#171 Turning building data complexity into AI-powered opportunity
Show notes
Before AI can optimize building performance, building data must first be structured and understood.
In this episode of Buildings of Tomorrow, we speak with Paul Baumann to unpack why data onboarding is one of the most complex steps in digitalizing buildings, and how AI is helping to simplify it.
Tune in to find out how AI is helping buildings make sense of their data, and what that means for the future of operations.
Transcript
Buildings generate an enormous amount of data every day from HVAC systems and lighting to occupancy sensors and energy meters, but before any analytics or AI can create value, there is a less visible step that must happen first. Adding all the data into a usable format. In this episode of Buildings of Tomorrow, we are going to unpack why onboarding building data has traditionally been one of the most complex parts of digitalizing buildings and how AI is starting to change that, and, more importantly, how this is paving the way to the next phase of building operations, self-learning systems. Hello, I'm Isabelle Fritzen, and I will be your host. Joining me today is Paul Baumann from Siemens Smart-Infrastructure Buildings. Thank you, Paul, for joining us. And let's start with the basics. Before we can analyze building data or generate insights, we first need to onboard it. In reality, how easy or difficult is that process? Thank you for having me. Yes, in reality, it's often the hardest part to start the journey. For instance, most buildings today have dozens of systems installed like heating, ventilation, cooling, lighting, meters, access control, all generating data. But the problem is that every system speaks a slightly different language. The naming conventions are different, the structure are different, and sometimes the commutation is incomplete or even missing. So before you can run analytics or AI, someone needs to make sense out of all this data. Traditionally, that means engineers manually going through thousands of points just to understand what each M represents to ensure that later AI can also understand the data. That's why onboarding can take weeks or even months in large buildings, and part of the reason for that complexity is that buildings rarely have just one system, right? You might have HVAC from one vendor, lighting from another, and different generations of technology layered on top of each other. But what typically happens when organizations try to connect all of that together. So yes, what usually happens is attempt to overcome this fragmentation. Each system produces valuable data, but it exists in its own structure and interface, so operators often end up juggling several dashboards just to understand what's happening in the building instead of undergoing costly integration of all systems. That remonstration is exactly why data on boarding and normalization are so important. And platforms like BillingX create a common data layer where information from different systems. Whether they are backnet, mothballs, Knes, or whatever proprietary protocols can be sanitized. Once the data is standardized and normalized, operators can move from managing individual systems to understanding the building as one connected environment. OK, Paul, but in other words, before a building can become smart, someone first has to translate all the raw information coming from the systems. When you actually connect a building system for the first time, what does that raw data typically look like? It's usually not very intuitive. You might see data points named like 1103. Or AI 147 for bid management system that originally created them, those names make sense. Furthermore, often these names are also written in native language of the one who created them. But for a platform trying to analyze billing performance, they don't intermediately tell you what they represent. Is it a chilled water sensor, a supply air temperature, which piece of equipment? Where is it located? So the first task is translating annotating those technical names into something meaningful like chilled water supply temperature sensor or it's a room temperature of zone 3. Is what needs to be done. That translation and annotation step is really what onboarding is about. So that was very helpful. And if you had to explain onboarding to someone outside the technical world like myself, how would you describe it? Well, I usually compare it to organizing, let's say, a books library before anyone can actually read or search for a book. You first need to label them, categorize them, and place them in the right section. Building data is very similar. Every sensor, a meter, a piece of equipment produces signals, but those signals need to be labeled and organized so that software applications can understand them. What platforms like Billing A do is help automate that process with the help of AI. We collect and analyze the information gathered from various connected systems. The information is gathered from technical floor plan drawings, from underlying automation systems and engineer configuration files of controllers, billing management stations, or even from documentation of files available with a customer. This allows us to build a data structure of the bill needed for further analytics and reporting. Using the building X data model, we create a common understanding of the complexities in the underlying billing data by adding meaning to it, defining the relationship within the structure, and identifying the dependencies across disciplines and domains. That's a very good example. So where does most of the time typically go during onboarding? Well, uh, the biggest time thing is annotating validation. So, uh, engineers need to identify every sensor, meter, valve, piece of equipment and the data stream. Then they verify that each data point is mapped correctly to a piece of equipment to which it belongs. In a large commercial building, you might be dealing with data points ranging from 1000 to 100,000 data points. Of course, not all data points in the underlying system running on site are necessarily for onboarding into the cloud. It becomes more important for the onboarding engineer to exactly understand the data available on site and then select the data points and assets which are useful for for analytics, insight generation, and autonomous operation of a building via a cloud platform like Building X. So yeah, that's why historically onboarding required a lot of mental effort. It was not unheard of that an onboarding technician may take 2 to 3 months to complete an onboarding process for such a building, with a lot of time spent in going back and forth between understanding what is available in the building and what is needed to be on boarded to the cloud. Yes, and that brings us to AI, the buzzword. If onboarding has traditionally been such a manual process, where does AI start to change this? So AI helps by recognizing patterns in data. Instead of a human manually identifying every data point, AI models can look at how signals behave, compare them to known equipment patterns, and suggest what those points likely represent. For instance, the system might detect a group of signals that behave like a typical air handling unit, supply air temperature, return