While the AI conversation remains dominated by speculation about what might be possible, some organizations have moved beyond experimentation to deploy AI systems that handle millions of transactions, automate complex workflows, and deliver measurable business value. Their experiences reveal both the extraordinary potential and sobering realities of implementing artificial intelligence at enterprise scale.
We spoke with seven industry leaders who have successfully integrated this technology in real-world production environments, with projects ranging from processing over 300,000 municipal documents a month to automating customer support for more than 500 representatives. Their insights go beyond the hype, offering a clear-eyed view of what it truly takes to unlock AI’s transformative potential at scale.
Our distinguished interviewees
Frederik Severin leads AI initiatives at SumUp, a major German fintech company, where he has overseen the simultaneous deployment of over 200 AI projects. Thanks to these efforts, the organization has achieved 50% automation in customer support while upholding enterprise-grade quality standards.
Clémentine Lalande is the co-founder and CEO of kelvin, a French company that has developed specialized AI engines for analyzing the energy efficiency of residential buildings. Her team has created entirely new frameworks for AI applications in the energy retrofit sector.
Kimmo Parviainen-Jalanko serves as Engineering Lead at Vainu, a Finnish company that processes vast amounts of business data for the Nordic region. They have been implementing AI and machine learning solutions for over eight years, well before the current GenAI boom.
Ariel Rosenfeld is CEO of 3d Signals, an IoT company that connects physical manufacturing assets to cloud systems. With a background spanning from founding M-Systems (the inventor of the USB flash drive) to ultra-marathon running, he brings a unique perspective on digital transformation in industrial environments.
Gil Matzliah is the CEO and co-founder of NoviSign, a global SaaS company specializing in digital signage. He has been integrating AI content generation capabilities into their platform, carefully balancing innovation with operational stability.
Rosaria Silipo is the Head of Data Science Evangelism at KNIME. She brings nearly three decades of experience in data science and neural networks to help organizations integrate AI capabilities into their data analytics workflows.
John Smith (a pseudonym), who has requested anonymity, is a seasoned professional in e-commerce advertising optimization. His company uses predictive AI to optimize PPC bidding across major platforms like Amazon, eBay, and Walmart, with a focus on aligning customer success with business outcomes.
Industry use cases: Where AI is delivering value
AI is no longer confined to the realm of experimental technology—it’s becoming a core part of how businesses operate. According to McKinsey, 78% of organizations report using AI in at least one business function. For leaders navigating this landscape, understanding the broader implications of current AI adoption statistics is critical for benchmarking strategies and identifying emerging opportunities.
From manufacturing and finance to marketing and customer service, companies across industries are finding concrete, value-generating use cases.
The following examples from industry leaders illustrate how different sectors are adopting AI not just to enhance efficiency, but to fundamentally rethink how work gets done.
Customer support: When AI handles 500 jobs
Customer support is one of the most resource-intensive functions in any large organization, often requiring hundreds of staff to manage inquiries quickly and accurately. It has traditionally been viewed as a necessary overhead with limited room for innovation. That’s what makes Frederik Severin’s story so remarkable—if the results weren’t measurable, they might seem almost unbelievable.
As the leader of AI initiatives at a major German fintech company, Severin has overseen the simultaneous deployment of over 200 AI projects. The transformation of customer support is the biggest initiative SumUp has done so far. And it clearly demonstrates the true scale of what’s possible with AI.
“At our company size, that would have required 1,000 customer support people. We used to have 1,000 or almost 1,000 customer support people in the past, and now 50% of the cases coming in are handled solely by AI.”
The mathematics is staggering: AI now manages the equivalent of 500 full-time customer service representatives. But this wasn’t achieved through a simple ChatGPT integration. “There are a couple of engineers working full-time on it for almost a year, because they have been connecting all relevant data sources,” Severin notes.
This level of implementation marks a fundamental shift from seeing AI as just a productivity tool to treating it as essential business infrastructure. Severin’s team has organized this approach into what he calls a “three-stream strategy”:
- Internal development led by dedicated engineering teams to build use cases.
- Collaborations with startups that apply AI on top of foundation models for rapid deployment.
- Providing democratized access to over 40 AI tools across all employee roles.
