Introducing AI to the Office
10,000 Images in Eight Weeks: ATP integrates AI on a Grand Scale
A company with 1,500 employees that is seeking to encourage and scale up innovation requires a focused structure. We report how ATP achieves this.
ATP employs more than 1,500 people in 14 locations. And it is integrating AI into every area. Learn more about the challenges and successes of the digital transformation of a large architecture office.
When ATP architects engineers introduced a so-called “AI hub,” its impact could be felt surprisingly quickly: Within eight weeks it had created over 10,000 AI-generated images, which have already been used in a range of projects. But this hub is only one part of the office’s comprehensive digitalization strategy as it expands the use of artificial intelligence (AI).
Various teams are working on the development and application of AI solutions in areas such as design, environmental analysis, and cost planning. They have created and implemented tools including chatbots and bespoke AI models. Albert Achammer, Managing Director of ATP in Hamburg, explained to competitionline how the company is adapting its structures so that it can make targeted use of these innovations.
Achammer studied in Zurich and worked with GMP in Hamburg before completing an MBA in Spain. Returning to Hamburg, he established a new subsidiary of ATP architects engineers.
This founding of an office within a large group offered us the opportunity to test new approaches and workflows, which we were then able to implement across the company.
Albert Achammer
Architect ETH, Executive Board, Partner
Computational Design Team
While ATP doesn’t have a specialized AI Department, a number of teams across the company are intensely involved in the subject. Four years ago, for example, the Hamburg office established a specialist, decentralized Computational Design Team. Programs such as Grasshopper or Dynamo had already been used in the company in specific situations, but never as a package. The new team unified such efforts and intensified the work on Computer Aided Design.
This group of experts made progress in a range of AI applications – from plan and volume generators to image generation algorithms and environmental simulations. “These technologies are now widely spread and work well,” reports Achammer.
The work of the team was divided into three areas that were covered by a Development Team, an Application Team, and a Superuser Team. The Development Team works almost exclusively on research & development, while the other teams use the new technologies in practice and provide feedback. “This structure enables us to ensure that the innovation process is supported from the bottom up and, eventually, leads to widespread application,” says Achammer.
“Two speeds”
The hype around GPT-4 and other AI tools led to increased interest in AI, even amongst employees who had little affinity towards IT but hoped that they would soon be able to automatically generate designs at the push of a button. “Since then, AI has developed within the company at two speeds: the first represented by the work of the experts and the second by all the other developments stimulated by AI,” explains Achammer. But these two tendencies are coming ever closer together.
In practice: the introduction of new technologies
ATP’s approach to the introduction of a new technology such as AI is as follows:
The process often emerges from the personal initiative of highly motivated employees.
The product owners from the four areas of the Digital Business Improvement Program established by the office (see box) – Integrated Design, Construction Management, Sustainability, and Building Services Engineering – develop specific applications in exchange with architects and engineers.
Employees who want to develop such ideas draw up a brief so-called “Research & Development Proposal” on a single page of A4 that sketches out the idea, implementation, costs, and expected result.
These suggestions from the operational team go directly to the Executive Board, which rapidly decides about implementation in line with the digitalization strategy.
This leads to the creation of interdisciplinary development teams consisting of designers, software developers, IT support, and further specialists.
The innovation is then introduced in steps. The new technology is first developed and tested in small teams; this, for example, is how a prototype of ATP’s own chatbot “ChatATP” was created.
This is followed by a beta phase in which 20 to 40 people work with the new technology. Eventually, several hundred employees use the application.
We move forward step-by-step in order to gather practical experience and optimize the technology before it is introduced more widely. This method often rapidly leads to far-reaching results.
Albert Achammer
Architect ETH, Executive Board, Partner
Tools
ATP is increasing its use of external software. For example, the Computational Design Team works with the modelling software Rhino Inside Revit, to which it adds its own codes for such features as the automatic configuration of ceiling plans or the connection of building services pipework. Tools such as iTWO and DBD-BIM and applications in the Autodesk Construction Cloud (ACC) also boost efficiency in calculating costs and writing tender texts. “The integration of this system is a huge step forwards,” says Achammer.
ChatATP
ChatATP is a chatbot assistant for ATP’s employees. It was developed in order to reorganize knowledge management in the company. The tool resembles common chatbots such as ChatGPT and Copilot.
“We have a huge variety of norms, standards, and in-house know-how. ChatATP is a powerful tool that simplifies and improves access to all this information and data.”
The development process is currently principally focused on norms and standardized data. In future, the office plans to further expand the abilities of the system. This is based on the Microsoft SharePoint database in combination with Microsoft Azure and API from OpenAI.
LoRa models
A hackathon in early 2024 expanded the AI server in the form of Low-Rank Adaptation (LoRa) trainings. LoRa is a method that adapts an AI model in line with specific tasks or data by training just a small part of a model. This permits ATP to individualize the model further and align it with a specific style. LoRa trainings reduce the volume of calculation and make it possible to generate diverse representational styles for a range of applications.
ATP uses LoRas in a different way to other offices, says Achammer. Rather than an ATP style enhancer, a range of style enhancers are used for a range of applications – from the local context in a visualization to a special style of display in a presentation. “After all, as architects we don’t only base our output on our own work but also allow external inspirations to flow in,” says Achammer. “This enriches us and improves outcomes.”
Here is an example of variations in a range of styles with a LoRa model from an AI-generated original image of an office building:
AI hub
One key objective for 2024 was the establishment of an in-house AI hub. The AI hub is an online AI server for the realization of Stable Diffusion. This platform enables ATP teams to rapidly create high-quality visualizations using simple prompts. These visualizations can be used in later work phases from concept development to presentation.
