AI Agents in Quality Management - Guest Article on QZ-Online

AI Agents in Quality Management: How Semantic Matching Reduces Processing Effort by 70%

Together with the renowned trade journal QZ, we discussed "AI agents in quality management". In this article, we demonstrate why the core issue with product compliance inquiries is not a lack of data, but rather inconsistent formats, and how AI agents resolve this problem.

The Problem: Identical Questions, Hundreds of Formats

Quality managers are all too familiar with this scenario: a customer sends an Excel spreadsheet with 150 rows, requesting compliance data regarding REACH, RoHS, or industry-specific standards. The next customer asks the exact same questions, but in an entirely different format. What follows is manual copy-paste work: questionnaire by questionnaire, column by column.

One customer asks about "SVHC > 0.1 wt.%", the next about "Candidate List Status", and a third about "Art. 33 REACH compliance". The same inquiry regarding RoHS compliance may arrive informally via email, as a sub-question in a PDF form, or as part of a comprehensive Excel questionnaire. The administrative burden lies not in the answer itself, but in the manual completion of these forms.

The Solution: Semantic Matching with AI Agents

Large Language Models (LLMs) enable a novel approach: they understand the layout and format of questionnaires, identifying questions from both a content and structural perspective. The AI analyses column headers, response fields, and sheet relationships. Concurrently, it comprehends the semantic meaning behind different formulations of the same question.

This enables the automated transfer of knowledge between internal corporate data and customer inquiries, as well as between individual customer inquiries themselves.

Practical Example: 70% Reduction in Effort within Intralogistics

An internationally operating intralogistics provider regularly receives extensive customer inquiries regarding supplier disclosures, substance declarations, and product compliance. Most of the information was already available, yet processing previously required several hours per inquiry.

Following the implementation of the AI system, it now reads incoming inquiries, identifies the semantic meaning of the questions, and searches the designated knowledge base for relevant information. On this basis, it generates a draft response, which the specialist department reviews and approves. The result: internal coordination overhead fell by approximately 70%.

Preventing Hallucinations: Security by Design

In a compliance context, generating plausible but incorrect statements would be critical. To prevent this, the systems access internally approved documents exclusively. Information based on the general world knowledge of the LLM is not incorporated into the responses.

Additionally, the system utilizes an "LLM-as-a-Judge" approach: a second language model critically evaluates the generated answers against the underlying source documents. If the database is insufficient, no response is generated, and the respective position is explicitly marked as open.

The Role of Humans in the Process

Even if only 70% of the information is structured, its automated transfer represent a massive time saving. The remaining 30% are systematically flagged for manual review — precisely where human expertise is indispensable.

Key insights from practice:

- The achievable level of automation is directly dependent on the quality of the existing data

- Previously answered questionnaires represent the most valuable data source

- Team acceptance increases when every generated answer transparently references its source

- Specialist personnel retain full responsibility for the final approval

- With every manual correction, the knowledge base expands, continuously improving accuracy

---

The complete article including a video interview is available here on QZ-Online: AI Agents in Quality Management