Requirements management at an automotive supplier: how a custom AI tool replaced Excel

Requirements management can be largely automated today: an AI-supported tool reads customer requirement specifications and technical documents, extracts the requirements they contain and structures them in an ISO-compliant way, where previously days were spent transferring everything into Excel by hand. We show what this looks like in practice with a project for an automotive supplier, where the solution runs entirely on-premise for data protection reasons.

The starting problem: requirements in Excel

At our client, a mid-sized automotive supplier, every development project starts with customer documents: requirement specifications, load requirements, technical specs, often hundreds of pages in different formats. Engineers had to extract the relevant requirements from these documents by hand, transfer them into Excel lists and bring them into a standards-compliant structure. This cost many hours of highly qualified working time per project, was error-prone, and whenever the specification changed, the reconciliation work started over. Excel had long reached its limits: no traceability, no clean versioning, no link between the customer requirement and its internal implementation.

Why the cloud was not an option

Requirement specifications are among the most sensitive material a supplier handles. They contain development details from vehicle manufacturers and are subject to strict non-disclosure agreements. Sending these documents to a cloud AI service was never up for discussion for our client, and would likely have violated the agreements with its customers. The requirement was therefore clear from the start: all processing must happen in-house. We describe in detail why local AI is the right answer for exactly these cases in our article on local AI and GDPR.

The solution: a custom tool on local AI

We built an AI system for the client that runs in its own rack on company premises and works without an internet connection. On top of it runs a requirements management tool tailored precisely to the engineering team''s workflow: customer documents are ingested, the local language model extracts the requirements they contain and proposes an ISO-compliant structured version. The engineers review, correct and approve. Professional responsibility stays with the humans, but the tedious groundwork is done by the system. Every requirement remains linked to its source in the original document, so it is always traceable where it came from.

The result

Creating the requirements structure, which previously cost many hours per project, now takes a fraction of the time. The engineers review and decide instead of transferring and formatting. Quality has gone up, not down: the system never skips a page and does not get tired in the afternoon, and the consistent link back to the source makes audits and customer queries considerably easier. And because everything runs locally, the confidentiality question stayed exactly where it belongs: entirely uncritical.

What other companies can take away from this

The pattern of this project transfers to many industries: wherever professionals manually transfer information from extensive documents into structured formats, whether requirements, contracts, test reports or tenders, a tailored tool with AI support can take over most of that work. What matters is that the solution fits the actual workflow and that humans stay in control. We show how we build such tools on our AI solutions and software development pages, or get in touch directly if you have a similar process in-house.