The excitement surrounding so-called agentic AI is clearly palpable within procurement organisations. In technical procurement departments in particular, there is a growing sense that autonomous AI agents are on the verge of taking over complex processes independently – from supplier selection to price evaluation. Yet there is a gaping chasm between vision and operational reality, which has less to do with technological limitations than with structural shortcomings within companies.
From analytical tool to autonomous actor
The current stage of development in AI is fundamentally shifting the role of systems: away from supportive analysis (“Show me the data”) towards delegated action (“Carry out the task”). Management consultancy McKinsey & Company describes this shift as a transition to agent-based systems that independently prepare decisions, run through scenarios and iteratively optimise processes (McKinsey, 2025).
This potential appears particularly attractive in technical procurement. Here, complex requirements – such as those from design, manufacturing and supplier management – meet high cost sensitivity. AI agents promise to master this complexity whilst simultaneously increasing speed.
The illusion of the working demo
Many companies are currently gaining initial experience with generative AI systems such as ChatGPT or comparable models. In isolated use cases, these systems deliver impressive results: they analyse technical specifications, formulate content or structure information.
However, this performance often leads to a misjudgement. What works in an experimental environment cannot simply be transferred to operational procurement processes. This is because different standards apply in procurement:
- Results must be reproducible
- Data must be standardised and interoperable
- Decisions must be auditable and robust
Plausible answers are not enough if they are not consistent and system-compatible.
The real bottleneck: a lack of data integration
The central weakness of many agent-based AI initiatives lies not in the AI itself, but in the underlying data structure.
Technical procurement is based on a multitude of heterogeneous data sources:
- CAD and design data
- ERP systems containing order and price information
- PLM data throughout the product lifecycle
- Supplier and manufacturing knowledge
In many companies, however, this data exists in isolation, in different formats and without a uniform semantic framework. For AI agents, this means they operate without a consistent foundation.
Without an integrated database, typical problems arise:
- inconsistent classifications
- lack of comparability between components
- inconsistent price analyses
- lack of integration into existing systems
The result: results remain fragmented and are of little use in operational practice.
Industrial data as a prerequisite for automation
For Agentic AI to actually create added value in technical procurement, an upstream transformation is required: the establishment of a structured industrial data layer.
This must be capable of:
- automatically recognising geometries and component features
- standardising technical specifications
- mapping manufacturing processes
- linking data from engineering and procurement across systems
Only when these prerequisites are met can AI agents demonstrate their strength – namely, making complex decisions based on a robust data foundation.
Leadership rather than a technology problem
The challenge is therefore less technical in nature and more a question of prioritisation and management. According to McKinsey & Company, the key bottleneck in many organisations lies at management level: data strategies, system integration and process harmonisation are often underestimated or addressed too late (McKinsey, 2025).
Agentic AI therefore requires a shift in thinking:
- away from isolated pilot projects
- towards holistic data and process architectures
Between hype and operational reality
The future of technical procurement will undoubtedly be shaped by AI. A hybrid approach, in which humans make strategic decisions and AI agents scale operational tasks, is considered the most likely scenario.
However, the difference between a functioning system and a failed project lies not in the performance of the models, but in the quality of the data.
Companies therefore face fundamental questions:
- Is our data structured and consistent enough for automated decisions?
- How can engineering and procurement data be meaningfully linked?
- Which systems form the basis for scalable AI applications?
Conclusion
Agentic AI has the potential to fundamentally transform technical procurement. However, many current initiatives will fall short of expectations – not due to a lack of technology, but because of inadequate data integration and a lack of strategic direction.
In future, the decisive competitive advantage will arise where companies master, structure and intelligently link their data. Only on this basis can an AI experiment become a productive system.
Sources: McKinsey & Company (2025): *Redefining procurement performance in the era of agentic AI
