Challenges of AI in manufacturing: what 52 practitioner conversations show
The challenges of AI in manufacturing are not primarily technological. In our analysis of 52 published practitioner conversations (September 2025 to August 2026), 42 of 206 pain point statements concern data, IT/OT and integration, and a further 32 concern leadership, culture and change. Technology as such hardly appears as a problem.
For decision makers in brief
- Core finding: Industry has an implementation problem, not a technology problem. The conversations show this just as clearly as the large surveys.
- Effort: The largest effort lies in data access and integrating heterogeneous machine fleets, not in choosing the model.
- Cost: Cost and ROI sit in the middle of the field with 10 mentions. If you do not plan for the effort of data integration, the cost question later turns into disappointment.
- Risk: Leadership and culture are the second largest bottleneck. Transformation that is imposed rather than shaped slows down even well prepared initiatives.
- Consequence: The first question before any use case is not which model, but which data comes from which system at what frequency, and who releases it.
How the analysis was carried out
Surveys measure agreement with predefined categories. To check whether the findings also appear where practitioners speak freely, we fully analysed twelve months of published expert conversations from industry: 52 conversations, 206 individual pain point statements, 19 user companies speaking. 27 % of the conversations come from users on the factory floor.
A methodological caveat belongs at the start. The analysis reflects how the speakers present themselves, not market reality. It is based on the published summaries of the conversations, not the full recordings. We use it as a qualitative cross-check against the representative surveys, not as statistics.
Challenges of AI in manufacturing by frequency
Each statement was assigned to one or more topic areas. The table shows the mentions per cluster.
| Rank | Cluster | Mentions |
|---|---|---|
| 1 | Data, IT/OT, integration | 42 |
| 2 | Leadership, culture, change | 32 |
| 3 | Complexity and variants | 18 |
| 4 | Implementation and scaling | 16 |
| 5 | AI, expectation versus reality | 11 |
| 5 | Skilled workers and competencies | 11 |
| 7 | Cost and ROI | 10 |
| 7 | Paper, knowledge, documentation | 10 |
| 7 | Location and competition | 10 |
Source: own analysis, mentions per cluster from 52 conversations, multiple assignment possible.
Cluster 1: data, IT/OT and integration
No other topic comes close. The statements mention data silos, missing standards, heterogeneous machine fleets, retrofitting old equipment, gaps in OPC UA and PackML, a patchwork of IT systems and technical debt. These challenges around IT/OT integration arise where machine data, OT networks and security requirements meet.
The surveys confirm the finding from another direction. In mechanical and plant engineering, 43 % use AI in the company (VDMA, the German mechanical engineering association, 2025), yet only 13 % in their own production (Fraunhofer ISI 2024). Adoption starts in software development, marketing and service, because manufacturing raises exactly the integration questions collected in cluster 1. How these requirements differ by industry is described on our page on AI for manufacturing companies.
Cluster 2: leadership, culture and change
In second place are statements about mindset as a buzzword, fake lean, resistance among the workforce, silo thinking, transformation that is imposed rather than shaped, and a lack of acceptance for AI decisions. Together with cluster 1, these two topics account for more than a third of all statements.
The spring 2026 survey by the ZEW Leibniz Centre for European Economic Research provides a matching figure: in manufacturing, only 30 % of companies actively provide AI applications to their employees, 39 % merely tolerate their use and 8 % expressly prohibit it. Change management is therefore not a side issue but part of the project.
Two perspectives, one finding
When it comes to the challenges of AI in manufacturing, representative surveys weight things somewhat differently. Bitkom, the German digital industry association (2026, n=604), names data protection requirements as the most frequent barrier to AI and digitalisation at 77 %, followed by the shortage of skilled workers at 70 %, lack of time at 66 % and IT security at 61 %.
| Perspective | First barrier named | Second barrier named | Source |
|---|---|---|---|
| Survey | Data protection, 77 % | Shortage of skilled workers, 70 % | Bitkom 2026 |
| Practitioner conversations | Data, IT/OT, integration, 42 of 206 | Leadership, culture, change, 32 of 206 | Own analysis |
The ranking in the survey is striking: data protection comes before skilled workers, skilled workers before time, time before security and security before money. Lack of funding is named by 48 %, significantly fewer than the organisational and legal barriers. The problem is rarely that a model does not work. The problem is that it cannot reach the data and nobody has time to run it.
This does not contradict the conversations. The survey names the risk, the conversations name the effort. Both perspectives point to the same place: data is worth protecting and at the same time hard to access. If you want to solve both, you need an architecture in which sensitive data does not have to leave the company. This is exactly what the approach of AI in your own data centre stands for, with models running behind your own firewall.
What remains unresolved
43 of the 52 conversations explicitly name something that remains open. That is a more honest maturity indicator than any success story, because successes are told in public while failures almost never are. Four questions come up particularly often:
- From lighthouse to scale: the most frequently asked question of the period. Four companies answer it differently, through templates, train the trainer, governance or architecture guardrails.
- Who owns the data: data sovereignty runs as a leitmotif through at least four conversations on Manufacturing X, Factory X and Process X.
- What in AI and robotics is real today: one AI vendor speaks of an implementation gap, the potential is there, broad deployment is not.
- Is the role of integrators changing: self-integration and in-house development in the German Mittelstand, mid-sized companies with roughly 50 to 1,000 employees, call classic models into question.
Three blind spots
Three topics come up conspicuously rarely during the entire period, even though they border directly on the dominant challenges. For decision makers they are the real to-do list, because no ready-made answers are circulating there.
First, operations and hosting. There is a lot of talk about use cases and almost none about where the model runs, who operates it and how operations are safeguarded. Regulated AI operations as a managed service close exactly this gap with defined ownership of model quality and retraining.
Second, the transition from prototype to production. Architecture guardrails, technical debt and multi-tenancy questions appear only in exceptional cases, even though the overall picture of the analysis shows that this is exactly where scaling fails.
Third, pure consulting is missing. Exactly one of 52 conversations comes from a classic consulting firm. If you want to be heard in this environment, you need a user story or a product, not a service promise.
How the biggest barrier, data and integration, can be reduced in concrete terms is described in our article on the barriers to AI adoption.
