Competing on efficiency: from 300 hours to 52Competing on efficiency: from 300 hours to 52

MAPSA S.A. is the only Argentine-owned company that manufactures cable harnesses for top-tier automakers. It competes in a demanding, globalized market where most inputs are imported and its main rivals are multinationals with greater scale, international footprints and operating advantages. The structural cost gap can reach 30% to 40%, which makes internal efficiency the only ground where the company can truly compete.

The most critical bottleneck was quoting new projects. Quoting a harness means analyzing highly complex electrical and dimensional drawings, identifying every component, quantifying operations against 89 internal standards and cross-checking all of it with costs in the ERP. A typical project of 30 to 40 harnesses took about 300 hours of engineering work.

The underlying problem was that quoting is, to a large extent, already producing: 70% of the technical development is done before the company knows whether it has won the project. Every lost bid meant hundreds of hours of work with no return, and responding in a reasonable time required putting several people on the same analysis at once.

Before reaching today's solution, MAPSA spent eight months building its own Python system with a dedicated systems engineer. It reached a solid technical level in image detection and data extraction from drawings, and delivered high-quality results in specific cases.

The problem wasn't the technology but scalability: the system had to be fed manually with every new drawing type or customer. It didn't learn on its own, didn't grow with the business and created a structural dependency. That experience became the conceptual and technical foundation for the scalable solution that followed.

The system takes electrical and dimensional drawings in standard formats (PDF, JPG) and analyzes them with computer vision and language models. It identifies components (connectors, terminals, wires, tape segments), infers the production operations each element requires under the 89 internal standards, and automatically generates both the bill of materials and the assembly time estimates.

The differentiator is in the prompts: the team built a layer of instructions that encodes "the MAPSA language," the company's own production and quoting logic. The system doesn't just detect what's on a drawing: it knows what each element means for MAPSA, how long it takes to process and how it maps to ERP costs. Human review remains part of the workflow.

The technology evolved gradually, from custom Python to integrated platforms such as Claude, Gemini and Codex. Today, staff with no programming background can generate working code using natural-language instructions.

The most visible impact is on time: a project that used to take about 300 engineering hours is now done in 52. That's a reduction of more than 80%, and with it goes one of the sales team's biggest opportunity costs: pouring massive resources into projects that aren't always won.

Accuracy also improved significantly. The maximum deviation across the projects analyzed is around 5%, and it keeps falling as the system takes in more cases. For a company whose quality standards demand zero defects, that is a structural change.

Beyond engineering, AI has spread to back-office work, internal communications, knowledge management and training new talent. Leadership has turned this into a concrete goal: shift from a 70% operational / 30% strategic split to 20% operational / 80% value creation.