Gunnebo is a global leader of security solutions headquartered in Sweden and on a mission to provide its customers with the safest devices possible. As part of this commitment, the company recently established Gunnebo Safe Storage Innovation Hub in Italy centered around researching new technologies and innovative ideas for its products. It has then established a partnership with Microsoft and embarked on a ground-breaking AI journey that’s led to the adoption of Azure OpenAI Service and its latest GPT-4 model. Currently in use to simplify, speed up and automate log analysis, the technology is already transforming operations at the company and driving efficiencies across the board.
“Sometimes I still find myself astonished by just how powerful and groundbreaking this AI model is.”
Bjorn Tore Nostdahl, DevOps Innovation Manager at Gunnebo, is reflecting on the extraordinary impact that a GPT-4 model is having on his company – and the initiative that led to its adoption.
“At Gunnebo, our mission is simple: to keep your valuables, life and world secure at all times,” says Nostdahl. “As the global leaders in security solutions and services, we feel the responsibility to keep our customers and their belongings safe – and to do that using the best technology available.”
It’s a commitment that Gunnebo has been solemnly devoted to for decades. One that recently culminated with the pioneering adoption of GPT-4 – OpenAI’s most advanced AI system – on Azure, as a means to improve logs analysis and drive efficiencies. Ultimately making its devices even more secure.
“In my 25-year career, I’ve seen many technology disruptors come to life,” he says. “The Internet of Things was one of course, and then the cloud, machine learning and artificial intelligence all have been disruptors too.
“But nothing has been quite as groundbreaking as GPT models. These are, by far, the biggest disruptors in my career.”
A company industrial by birth, innovative by choice
Gunnebo is a global leader in security, offering innovative products and services to protect and control the flow of people, and to safely secure valuables. Headquartered in Gothenburg, Sweden, Gunnebo operates in 24 countries with 10 production facilities and employs more than 3,500 people serving customers in more than 100 markets.
“We are what you would describe as a traditional, heavily industrial company producing anything from safety boxes to electronic gates and software solutions,” says Nostdahl. “At the same time, we pride ourselves on being a highly innovative and modern organization.”
He says so while speaking from the University of Bari in southern Italy, where Gunnebo recently opened the Gunnebo Safe Storage Innovation Hub which specializes in researching new methods for valuables protection and developing digital solutions. These include a connected safe that can alert the owner if it is under attack, an app for drive-up products, as well as the company’s EverydaySafe solution.
“The innovation hub was created to foster R&D activities on a range of topics including IoT and cloud technologies, which are my area of expertise,” he explains. “This is a relatively new area for Gunnebo, which has long been dedicated to on-prem installations rather than intelligent devices and software.”
But that’s increasingly changing. “On-prem solutions are still our core offering,” he says. “But in recent times we’ve been fitting them up with electronic software, gradually paving the way for a more IoT-ready set of products.”
Things have ramped up significantly over the past three years, during which Nostdahl and his team have moved much of the software onto the Azure cloud. “We have totally redone pretty much half of our software offering and made it cloud-native,” he adds.
“That has opened our doors to an entirely new player: the data they generate and the analytics that make sense of it.
“And has marked the very beginning of our AI journey.”
In search for the perfect solution
When Gunnebo turned to long-term partner Microsoft for support implementing Azure OpenAI Service, their request was simple: a solution that would allow faster and more efficient log analysis for their devices.
“Many of our products are fitted with software that generates logs – a particular set of data informing us on device performance, events happening within it, health and so on,” says Nostdahl. “Of course, this is the kind of data that we’re highly interested in understanding – showing us what’s working and where we need to intervene, but also new ways to support our engineers and customers.”
The challenge, he says, is analyzing the data itself. “We have very skilled engineers working for us, but log analysis is a tedious, long, and complex process that only a handful of people can perform.
“So we had bottlenecks within the organization standing in the way. That’s what led us to Microsoft and Azure OpenAI Service.”
At the time, a variety of GPT models were rapidly gaining traction – though none of them initially seemed to satisfy Gunnebo’s needs. “We tried some of the algorithms that were available without any success,” he says. “Some were too simple, others couldn’t understand our context well enough.
“The closest one to our requirements turned out to be gpt-3.5-turbo (also known as ChatGPT) which quickly surprised us with its ability to extract so much information from very little context."
But even that came with its challenges, with Nostdahl recalling experiencing AI hallucination – a phenomenon that occurs when an AI model fabricates an artificial story about log data instead of providing its summary.