“We categorize our strategy into three different streams,” he explains. “One workstream is building out use cases internally. The second workstream, AI projects, utilizes an AI layer based on the foundation model, with the application case being conducted by a startup. The third one is that we try to get contracts with big enabling tools where everybody inside the company can apply for certain seats.”
The sophistication becomes clear when he explains the security requirements: the requester must be signed into their account, and only a specific set of information is accessible. “To achieve a 50% success rate, it is usually necessary to access their financial history.”
Perhaps most importantly, Severin has recognized that enterprise AI requires different quality standards than those of startup implementations. “If you have a big operating business, in contrast to running a startup, you need to work towards 99 point something in accuracy, in uptime, in non-hallucinating. It requires way more testing, training, reviewing, and doing QA than in a startup.”
Building AI for problems that don’t exist yet
While Severin’s organization focuses on scaling proven AI applications, Clémentine Lalande, co-founder of kelvin, has built something entirely unprecedented: an AI system that can analyze any residential building in France and recommend optimal energy efficiency improvements.
“We have developed an artificial intelligence to massify energy retrofits in residential buildings,” Lalande explains. Her company has created three interconnected AI engines that work together in ways that would be impossible for humans to replicate at scale.
The first engine performs geo-statistical analysis, cross-checking dozens of databases to create comprehensive building profiles. “We are connected to databases that range from land registers and Google Maps, to more confidential databases, and they give us estimations about the building’s construction date, its area, wall material, and the orientation—whether it faces south or east.”
At this stage, data availability isn’t the main challenge—consistency is. “We have a lot of data, but the different sources don’t really talk to each other,” Lalande explains. This forces the engineering team to continuously find “common ground across all the data sets. The quality is very heterogeneous, so the first AI layer focuses on predicting a high-performing building profile and its 1,000+ features.”
The second AI engine takes this data and creates 3D models of buildings, making intelligent assumptions about shape, height, and roof configuration. “We go from something that’s just a data set to completing the missing parts and finding what the house actually looks like,” Lalande describes.
The third engine models energy efficiency and calculates optimal retrofit scenarios. Together, these systems can analyze buildings and generate recommendations that would take human experts days or weeks to produce manually.
What makes kelvin’s approach remarkable is that her team had to invent entirely new frameworks.
“The field of application is so new for AI that we have to come up with our own frameworks and models. We have to go back to the existing literature on how to calculate energy efficiency, but all of that is just literature, not data models.”
This cutting-edge development carries real risks. “One of the complexity factors of what we are trying to achieve is that there is close to little ground truth, or at least not at scale. Plus, this is a massive unsupervised ML problem: there is no right or wrong in the energy retrofit space. So we had to aggregate the data for our models to learn and iterate,” she recalls.
The complexity extends beyond technical challenges to regulatory differences across European countries. “The way we calculate energy performance in France is different from Germany, and it’s different from Poland,” Lalande notes, highlighting how AI applications often face obstacles unrelated to the technology itself.
From clicks to revenue: The next generation of AI ad optimization
In the high-stakes world of e-commerce advertising—where profit margins are razor-thin and competition is relentless—artificial intelligence is becoming a powerful tool for gaining an edge. John Smith, an industry expert, shared insights into how advanced AI is being used to optimize bidding systems, revealing just how sophisticated these applications have become in today’s most dynamic markets.
“Our job is to set the optimal bids,” he explains. “If your bid is too high, then you lose money because you pay too much for a click. If your bid is too low, then there’s no volume, you won’t win an auction.”
The challenge requires predictive AI that can make forecasts while incorporating seasonality, product attributes, audience behavior, location data, and device information. “We really need to do two things: forecast the conversion rate and forecast the market volume, depending on the bids or number of impressions and clicks.”
This AI platform stands out for its ability to address a fundamental misalignment in digital advertising. “Google’s interest is not to make the customer profitable. They want to earn money, “the expert explains. In response, his company shifted to a model that charges clients based on the revenue generated from ads rather than the ad spend itself, showing how AI can enable business models that better align the interests of providers and customers.
Behind the scenes, the technical implementation relies on sophisticated risk management practices. “We usually ask customers if they want to join a beta, and we have a soft rollout system. We usually have three versions running at the same time, just to switch back and forth if the new version causes some problem.”
