Extracting the required data from the puchase order, price lists and other product data to produce a traceable form for a review
Company
Luvata oy
Industry
Manufacturing & Supply Chain
Problem
Luvata manufactures high-quality copper products, such as wires, tubes, conductors and customised components for industrial customers.
Purchase orders arrive by email or portal upload, primarily as PDFs and sometimes with accompanying bills of materials. Each customer produces purchase orders from its own system, resulting in different formats and layouts. Operations staff must manually extract and interpret the required products or materials, product codes, parameters, and volumes. The current manual process involves extracting top-level information and item-level details from purchase orders, and then extracting item-level supplementary details from the accompanying bill of materials. They then match this information against internal material records and price lists, apply relevant fees, and prepare a sales order and order confirmation with verified details and correct prices.
Customer-specific formats and pricing, documents with complex layouts with arbitrarily located information, and frequent exceptions make this knowledge-capture process repetitive and time-consuming, increasing the risk of human error and the workload of sales staff. The business need is to spend less time processing purchase orders, eliminate human errors, and reduce salespeople’s workload.
Solution
Sales staff upload the customer’s purchase order and Luvata’s own price list to the tool. The AI extracts the required products, their codes, parameters, and volumes from the purchase order.
The solution uses reusable components from the GAIK GenAI Toolkit to process the documents. A parser first reads the document and its structure, after which AI-powered data extraction converts the relevant content into structured information with confidence scores and flags for human review. For documents where the layout is important, visual AI can read and extract the information directly. The structured data is then used to match products and calculate prices according to Luvata’s pricing rules.
The interface presents the extracted information and price calculations in a traceable form for review. Using a human-in-the-loop approach, sales staff verify the results before approving the final output.
Further development towards a production system includes integration with Luvata’s existing systems, such as direct access to current pricing information, order confirmation generation, and data transfer to the ERP system.
The solution was co-developed by Luvata and Haaga-Helia University of Applied Sciences within the GAIK project. Luvata contributed practical requirements, test materials and feedback, while reusable GAIK GenAI Toolkit components were configured and combined for the company-specific pilot.
Outcome
The pilot successfully demonstrated the feasibility of the solution and provided a clear basis for continuing development towards a production system. The tool was tested with more than 30+purchase order cases. It worked particularly well with purchase orders from one key customer, while testing with two other customers identified opportunities to tailor the solution to variations in customer-specific input data.
The tool was considered simple and easy to use, receiving an ease-of-use rating of 5 out of 5 from both company representatives. Its interface provides traceability and makes the extracted information and price calculations easier to verify.
Because errors in purchase order processing could lead to financial and reputational consequences, the solution uses a human-in-the-loop approach, where the system facilitates the sales staff in reviewing the extracted order details and price calculations before approving the final output. As accuracy becomes more consistent across different customer formats, the amount of manual review can be gradually reduced.
Overall, the pilot achieved its purpose. The concept was considered feasible, and further investment towards an MVP was rated as definitely worthwhile.
ROI
Quantitative ROI has not yet been measured. The expected return would come primarily from reducing the time and manual work required to extract and interpret purchase order information, identify product codes, verify order details and calculate prices.
Further value is expected from fewer data-entry and calculation errors, more consistent pricing and a reduced workload for sales staff. These benefits would be realised more fully as the solution achieves consistent accuracy across different customer formats and is integrated with Luvata’s pricing information and ERP-based order confirmation process.
Technologies
- AI: Natural Language Processing (NLP)
- From the list: Azure
Free text: OpenAI’s, Anthropic’s, and Google’s LLMs for extracting structured information from purchase orders and BOM documents, GAIK schema generator, GAIK parsers, GAIK data extractor, cross-document information extraction, structured-output generation.
Partners
Haaga-Helia University of Applied Sciences: primary development partner; co-developed the company-specific pilot and reusable GAIK GenAI Toolkit components. Tampere University: conducted an AI adoption readiness assessment.
Process Impact
- Sales & Business Development
- Supply Chain & Logistics
Languages
- Finnish
- English