Festo and Zeta Alpha: Scaling AI through Collaboration

How can generative AI create real value in industrial environments? Festo and Zeta Alpha share how a focused collaboration turned a proof of concept into a productive solution - and what startups and corporates can learn from it. An interview by Rebecca Pini - original version was published on 16.02.2026 on the VDMA website.

Contact us to learn more...

VDMA is the largest network organization and an important voice for the machinery and equipment manufacturing industry in Europe and Germany.

What triggered the cooperation between Festo and Zeta Alpha?

Jacob Decker (Festo):

The ChatGPT moment changed the discussion. Suddenly it was no longer about the question "Does AI work?", but "How can we use it?".

Festo identified generative AI as a strategic lever. However, customers don't pay us to be the best AI developers. They pay us to deliver the best automation solutions.

Therefore, we distinguish the big topic of "AI" into competencies that are close to our core business, e.g. machine learning models for detecting and classifying anomalies in automation components, and other fields, e.g. Large Language Models (LLMs), which are not our core competencies but non the less critical for sucessful business.

Instead of developing everything internally, Festo started a targeted scouting process and invited startups to propose solutions. Zeta Alpha proved to be the right match.

From the beginning, the goal was not experimantation, but realizing scalable and measurable value for multiple business units and products.

Jakub Zavrel (Zeta Alpha):

We are dealing with tens of thousands of technical documents - specifications, diagrams, engineering tables. The challenge is not just to answer a question, but to find the right document and to reference the exact source.

The teams developed a virtual assistant based on Retrieval Augmented Generation (RAG) that securely connects large language models with Festo's internal knowledge database.

Transparency and traceability were essential: the answers always refer to the original source document. While hallucinations cannot be eliminated entirely, quality and reliability can be significantly increased through architecture and governance.

How did you move from proof of concept (POC) to productive deployment?

Jakub Zavrel:

The initial project was planned as a three-month POC. Instead of delivering one use case, five were implemented across different business units. That Festo-internal momentum made scaling much easier. Several departments wanted to use the solution.

Because multiple stakeholders were involved at an early stage, follow-up funding was not a struggle. The demand for the solution came from within the organization.

Jacob Decker:

After the three-month development phase, evaluation, legal alignment and procurement processes followed before the productive deployment. Roughly one year after the POC started, the virtual assistant went live on festo.com. Since then, additional integrations into internal applications and saleable products have been rolled out. The architecture continues to serve as a scalable foundation for further AI applications.

What has made the collaboration work?

Jakub Zavrel:

Executive mandate and stakeholder access were critical.

In many startup-coroporate collaborations, the mandate ist not really there. The startup talks to innovation teams, but is not given access to real business stakeholders. In this case, all doors were open.

Having direct access to domain experts was essential to understand the industrial context and refine the solution beyond a generic AI application.

Festo's existing IT infrastructure also accelerated the implementation. Clear ownership prevented complexity from slowing down the process.

Festo almost felt like a very large startup - family-run, continuity-driven, focused on product quality and customer relationships. That value alignment led to a strong partnership.

Jacob Decker:

We already had a strong global IT foundation, clear APIs, expertise and infrastructure in place. This allowed us to focus on value instead of spending months on basic integration.

Without this foundation, also from our research colleagues, speed would not have been possible.

In a company with more than 20,000 employees, you need one clear contact person. Otherwise you lose speed.

Involving purchasing, legal and compliance at an early stage helped to prevent friction later on in the process.

What advice would you give to companies working with startups?

Jakub Zavrel:

Look for the middle ground. You don't need to build everything yourself - but standardized AI tools are often not enough for industrial complexity.

There is a growing ecosystem of specialized, highly innovative startups that understand industrial requirements. Leveraging that ecosystem can significantly accelerate innovation.

Jacob Decker:

Working with startups is not just about measurable results. It also keeps the entrepreneurial spirit alive inside the company. However, measurable results are where the "fun part" starts:

Anchor open innovation strategically. Define clear focus fields and give your partners a real mandate.

Without strategic anchoring, POCs risk of remaining isolated experiments rather than becoming scalable solutions.

With clarity, ownership and technical readiness, startup-corporate collaboration can move from an experiment to a scalable solution.

About Zeta Alpha: Headquartered in Amsterdam, Zeta Alpha develops AI-powered search systems for expert knowledge complex industrial organizations. The company focuses on secure and customizable AI for industry. Further information: www.zeta-alpha.com

About Festo: Festo, headquartered in Esslingen am Neckar, Germany, is a global supplier of automation technology and technical education solutions. With about 20,600 employees in over 250 branch offices in around 60 countries worldwide, Festo achieved a turnover of around 3.45 billion euros in 2024. Further information: www.festo.com