Looking ahead, Smith sees critical challenges that extend beyond technical capabilities. “I think these are the challenges for the future: to somehow enable users to fine-tune the system, to talk about data privacy, to control it, to know what’s going on, to tweak it to make sure it’s compliant.”
Smith’s concerns about authenticity are especially timely:
“One big challenge in the future is not knowing anymore what’s real and what’s not. Is it a fake product or a real product? Is it a fake company or a real company?”
A related development gaining momentum is what Smith refers to as paid agentic commerce. Modern AI agents can already assist with product discovery, and they have also started to integrate a direct purchasing option, allowing customers to buy desired items without leaving the conversation or interface. The next step, Smith believes, will be enabling sellers to actively advertise through these agents. This evolution will blur the line between organic recommendations and paid placements, creating fresh challenges around bias, transparency, and fairness in AI-driven commerce.
Data intelligence at unprecedented scale
Kimmo Parviainen-Jalanko from Vainu offers insights into processing business data across Nordic countries. “We collect company data from multiple sources and provide it to sales teams and financial institutions,” he explains.
Vainu’s AI has been performing named entity recognition on news articles for years. It analyzes streams where “95% don’t mention any companies” to link relevant mentions to specific businesses in their company ID database.
The real innovation, however, lies in their emerging generative AI capabilities. “We use large language models to generate company summaries automatically, with zero human effort,” Parviainen-Jalanko says. “Users can also create their own prompts and select which data sources to include.”
More significantly, they’re developing systems that can answer questions impossible to address through traditional database queries: “Does a Swedish company do business in Poland? That’s something that cannot be a direct data point, because it’s impossible to get anything there. That’s something you can do with LLM.”
This level of innovation, however, comes with its own set of challenges, particularly the rapid pace of change in underlying platforms. “All the vendors in the space are very fast, but also very slow,” he explains. “They bring out new, improved models, but then they have limitations. So not everything built for old models works for new ones.”
Looking ahead, he offers a prediction about the future of business communication that is both amusing and concerning:
“I think there’s gonna be this thing where models sell to other models. Somebody writes three bullet points, then you use some fancy GenAI to turn it into a three-page presentation, and then the other person puts that into another model and gets the three bullet points back out.”
When AI meets manufacturing reality
Ariel Rosenfeld, CEO of 3d Signals, faces a unique challenge: applying AI in physical-world manufacturing environments, where data quality and real-time decision-making are critical. His team connects physical assets, which are machines on the shop floor, to the cloud world behind end-to-end IoT.
What sets their approach apart is the exceptional quality of data their systems generate. “The data that we have is clean,” Rosenfeld notes, describing how they process sensor noise to determine machine states with precision. This clean data foundation unlocks AI applications that would be impossible with messy or unstructured inputs. It’s also paving the way for generative AI to become a seamless part of factory operations.
“Since we generate this data, we are going to implement Gen AI as another layer of our solution,” Rosenfeld explains. “You can ask questions in the context of our system, like: Where are my bottlenecks in the factory? Where should I pay attention?”
He argues that successful AI in manufacturing depends not just on algorithms, but on having the right data architecture in place from the start. “If an AI company comes to the factory, they have no data, so they cannot ask that. We generate this data, and we can do that.”
But while the technical capability is clear, Rosenfeld emphasizes that the real hurdle is not the technology—it’s user behavior. “The biggest challenge is how users will leverage that,” he observes. “We saw users who learn from the system, use the data, and regenerate. We never imagined the other extreme: users who don’t use the data at all.”
This variability in adoption has shaped his broader view of AI’s role in the workforce. Rather than replacing people outright, he believes AI is transforming how they work—and what is expected of them. “There are so many things that I needed to use consultants and experts. And now we have all the experts in there.”
The content generation revolution
Gil Matzliah, CEO of NoviSign, shares insights from the digital signage industry that highlight how artificial intelligence is reshaping creative processes across sectors. His company has taken a forward-looking approach, embedding AI capabilities directly into its platform to move beyond static templates and enable dynamic content generation.
“Over the past year, we’ve been trying to adapt AI to our needs,” Matzliah explains.”We embedded it within the product—not only from scratch, from templates—we also added the capability to build content screens using AI.” This integration shows how established companies can bring artificial intelligence into their workflows without disrupting existing systems.