This is a widely familiar issue in large language models. An issue that Gunnebo decided to resolve using Microsoft’s advice – eventually landing on GPT-4. “We quickly saw just how much better GPT-4 was at processing bigger documents and understanding human instruction, as well as reasoning upon complex data,” he says.
“We had finally found a model that would help us not just summarize events, but also detect abnormal behaviors in log activities.
“It was the breakthrough that we had long been looking for.”
The solution in a nutshell
Log analysis is the practice of searching for, visualizing and studying information generated by a machine in order to drive insights. At the core of it is the idea that these systems generate data – or logs – related to events happening inside the machine, such as malfunctions, updates, machine behavior and more.
The basic concept behind Gunnebo’s solution is that log analysis could be sped up and improved using machine learning. “Most Gunnebo products generate data for us to collect in three main ways: telemetry using Azure IoT Hub, raw log files using Azure Functions and information from our mobile app via Azure Application Insights, all visualized in Power BI,” explains Nostdahl.
“The data then needs to be transformed using Azure Data Factory, which makes it uniform and standardized. That way it’s easier to understand it and to work on it.”
That’s when GPT-4, as part of Azure OpenAI Service, comes into play. “We have created a web-based application that allows us to view the raw log files and analyze them using GPT-4 in the form of a chatbot,” he says. “This then feeds information to the engineers and responds with a short and easy-to-understand summary of log events.
“Our engineers can later visualize the results on our SafeControl Platform, easily detect break-down patterns, and use them to identify areas needing intervention. All with a view to increasing the product’s lifetime and quality.”
But adopting a state-of-the art AI model is only part of Gunnebo’s success. Instrumental to the project, says Nostdahl, has been feeding the GPT-4 model with consistent and qualitative data “We have spent the past two years gathering as much data as possible from 10 to 15 different products into our cloud, and we’re now using Azure Data Factory to normalize and standardize them,” he adds. “That’s forming the very basis of our proactive and predictive data analysis.”
Delivering analysis 10 times more precise than ever before
Gunnebo’s log analysis solution may not be fully complete yet, but its benefits are already more than evident.
“To give you an idea, from a typical log that we receive, the majority of its data is completely useless for us,” says Nostdahl. “So to visualize which information is useful for our work is a really tough job for anyone. It’s the kind of task that you do once and never forget it.”
That’s where the greatness of GPT-4 lies. “We can provide it with very little background information and still get great results,” he says. “Compared to what we had before it must have increased the performance and the correct analysis of the logs by 10 times at least.”
He says it’s saved engineers a lot of time too. “Before, an engineer would take at least a couple of hours to analyze log information from our devices,” he comments. “Thanks to GPT-4, that same analysis takes five to ten seconds.”
On the customer side, that means that the devices are easier to monitor, operate and maintain – and therefore better at guaranteeing security.
“Once I gave GPT-4 a log file in German without any context and I asked it to explain what this log file was for,” he says. “In a matter of seconds, it explained to me everything there was to know about it and that I couldn’t understand.
“That’s ultimately the beauty of this model: it’s helping us to protect our customers in a truly concrete and practical way.”
Next steps
This is just the beginning for Gunnebo. “Log analysis is merely the first step for us, our main goal is to have all our products connected to this solution in the future, especially to better meet our customer demands.
“They want to know their health status, the temperature inside, humidity, everything. Data is important for them and it’s therefore important for us. That’s precisely what we’re working on now: gradually making even the most complex ones – like those made of concrete and steel – IoT-enabled.”
To do this, Nostdahl is considering adding even more Azure services to the ones they’re already using, such as Azure Synapse Analytics. And also bring new layers of security thanks to a Zero Trust approach.
He says that changes are planned from a skilling perspective too. Implementing the Azure OpenAI Service and GPT-4 model has improved performance and efficiency for Gunnebo, but it’s also made it realize the need of a new competence within the organization: prompt engineering.
“The truth is that you don’t need to feed a lot of information for GPT-4 to find you a solution,” he continues. “What’s new is rather the quality of prompts and information you feed in the first place.
"That's why we hired our new Prompt Engineer trainee, Bahador Amjadi, whose focus will be solemnly on ensuring GPT-4 receives high-quality prompts and input to further improve log analysis.
"For us, that is a clear sign of where our future is headed: towards more integration with foundation models like GPT-4,” he concludes. “There is no doubt in mind on it. With Microsoft at our side, we know we can do incredible things with it.”
“Before, an engineer would take at least a couple of hours to analyze log information from our devices. Thanks to GPT-4, that same analysis takes five to ten seconds.”
Bjorn Tore Nostdahl, DevOps Innovation Manager, Gunnebo
Follow Microsoft