Joutsenon Elementti Utilized Artificial Intelligence and Google Cloud in Order Processing
AI in Order Processing – Utilization of Computer Vision Played a Central Role in Analyzing Technical Requests for Proposals
Joutsenon Elementti Oy is a Nordic expert in concrete elements and a manufacturer of high-quality elements. Nearly a hundred employees work across three Joutseno Element factories.
What was the challenge?
In the order processing at Joutsenon Elementti, a significant and time-consuming phase was extracting data from technical drawings in PDF format. Despite clear industry standards, technical documents arriving from various design offices often came in vastly different formats.
Technical specifications, such as the strength class of concrete used and stress classes of concrete elements, were critical information for manufacturing the right kind of concrete element. Another challenge was ensuring the correctness of the collected data format. The gathered information had to adhere to technical standards in the concrete industry.
Previously, Joutsenon Elementti experts searched for information from images in the documents and manually inputted them one document at a time into the production management system. This was both time-consuming and susceptible to human errors.
The challenge presented by Joutseno Element to Codento was to explore whether this process could be reliably automated using modern AI tools.
Our Solution
Codento demonstrated to Joutsenon Elementti the use of Google Cloud, particularly the Document AI tool, in demanding text recognition. Document AI is a no-code text recognition model (OCR, optical character recognition) developed within Google Cloud that can be trained with very little data.
At its minimum, for a standard form, a reliable OCR model can be trained using Document AI with less than ten examples.
In this case, the examples were significantly more challenging. Layouts in technical drawings vary from one engineering office to another. Therefore, it couldn’t be assumed that technical drawings were standard forms. However, training the model only required about 30 technical drawings, enabling rapid progress.
Why Codento?
Codento was chosen for collaboration based on a clear proposal, extensive and diverse AI consulting experience. Codento’s previous similar implementations also positively influenced the decision.
Surprisingly Positive Reliability of the OCR Model
The reliability of OCR models is generally measured by F1 scores, which range from [0…1], with zero being comparable to tossing a coin and one being perfect accuracy. F1 scores are a kind of average of the model’s precision (how much of the results are correct) and recall (how much of the data is found). With the model developed by Codento, the F1 scores were nearly 0.9. For some sought-after information, the scores were even significantly higher.
As a conclusion in the Proof of Concept (PoC), it was feasible to automate the extraction of data from PDF files and input them into Joutsenon Elementti’s own systems using Google Cloud’s modern OCR tools.
In Customer Own Words
“Codento succeeded in a short time in solving a practical challenge that had been troubling us, leveraging their expertise, artificial intelligence, and the capabilities of Google Cloud. We gladly recommend Codento for similar assignments aiming to automate sales and order processing.”
Teemu Liimatainen, Sales Director, Joutsenon Elementti
Additional Information
- Document AI: https://cloud.google.com/document-ai?hl=fi
- Examples of technical drawings for concrete elements: https://www.elementtisuunnittelu.fi/suunnitteluprosessi/mallipiirustukset
Contact us for more information on using computer vision:
