do things or cannot do things. All of the things you have seen is real, but not everything is North America, not everything is in Germany, not everything
[26:03] is in Korea. But we want to scale as much as possible per country and across countries what's possible. Second of all, uh Christian mentioned right from get-go he's always making sure that we don't build it once and then have to
[26:19] throw it away, but we have to that we're building it in building blocks, reusable artifacts that we can scale. And for sure in this new world, we heard it this morning, the governance is even more important than ever before just because of the power of the tools that
[26:35] is available to experts or semi-experts. So, we got to be careful about the governance and that's what what we will do. For sure, we will focus on scalability, on security, we'll focus on cost management because we don't only
[26:52] want to do AI the first quarter of the year, but also the rest of the year. And we want to do enterprise great by design. That's one of the biggest takeaways I had this morning that Databricks is so much focusing on
[27:07] on global scaling, on enterprise great design. So, we need to be ready for that, and we need to make sure that whatever we do, we think scaling, and we think modularity. All of that brings us to hopefully an intense and fun journey that all of
[27:23] us can go together. And as I said when we started, we want to bring together the elevator world with our wonderful, strong service technicians and our great technology with this data, AI, cloud world to really make sure
[27:39] that we can transform the service with new technology helping the service technicians. So, I hope after our Christian and my talk, when you walk out, you have a better eye for all the elevators, escalators that serve you every day. So, when you walk San Francisco, when
[27:54] you walk New York, when you walk I don't know, daily or so, just check the elevators, think about this session, think about us, think about TK Elevator, and we hope that you will enjoy the ride.

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.