air, fan speed, valve position, and automatically suggest that equipment structure. That is where platforms like Billing X are increasingly using AI system on boarding to access this process. We use a combination of techniques to ensure accurate and reliable results. Natural language processing like NLP helps us understand the meaning of each data point and description. While machine learning classification models then map these data points to the right equipment systems and assign confidence scores, rules based on billing X data model further improve classification accuracy. In more complex cases, again, large language models, LLM helps resolve tiebreaks and increase coverage while maintaining accuracy. And we are also starting to hear about things like Ask Building X, which introduce generative AI into the building space. How does that fit into this evolution? Well, it fits perfectly. Ask BuildingX is a good example of how to generate AI makes building information much easier to access and use. In building operations, engineers and facility teams often need to look up details about equipment, things like specifications, installation requirements, or documentation. And instead of searching through manual or product catalogs, they can simply ask questions in natural language. For instance, someone might ask, Can you show me the dimension of my humidity sensor installed in a giving building? AI can then retrieve that information instantly from relevant documents or system knowledge, so it acts like a digital system for billing systems, helping operators quickly access the information they need without digging through multiple sources. What is also important is how we support the democratization of skills. A junior technician, for instance, can onboard data or connect systems with more confidence using semantic enrichment. When paired with ask building acts, they have the right knowledge at the fingertip based on the context. This helps close the knowledge gap we often see in facility management today. And of course this becomes even more powerful when the underlying data and devices are properly structured through platforms such as Building X. So once the data is structured and all those systems are connected, that's when the real value starts to appear. What changes for operators at that point? What becomes possible that wasn't possible before? Well, a lot changes. A lot becomes possible. First, operators gain true operational visibility. Instead of looking at isolated systems, they can see how equipment across the building is performing in one place. Second, platforms can run advanced analytics such as fault detection and diagnostics. This means the system can automatically identify inefficiencies or equipment issues before they even become a major problem. And third, You can start optimizing performance across a building or even across the entire portfolio. For instance, in building acts, applications like Operations Manager can identify operational anomalies, while Energy Manager can benchmark energy performance across multiple buildings. This is when buildings start moving from reactive operations towards data-driven operations. Yes, and this is also where we start moving from simply smart buildings towards something more advanced like self-learning models that are foundational to autonomous buildings. How does the system actually get smarter over time? Well, it happens gradually as the system learns from operational data. Once building data is structured and continuously collected, AI models can start recognizing patterns in how the building behaves, things like occupancy pattern, where the impact, equipment performance, or even energy consumption. Over time, the system can identify what normal looks like. That means it can detect deviation earlier, predict potential issues, and even recommend operational improvements. So instead of reacting to alarms, operators can start managing buildings proactively. That's really the transition from a smart building to an autonomous building. For instance, a software engineer office hosts a quarterly planning meeting with a lot of stakeholders visiting the office for 3 days every quarter. When the occupancy of a building is more than doubled. The system will learn this behavior of a building and ensure that the various systems of a building, such as comfort, lighting, fire safety, security, energy, are prepared and ready for the spike in occupancy and address the demands for the occupant accordingly. As another example, seasonality in the hospitality industry is an important consideration for a hotel. The service technicians need to be made aware by our work orders to perform preventive maintenance checks in time for peak season when a hotel building needs to be in prime working condition. The system learns this based on the billing behavior over time and can schedule these work orders for the service technicians autonomously or via facility management teams. Paul, I believe for the operators to rely on that data they need to trust it, right? So how is that trust established? Trust obviously is built in several ways. So first, through transparent data models where operators can clearly see how data points are structured and mapped to equipment. Second, through industry standards and interoperability, which ensures systems can communicate reliably, and 3rd, through security and independent verification. For instance, digital platforms like BillingX and like have achieved Smart System verified platinum certification. Which validates both the building management layer and the cloud platform against strict cybersecurity and performance criteria. That kind of independent verification gives building owners confidence that their data and operations are protected. In the spirit of openness and flexibility, the billingX data model has been designed to be flexible and extensible while also maintaining a mapping of all major industry standards anthologies. And data models such as Project Haystack, BRI schema, Ashtray 223P, Real Estate core, and so on and so forth. Thank you so much for already sharing a lot of insights today, but before we wrap up, let's make this really practical. If a building owner or facility manager listening to this conversation wants to start their digital journey tomorrow, what are the three things they should think about first? Well, I would suggest the following 3 things. First, treat billing data as a strategic asset, not just a byproduct of a billing system. Second, Choose platforms that support open standards and interoperability so systems can integrate easily, and 3. Invest early into a strong data foundation, because once your billing data is structured, it becomes much easier to unlock analytics, automation, and eventually self-learning capabilities. So the messy truth about building data is this most buildings already have the information they need. The real challenge is turning that information into something usable. Once that happens, AI can start connecting the dots, helping operators see patterns, make better decisions, and gradually move toward buildings that improve themselves over time. And that's where the future of self-learning autonomous buildings really begins. Paul, thanks so much for joining us today. Thank you.