01One job
Upload the request
A user uploads the PDF, adds the customer's name and starts a job. The document becomes the starting point for a structured workflow in one application.
- PDF upload
- Customer
- Job
Case study / Swiss HVAC manufacturer
From three-day waits to same-day RFQ responses.
A Swiss HVAC manufacturer needed to turn long customer documents into usable product requirements. Werkon built an AI application that finds the relevant sections, extracts the requirements and matches them to the product database. People review and approve both stages. Reported average processing time fell from more than eight hours to under 30 minutes, including human verification.
The workflow
The right requirements, without reading every page by hand.
An incoming RFQ could mean a PDF of around 300 pages from an architect or builder. A six-person team had to find the relevant section, extract requirements, interpret unfamiliar product names and prepare ERP entries. Before anyone could respond, hours had already gone into understanding the request. We built a complete web application around that work, with a product database, jobs, users, settings and audit logs.
02AI extraction
Find and extract
The application breaks the document into chunks. AI locates the relevant sections and extracts the required product data into a structure aligned with the manufacturer's ERP.
- Relevant sections
- Required fields
03Human verification
Verify the requirements
Reviewers inspect and edit the extracted values in a table. Product matching starts only after verification, so the next stage works from reviewed requirements.
- Inspect
- Edit
- Verify
04AI matching
Match to real products
AI matches the verified entries against the product database. Each product's keyword field supplies alternative names and phrases that may appear in customer documents.
- Product database
- Keywords
05Human approval
Compare and correct
Reviewers can use the standard matching view or switch to Compare, which shows the fields that determine the product. They can edit values directly in the application.
- Matching view
- Compare
- Edit
06Structured output
Prepare ERP-ready data
The matched entries are structured in the format the manufacturer needs for its ERP workflow. AI handles extraction and matching; people retain approval of the result.
- ERP format
- Reviewed entries
Inside the application
Built for the people who check the work.
The useful interface is the one that makes a questionable entry easy to inspect and correct. These four application views show where reviewers work, from the source document to product comparison and the vocabulary used for matching. Select any screenshot to take a closer look.
- 01
Extraction review
The extracted requirements remain editable before the reviewer moves the job into matching.
Source document and extracted requirements, side by side. Enlarge screenshot: Extraction review - 02
Product matching
The standard view brings proposed product matches into a table the reviewer can inspect and edit.
The standard results table, with structured product fields. Enlarge screenshot: Product matching - 03
Compare view
A focused view shows the fields that determine the product, helping the reviewer check the proposed match.
Matched and unmatched entries remain visible for review. Enlarge screenshot: Compare view - 04
Product keywords
Reviewers can add an unfamiliar product name to the keyword field so it is available as context for later matching.
The Synonyms field stores alternative product names. Enlarge screenshot: Product keywords
What made the difference
Faster processing. People still make the final call.
Werkon's delivery account reports average processing of under 30 minutes per RFQ pack, including human verification, compared with more than eight hours before. Customer responses moved from roughly three days to the same day. These are reported operating figures; a formal measurement period and sample size have not been supplied.
- A small field solved a difficult problem
- Product names made matching difficult: documents often used different names for the same product. An editable keyword field made that terminology explicit. The team can maintain the mapping themselves, and AI receives those keywords as matching context.
- A different way to spend the day
- Four people now handle the workflow that previously involved six. Alongside faster responses, the reported improvements are fewer errors and easier verification. No quantified error-rate reduction is claimed.
- Three months, built with the customer
- We began with repeated conversations to understand the scope and working process, then held monthly progress meetings through delivery. The result was a complete application shaped around the customer's actual RFQ work.
- In daily use, owned by the client
- The manufacturer uses the application every day and owns the solution. Werkon remains available for support and updates the AI models under an agreed maintenance arrangement. The client pays for and controls the system.
[ WORKFLOW / SYSTEMS AUDIT ]
THE FIRST ENGAGEMENT
Start with one real workflow
A Systems Audit is the usual starting point. If the opportunity is already clear, we can move directly into a focused build.
Show Us the WorkflowStart with the free automation readiness checklistOBSERVEQUANTIFYDECIDEBUILD
Continue with the work
Choose the next useful path.
Stay with the operating question. The technology can wait until the work is clear.