The adoption of artificial intelligence (AI) technologies has become one of the main objectives of high-tech companies and start-ups in recent years. However, once these technologies reach the factory floor of physical manufacturing plants, reality proves to be far more complex. The gap between the quiet, air-conditioned development environment and the operational conditions on the ground creates ongoing friction between entrepreneurs and industrialists.
“I wear two hats - a high-tech hat and an industrialist’s hat - and connecting the two is full of challenges,” explains Tzuri Dabbah. According to him, the root of the adoption problem lies in a fundamental lack of understanding among technology professionals regarding the nature of work in manufacturing.
One of the key issues Dvush points to is the dramatic gap between demo presentations and real-world conditions in a factory. “Someone presents a demo where everything works and looks great, but when the solution reaches the production line, you discover a completely different world. The environment is noisy, dusty, humid, hot or cold, and the operational intensity is unforgiving. People are constantly moving, there are forklifts and cranes - these are things that simply don’t happen in an air-conditioned office.”
In addition to the physical environment, there are also issues involving terminology and regulation. Many entrepreneurs arrive without familiarity with the strict standards that apply to manufacturing plants, and often without command of the industry’s professional language. Dvush recommends that developers themselves use AI tools to learn the relevant regulations and local terminology in advance: “Even a simple word like ‘mold’ is called something different in almost every industrial sector. To establish the right initial connection, you have to speak the language of the factory.”
While the high-tech industry is accustomed to launching products in beta and improving them along the way, this model simply does not work in manufacturing.
“start-ups come in and say, ‘We’re 80% accurate,’ or ‘The system will learn as it goes.’ That’s not enough. Industry needs something that works at accuracy levels approaching 100%,” Dvush emphasizes. In manufacturing, even a small error rate does not mean a mere “bug” in the system. It can mean defective products, production-line shutdowns, damage to expensive equipment, reputational harm, and costs amounting to enormous sums.
Another obstacle gaining momentum is the “token economy” and the high usage costs associated with AI models. Without defining the payment model in advance and clarifying who is responsible for processing costs, factories may face inflated and unexpected monthly bills. In addition, the issue of ownership of the data fed into AI systems requires careful legal and regulatory arrangements from the outset.
Despite these challenges, the adoption of AI in industry is essential, and the path to success runs through realistic expectation management. Dvush outlines several key principles for entrepreneurs and factory managers:
Stop promising “we’ll solve everything”: Entrepreneurs should focus on a specific, deep-rooted problem and offer niche expertise, rather than presenting themselves as capable of solving every possible challenge.
Define clear success metrics (KPIs): The factory and the technology provider should establish the project’s success metrics in advance and provide a transparent dashboard for ongoing monitoring.
Start small: It is recommended to begin with a pilot in a defined area that does not put critical production-floor operations at risk.
Partnership models: Working through a model that reduces initial costs or shares the risk can help alleviate industrialists’ concerns.
Alongside these warnings, the market appears to recognize the potential inherent in this connection. Organizations and investment funds, including the i4Valley incubator, which specializes in industrial AI, and IL Ventures, which focuses on later-stage investments, are investing significant resources in ventures that manage to overcome the adoption gap.
“Entrepreneurs have a responsibility that extends beyond their individual project,” Dvush concludes. “Every failure of an AI system on the production floor erodes industrialists’ trust and makes things more difficult for those who come next. Transparency about the risks and clearly defined boundaries from the outset are the only way to build long-term trust.”
This article was written in cooperation with Klil