Rapid accurate quotations with Visual Intelligence solution

A waterproofing products manufacturer for commercial buildings were facing challenges in quoting for new construction bids as their existing processes were slow and at times error prone.

Manual analysis of architectural plans took weeks per bid, causing lost opportunities and unpredictable utilisation of valuable resources.

We developed a bespoke Computer Vision quotation solution that used fine-tuned object detection and OCR models that identified relevant surfaces, calculated area measurements and mapped products to each surface.

The fine-tuned object detection model was trained on a curated set of GA architectural plans which were labelled for specific objects relevant to the manufacturer. In addition, the AI inference server used a real-time inferencing workflow combining object detection, character recognition and visual capabilities of a large language model to achieve the required outcome.

Estimators significantly reduced their time spent on delivering quotes from weeks to days and hours. They were able to refine AI-inferred measurements through a simple UI and exported BoQ quotations and annotated assets with ease matching the clients' specifications.

• 90% model accuracy (using mean average precision metrics), reducing a lot of rework and bid submission errors.
• 12% increase in estimator team's capacity was heavily utilised in supporting additional bids and improving quote response quality.