A team with expertise in both AI and machine learning tested a range of systems during a two-year period before finally selecting the open source model Stable Diffusion.
During the introduction of the AI hub, employees developed five tutorials and generated over 10,000 AI images in just eight weeks. “This volume, and the excellent quality of the results, surprised even us,” says Achammer.
The integration of Low-Rank Adaptation (LoRA) models enables the AI hub to optimally adapt the results in line with the project requirements of the office – for specific building types, facades, and visual styles.
“This innovative hub is changing the way we visualize and refine designs and puts us at the forefront of AI-supported architecture. The server is part of our AI research,” says Albert Achammer.
Environmental and data analysis
ATP worked with Autodesk to integrate the AI-based tool Forma for environmental analyses in the early phases of projects. Thanks to AI surrogate models, this delivers rapid results regarding noise, sun, and wind conditions and helps designers to make well-founded decisions. The company then turns to other, more complex tools to carry out more precise calculations in later work phases.
ATP also uses Large Language Models (LLM) such as GPT and Copilot to analyze larger, text-based volumes of data. This simplifies such tasks as evaluating competition parameters and analyzing user data – always in line with data protection regulations. Smaller applications such as the organization of internal competitions also benefit from these tools.
On the building site
ATP uses sensors and scanners to compare the latest model of the project with progress on site. “These developments are a game changer,” reports Achammer. Tools like the Autodesk Construction Cloud (ACC) and PlanRadar simplify communication by directly integrating information into the model. Due to the complexity of developing in-house solutions, ATP uses tried-and-tested suppliers in such situations.
Networks and clients
ATP hugely values exchange and cooperation as a way of jointly driving developments, says Managing Director Albert Achammer. “We’ve always been very transparent with our innovations because competition in our sector is based not on workflows but on good ideas. That’s why we published our entire BIM content years ago at BIMpedia.eu.”
One example of this is the Rhino User Meeting in Berlin, which had 170 participants and was organized by ATP. “We believe that it’s more effective for everyone to develop jointly and benefit from each other rather than keeping everything in house.”
The office also organizes the “DigitalOpenHaus,” an online event, at which ATP shares ideas with 25 companies and seven universities. Achammer believes that networks can also help small offices for whom investment in AI or BIM simply seems to be too high.
The Managing Director is constantly meeting clients, who have little understanding for or interest in new technologies. “They expect us to work quickly and well, regardless of how we do so – of whether we draw using AI or BIM or by hand.”
However, Achammer adds that there is a growing number of technology-conscious clients, who recognize the added value offered by digital tools that, by contributing to increased transparency, enable robust decisions to be taken earlier.
Advantages and challenges
Achammer sees two principal advantages in AI: additional inspiration and increased efficiency. “In the early phases, inspiration is decisive and efficiency emerges when the entire process works as one.” A key example of this is BIM: “Complete efficiency gains are only possible when it is consistently used throughout the entire process.”
Achammer hopes that AI will act as a sparring partner in future. Text-rich areas provide particularly simple use cases. Over the course of the past 70 years, ATP has gathered enormous amounts of data. One of the problems is how to deliver this know-how to the right employee at the right moment in the right data format. The company sees this as a huge opportunity to use AI in simple applications such as an intelligent search function.
“To date, AI has relied on human decisions in a specific context. In future, it can act as a member of the team that rapidly provides us with better information,” says Achammer.
He sees the initial work phase as one of the greatest challenges, because, as a rule, clients don’t know exactly what they want. AI can provide powerful support here by helping to process this vast variety of data, says the Managing Director. This will remove many of the loops in the project process that arise due to shortfalls in the requirements planning.
Digitalization strategy and corporate culture
About a year ago, ATP decided to fundamentally realign its digitalization strategy in order to deal with the digital transformation of the sector over the long term. As a result, the company launched its “Digital Business Improvement” (DIBI) program. DIBI aims to not only modernize the planning and design process but also, and most importantly, digitally reinforce and improve the efficiency of every area of business – from architecture and engineering, via knowledge management and HR, to legal processes.
“Merely employing a couple of software engineers with AI expertise isn’t enough,” emphasized Managing Director Albert Achammer. “We have to address the issue company-wide.” It is particularly important for a group like ATP, with its 1,500 employees, to establish a solid foundation in order to ensure that innovations really have room to take hold. Technologies can be rolled out quickly in a small office; but in a large company you need a bespoke structure if you want to encourage and scale up innovation without holding it back.
In order to create such a basis, ATP focuses on four central elements: culture, organization, process, and technology. These elements have to be seamlessly coordinated if we want to ensure that innovative approaches and technologies fall on fertile ground. Achammer explains: “Firstly, the corporate culture must be open to new approaches. Then we adapt the organizational structures to ensure that innovations are viable. In parallel with this, outdated processes also have to be redefined, because adapting the culture isn’t enough if the working methods lag behind.”
One decisive aspect of this strategy is the cultural transformation in the context of digital tools and information searches. While the focus used to be the Google search, AI-based requests will become increasingly central. This form of “cultural transformation” is being systematically encouraged by ATP because, while the technology is generally well developed, it often takes longer for our own thinking and processes to change. A similar phenomenon can be seen with BIM: The technology offers many possibilities, but in the absence of the corresponding mindset and structures, it rarely achieves its full potential.
A company-wide AI codex provides employees with guidelines for using AI tools such as GPT or Copilot and ensures that these technologies are applied responsibly and efficiently. In order to prepare employees for the challenges of digitalization and the use of AI, ATP is also building on the expertise of software engineers and further experts. “The basis for successful cooperation – both physical and digital – begins with the right mental attitude,” summarizes Achammer.