Matzliah’s vision of the future also touches on how user interfaces will evolve: “More and more things that you’re doing today with UI, you’ll probably do with a prompt. I’ve already seen people speaking with the NoviSign interface, instead of opening a designer tool.” He acknowledges this shift won’t happen overnight, but sees it as inevitable: “It won’t happen next year, but if I look into the crystal ball, I’d say it will likely happen within the next five years.”
As NoviSign continues evolving its platform, Matzliah is also focused on the broader strategic implications of adopting AI, particularly the need to balance bold innovation with business continuity.
“What’s the right balance? If we invest 100% in AI, it’s probably too much, and then we can go the wrong way. If we don’t invest at all, then competitors will pass us by.”
This thoughtful approach extends to how the company views its workforce. Rather than using AI to replace people, Matzliah sees it as a way to amplify human capability: “Internally, we will not shrink the number of employees. We see ourselves developing better, faster, more accurately—and being able to expand the solution.”
Making AI work: The integration challenge
Rosaria Silipo, head of data science evangelism at KNIME, offers a different perspective. Rather than directly building AI solutions, her work focuses on helping organizations integrate them into real-world processes. KNIME’s platform supports a wide range of use cases—from simple LLM prompting to complex AI agents—making the technology accessible and actionable for non-specialists.
As part of this broader mission to support more intelligent and responsive systems, KNIME recently introduced new capabilities that enable more flexible workflows. One such development is the introduction of AI agent functionality, marking a significant step toward more autonomous and adaptive systems.
“We have a node that can implement your AI agent. It finds the necessary tools, rights, prompts, and intermediate prompts to generate this application. It’s fantastic,” she shares.
Silipo, who has worked with neural networks since the 1990s, brings decades of experience and historical depth to the current wave of AI enthusiasm: “It’s such a new world that everybody thinks they know it, but I don’t think they’ve explored it completely.”
Her insight about AI’s primary value cuts through much of the hype surrounding generation capabilities:
“GenAI is sort of like a term for the thing that it does. While its generative abilities have undeniable advantages, I believe there’s an untapped business opportunity in using it to distill information. For many businesses, the shorter the answer, the better.”
She also highlights one of the most pressing obstacles for organizations trying to work with AI at scale: a lack of standardization in how AI is used, particularly with large language models: “You write a prompt, you write stuff, but it’s very sensitive: you change one thing, and the result changes. I think one of the main challenges now is standardizing how we ask things of AI so that we get repeatable results.”
To help others make sense of today’s AI anxieties, Silipo draws a useful comparison to past tech transitions, such as the early days of cloud computing: “When the cloud came up, it was the same issue. The banks were saying, ‘Do I have to put the data of my customers in the cloud. And what about if it’s secure or not secure?’ There was a reasonable doubt, but then the cloud providers ended up defining business offers that would give protected spaces.”
Key lessons for successful AI adoption
At Neontri, we believe that AI is not just a tool but a strategic capability that must be thoughtfully integrated into the core of an organization. Successful implementation requires more than technical expertise—it demands clear objectives, strong leadership, and a culture ready to embrace change.
The insights shared by industry leaders align closely with our own experience and highlight several critical lessons for any organization looking to implement AI effectively.
Start with clean data and clear use cases
Attempting to apply AI to unstructured or noisy data often leads to underwhelming results. Rosenfeld’s success comes from designing systems that generate clean, structured, and highly relevant data rather than trying to apply AI to messy existing datasets.
This data quality enables the development of meaningful and accurate AI applications, such as identifying production bottlenecks or providing operational insights. Without this foundation, even the most advanced AI models can fall short.
Balance innovation with operational requirements
Innovation alone isn’t enough—especially in enterprise environments. Severin underscores that, unlike startups, established companies must prioritize operational stability, uptime, accuracy, and compliance. This means devoting significant resources to testing, validation, and quality assurance. AI must meet enterprise-grade expectations, which can demand more engineering and infrastructure investment than many anticipate.
Invest in organizational learning and culture
Successful adoption isn’t just about technology—it’s also about people. Severin’s company doesn’t just roll out solutions; it builds AI literacy across the organization. This includes executive education programs, training for non-technical employees, and democratized tool access. The goal is to empower every team to explore and use AI effectively.
