Iterations with salespeople
The project evolved iteratively in direct cooperation with the salespeople who tested on real materials. Hundreds of small refinements in short cycles: an adjustment, a test on an actual file, feedback.
FENESTRA salespeople received price inputs from two calculation programs: aluminium windows and doors were calculated in Logical, plastic ones in Klaes. The exports were technically correct but unusable for the end customer.
FENESTRA SK is a manufacturer and supplier of aluminium and plastic windows and doors. Aluminium windows and doors are calculated in Logical, plastic ones in Klaes, and prices will continue to be computed in the original calculators.
Logical listed the full price for every item and recalculated the discount (e.g. 54% or 59%) only at the end of the document. Every export had different units, a different structure and different drawing quality, aluminium and plastic went out as two separate quotes, and the look did not match the brand. This was not about replacing the calculators: the goal was for raw exports to become a quote a salesperson can send to the customer without manual adjustments.
The project evolved iteratively in direct cooperation with the salespeople who tested on real materials. Hundreds of small refinements in short cycles: an adjustment, a test on an actual file, feedback.
Drawings as binary EMF/WMF data in RTF, hidden bookmarks, positions with a slash, long texts in parameters: the first weeks belonged to the parser, not the visuals. At the same time we added the content that decides whether a quote is accepted: service, delivery times and the answer to why exactly this solution.
We ran it on 607 files from the archive of past quotes (34 projects). We found format variants that would never surface on a single sample file. Only then did we introduce the new structures.
Pagination to A4, margins, repeating headers and table clipping were tuned in small steps, each with the option to revert. PDF generation via headless Chrome (Puppeteer) was tested while preparing for the server.
AI parsing, Puppeteer, A4/PDF
From a raw export from Logical, Klaes or a partner a unified quote is created without manually recalculating items. The user does not depend on automation alone: they see the source, the result and the differences, and can fix them.
Checking is part of the product. The application verifies totals against the source, flags suspicious items and has a human evaluate them in a panel next to the PDF preview. A drawing is never generated without a real source and a number from our own template is never presented as data of a specific quote.
Input quality was the hardest part of the project. The first weeks went not into looks but into reliably reading the formats, from binary EMF/WMF drawings inside RTF through hidden bookmarks to processing large files in batches with a progress indicator.
batch test on a real archive of quotes
Logical, Klaes and an external quote
from raw export to finished quote


Today a source export from Logical, Klaes or a partner turns into a unified quote without manually recalculating items. The solution is verified on a real archive, the user keeps control over the result and the source code lives in a separate repository with a change history.
Look at your processes through the eyes of automation. Tell us what slows you down, or book a 30-minute call right away.

Enter what your company does and AI generates 5 concrete ideas where automation saves you the most time. It takes a minute.
Generate 5 AI ideas"We see Invelity as a team of professionals who are always willing and able to help quickly with solutions across many areas. Today we cannot imagine our digital marketing without this partnership."

Tell us about your process. We will help identify what can be automated and where to gain efficiency.
Let's go!