Case Study - How a Metal Fabricator Cut Quoting Time from 3 Days to 1 and Won 20% More Jobs

Client Context

A metal fabrication company in Northern Ireland with 51-200 employees, processing around 5 quotes per week. Each quote required extracting dimensions from DXF engineering drawings, matching them against a bill of materials, and preparing data for nesting software - a process that consumed days of skilled time and still produced errors.

The Problem

Quoting was eating the business alive.

Each quote took 2-3 days depending on staff availability. That meant skilled engineers buried in manual data extraction instead of higher-value work. It meant losing jobs to competitors who responded faster. And it meant errors creeping in when clients updated drawings mid-quote - there was no system to track versions or flag changes.

The volume wasn't the issue. Five quotes a week is manageable. The problem was that each quote was a resource-intensive, error-prone process with no consistency. Win or lose, the cost of quoting was the same.

Speed wasn't just about efficiency - it was about winning work. In metal fabrication, the first accurate quote often wins the job.

Why Previous Solutions Failed

The team had tried using off-the-shelf LLMs to extract data from DXF files. The results were inconsistent - around 50/50 accuracy. Sometimes it worked, sometimes it didn't.

That's worse than useless. If you can't trust the output, you have to check everything manually anyway. The time savings disappear and you've added a new failure point. They needed extraction they could rely on, not a coin flip.

Our Approach

The core challenge was accuracy. Anyone can throw an LLM at a DXF file and get something out. The question is whether you can trust it.

We built a system designed for reliability, not just extraction:

• Cluster analysis groups similar shapes, layers, and text blocks, finding repeated patterns across drawings
• A knowledge graph stores the relationships between parts, dimensions, and materials - trained on hundreds of real DXF files from the industry
• A scoring algorithm assigns confidence levels to each extracted item, so the system knows when it's certain and when it needs human review

The pipeline takes DXF data in, clusters features, maps them to the knowledge graph, and outputs a structured list of drawing contents with confidence scores. No black box - the system shows its working.

The Solution

QuoteFlow RFQ - a tool that extracts dimensions from DXF files, matches them against the bill of materials, and outputs clean JSON ready for nesting software.

But extraction was only part of the problem. We also built version control into the quoting process. If a client updates a drawing mid-quote, the system tracks it. If the nesting stage is already in progress, the system forces confirmation that the new drawing - not the original - is being used. No more errors from outdated files slipping through.

The output isn't just faster - it's auditable. The team knows exactly what data went into each quote and can trace any issues back to source.

The Results

Thousands in additional revenue from jobs that would have gone to faster competitors.

Proof

Delivered in partnership with Sans Souci, a digital transformation consultancy specialising in manufacturing technology.