Prepare for regulatory challenges
Expanding AI applications across different regions, especially in highly regulated industries like insurance or finance, introduces complex legal and compliance issues. Lalande’s experience navigating European markets shows how regulatory environments can vary dramatically and create unexpected roadblocks. AI leaders need to be proactive about understanding local rules and be ready to adapt their solutions accordingly.
Focus on integration
Many companies overemphasize developing cutting-edge models without considering how these systems will integrate into day-to-day operations. Silipo points out that success comes from embedding AI into real workflows, making tools accessible to non-experts, and ensuring consistency in how they’re used.
The workforce reality: Adaptation, not elimination
Across all seven interviews, a consistent theme emerges: AI changes the nature of work rather than simply eliminating jobs. However, this transformation demands active adaptation from both organizations and individuals.
Severin’s workforce philosophy encapsulates this perspective: “It’s not AI that is replacing people, but people knowing how to work with AI replacing people not knowing how to work with AI.” His organization has invested heavily in education, developing an onboarding course and a general three-level guide on AI, which is currently being prepared.
Lalande predicts differentiated impacts across job categories: “Part of the blue collar jobs, for sure, won’t disappear. And if everything goes well for us, they should multiply. But a lot of the intermediaries currently in the process—some of them will disappear or will be made so efficient that the volume will reduce.”
Her outlook is more sobering when it comes to knowledge workers. “I do believe that blue-collar jobs are fine, but white-collar jobs should get worried.” This perspective stems from her experience with AI systems that excel at identifying patterns and anomalies that often go unnoticed by humans, particularly in areas like fraud detection and process optimization.
Rosenfeld emphasizes that successful AI implementation is about augmenting human capabilities, not replacing them. “It shortens cycles, brings more expertise, and increases the quality of work,” he explains. “That said, because we’re now working more efficiently and professionally, there are certain positions where I don’t need to recruit more people to do the same amount of tasks.”
Embrace the future of work with AI
AI is reshaping roles across industries, offering opportunities for growth and efficiency—but only for those ready to adapt. Discover how your organization can prepare teams to thrive alongside AI and unlock new potential.
What’s next: Expert predictions on the future of AI
As AI moves from experimentation to enterprise-wide deployment, leaders across industries are beginning to chart the road ahead. The experts interviewed offer not only real-world examples but also valuable foresight into what’s coming next. From ethical concerns and data privacy to the evolving role of human workers, their insights paint a nuanced picture of the future AI landscape.
- Platform commoditization and competition. Parviainen-Jalanko observes that switching vendors is very common between AI providers, suggesting that competitive advantages will come from application sophistication rather than platform selection. “I’m pretty sure that part of the competition being so fierce is that for the technologies that we use, switching the vendor is pretty much trivial.”
- Interface evolution. Matzliah envisions a fundamental shift in human-computer interaction. He suggests that conversational interfaces may gradually replace traditional graphical user interfaces across many applications.
- Quality and standardization. Silipo highlights repeatability as one of the key unresolved challenges. This underscores the need for further progress in building AI systems that produce consistent, reliable results, especially in enterprise settings where accuracy and predictability are critical.
- Infrastructure maturation. The anonymous advertising expert anticipates AI becoming “the absolute standard” in many applications, with success determined by how well it is integrated rather than whether it’s used at all.
From possibility to practice: How to make AI work at scale
The experiences of these seven experts show that scaling AI successfully requires much more than adopting the latest models. It calls for robust technical architectures, rigorous quality assurance, strategic workforce development, and careful attention to regulatory and operational demands.
More importantly, their stories reveal that AI’s greatest impact doesn’t lie in automating existing tasks, but in unlocking capabilities that were previously out of reach. From analyzing building energy efficiency at a national scale, to optimizing ad bids across platforms in real time, to enabling natural language queries about factory bottlenecks, the most effective implementations don’t just reduce costs—they create entirely new forms of business value.
For organizations beginning their AI journey, the key takeaway is clear: success doesn’t come from racing to adopt new tools. It comes from deliberately building the data infrastructure, organizational maturity, and strategic clarity needed to translate AI’s potential into measurable outcomes.
If you’re looking to put these lessons into action, don’t go it alone. Reach out to explore how your organization can build the capabilities needed to scale AI effectively and turn it into lasting business value.
