Transforming Luxury Rug Operations with AI

A long-established manufacturer of bespoke luxury rugs needed to streamline how orders moved from customer requests to finished products. Critical steps, from design intake and vectorization to production scheduling were spread across manual processes and disconnected systems. This created unnecessary friction, slowed decision-making, and limited the ability to meet demanding lead-time commitments.

The approach consisted of three main initiatives that changed the operations day-to-day work:

‍

1. Production Time Estimation

In the first initiative, we  introduced predictive models to estimate production times with greater accuracy and visibility across the value chain. Three machine learning models were trained on historical data to predict total production time, injection cadence, and vectorization effort.

ERP integration: Predictions were made available directly in the ERP, both through monthly retrained models. On-demand calculations are available whenever a new order is created.

The machine learning models reduced MAPE (Mean Absolute Percentage Error) by 10 percentage points, improving accuracy from 31% to 21% compared to the company’s baseline. As a result, sales and planning teams can quote and negotiate with greater confidence, while production benefits from ERP-embedded forecasts that enhance workload balancing and capacity planning discussions.‍

‍‍

2. Integrated Production Planning

In the second initiative, we introduced an optimization model to bring more rigor and transparency to production planning. The company manufactures rugs using injection/tufting processes, guided by robots that move along an X/Y plane to inject yarn according to a predefined pattern. The model allocates orders across available production lines and determines the assignment and sequencing of robots and screens, respecting constraints such as yarn availability, equipment compatibility, and production calendars.

The order allocation step decides which production line or machine should handle each rug order, while robot/screen assignment focuses on which specific robot and corresponding injection screen will execute each design within that line.

Finally, the system allows planners to stress-test scenarios and compare alternative strategies, analyzing objective functions such as production efficiency, load balancing between manufacturing lines, and service level. Optimization consistently matched or outperformed baseline planning.

• On-time completions improved by +3.7 percentage points.
• Plans are produced in ~90 minutes, enabling planners to generate daily or weekly runs within operational constraints.

Today, planners can rely on this optimization model to generate constraint-aware, data-driven production plans that reduce delays, improve visibility, and free time previously spent on manual adjustments.

3. Automated Vectorization Validation

Vectorization is the process of preparing paths the rug will follow during production designs for production. It is labor-intensive, relying on manual reviews to catch issues like misaligned angles or missing layers. We designed a system that automated the identification of non-conformities in design files, running rule-based validations on G-Code and generating continuous visual overlays so designers could see issues emerge in real time:

• Rules detect common errors (e.g., angles below thresholds, overlaps, missing elements).
• The interface allows users to configure detection thresholds and choose which error types to flag.
• Output is a marked design file plus a summary of detected issues, reducing ambiguity in correction cycles.

‍

Processing a complex rug now takes an average of about 30 minutes, significantly faster than the manual review process, which typically ranges from 3–4 hours and can extend to several days for intricate designs. In addition, the tool standardizes quality assurance during design preparation, establishing a more consistent foundation for reliability in subsequent planning and production stages.