Artificial intelligence (AI) is considered one of the biggest growth drivers of the coming years. But the reality in many companies is sobering: the majority of pilot projects fail to achieve lasting success. While billions are being poured into generative AI (GenAI), the effects often fizzle out in practice – especially in industrial SMEs.

The high failure rate of AI initiatives

Studies by renowned research institutions paint a clear picture: around 95 per cent of all GenAI pilot projects aimed at increasing sales do not deliver any measurable added value. The reason for this is a discrepancy between the possibilities of modern AI technologies and the specific processes in companies.

While flexible, generic AI tools work well in private or individual contexts, they struggle in a corporate environment. They cannot adapt to specific production processes, supply chains or service processes – and thus fail to achieve their actual goal: to increase value creation.

Legacy IT systems slow down progress

This problem is particularly acute in the manufacturing industry. Here, innovative AI approaches encounter outdated ERP systems, MES platforms or machine parks that often lack modern interfaces or compatible data structures.

An analysis by the WIRKsam competence centre concludes that almost every second AI project fails not because of the technology itself, but because of insufficient data quality, a lack of integration options and organisational hurdles.

Added to this is a competence problem: while many IT departments invest in customised developments, the specialist departments often lack the know-how to use these solutions effectively. According to a study by Stifterverband and McKinsey, around 80 per cent of companies complain of significant deficits in their use of AI – even though management recognises its potential.

Info box: Typical stumbling blocks in AI projects

Where companies fail particularly often

• Lack of data quality: Incomplete master data or inconsistent item numbers lead to incorrect forecasts.
• IT-driven projects: When AI projects are managed exclusively by IT departments, there is often a lack of understanding of operational challenges.
• Oversized custom solutions: Complex in-house developments result in high costs and long project durations, without any guaranteed benefits.
• Legacy systems without interfaces: Old ERP and production systems block smooth integration.
• Competence deficits within the company: A lack of expertise in specialist departments prevents the effective use of AI.

Practical solutions instead of expensive experiments

A promising way out lies in preconfigured AI applications for industry. These so-called industrial AI apps are specially tailored to production and business processes and can be used immediately.

The advantage: instead of years of development phases, companies can start immediately and obtain reliable results from day one.

Practical examples:

• An automation specialist used AI-supported inventory planning to optimise several hundred components and significantly reduce storage costs.
• A global robot manufacturer implemented an AI-based search solution that gives tens of thousands of employees worldwide faster access to technical documents.
• A medium-sized electronics manufacturer used AI-supported production planning to increase its delivery reliability from 86 to 96 percent within a few months.

Four success factors for AI in medium-sized businesses

To ensure that AI projects do not get stuck in the lab but deliver benefits in everyday life, four key success factors can be identified:

1. Give specialist departments responsibilityIt is not the IT department but purchasing, production or sales that should lead AI projects. They are familiar with the bottlenecks in day-to-day business and can define the benefits in tangible terms.
2. Ensure data qualityOnly with clean master data, unique item numbers and clear governance rules can AI systems deliver reliable results.
3. Gradual integrationInstead of abruptly replacing complex legacy systems, an evolutionary approach is recommended: AI functions are gradually connected via interfaces and middleware without jeopardising operations.

From experiment to transformation

The era of pure AI experiments is coming to an end. What is needed now are scalable, industry-specific solutions that integrate seamlessly into existing systems and create immediate added value.

For medium-sized industrial companies, this marks a turning point: those who now rely on tried-and-tested industrial AI apps can realise competitive advantages, increase process efficiency and, at the same time, consistently complete the step towards digital transformation.

The crucial question is no longer whether AI will reach medium-sized companies – but who will successfully master the transition to systematic use.

Background

The MIT study cited in the article is entitled ‘The GenAI Divide – State of AI in Business 2025’ and was published in August 2025 by the MIT Media Lab, specifically by the NANDA initiative (Networked Agents and Decentralised AI). The aim of the study was to measure the actual business benefits of generative AI projects in companies and to reveal the reasons for their frequent failure.

The researchers based their findings on the analysis of 300 documented AI implementations worldwide, supplemented by 150 structured interviews with CIOs, CTOs and department heads, as well as a survey of 350 employees who were directly involved in pilot projects. The results were sobering: 95 per cent of all generative AI pilot projects had no measurable effect on revenue or profitability. Only around five per cent led to directly demonstrable growth or significant efficiency gains. The MIT identified a fundamental ‘learning gap’ between generic AI models and specific business processes as the main cause. While tools such as ChatGPT score highly for individual users due to their flexibility, they fail in a corporate environment because they cannot learn from operational processes or adapt dynamically.

Other key findings of the study concern implementation strategies. Many companies continue to rely on tailor-made AI developments, even though off-the-shelf solutions often have a faster impact. Added to this are classic obstacles such as insufficient data quality, a lack of interfaces to legacy systems and organisational barriers. A skills gap was also particularly evident: 79 per cent of the companies surveyed stated that they did not have sufficient AI expertise. At the same time, 86 per cent of executives recognise untapped potential, but lack the necessary skills and resources for successful implementation.

The MIT researchers therefore recommend placing greater focus on application-oriented, specialised AI solutions that are tailored to specific industry and use case needs. This requires a robust database with clear governance and clean master data. In addition, pilot projects should be more firmly anchored in the specialist departments, as these have the greatest practical relevance, while IT takes on a supporting role. Finally, the study advocates evolutionary integration into existing systems rather than disruptive complete conversions – only in this way can the gap between technological possibilities and operational reality in everyday business life be closed in the long term.