Use this article when you are bootstrapping a new shop in GelatoConnect Estimator and have machine lists, paper catalogs, or category notes you want to import without typing every row by hand. After this you will have substrates, presses, and finishing machines staged in Pending Changes from the AI Configurator chat, reviewed, and applied to live setup.
Steps
1. Open AI Configurator
Open Estimate Setup. The AI Configurator chat dock opens automatically the first time you visit. If it is collapsed, select the AI Configurator pill at the bottom of the screen to expand it.
For a fresh import session, select New conversation in the chat header. This clears prior context so the importer does not mix old answers with new files.
2. Describe what you are importing
In the chat input, say what you are importing in one sentence — for example, "Sheet-fed digital presses and commercial uncoated stocks", or "Saddle stitcher and perfect binder for our new bindery line". The chat uses this description to set extraction expectations.
Drop the source files into the chat window. The drop target appears as Drop files here during drag-over. Accepted formats include CSV, XLSX, and PDF.
Wait for processing. The progress strip shows the current step (uploading, parsing, extracting, summarising). Processing continues even if you navigate to a different page in Estimate Setup — the chat dock remembers state per conversation.
3. Review extracted rows
When extraction completes, the chat surfaces a summary of what it found. Open the summary and check three things:
New rows — entities the importer wants to create. Confirm the names match what you expect (no garbled rows from a bad PDF column).
Updated rows — entities the importer wants to modify because it found a match against existing setup. Inspect changed fields before staging.
Removed rows — entities the importer wants to delete. The most common source of false positives — confirm each one before staging.
Drill into individual rows by selecting the row link. The detail view shows every field the importer extracted; fix obvious parser errors (wrong gsm, swapped width/height) in the row editor before staging.
4. Stage in Pending Changes
Once the review is clean, select Stage changes in the chat. The chat closes the extraction and the staged rows appear in Pending Changes in the sidebar. The Pending Changes count increments by the number of rows you staged.
If Pending Changes still reads zero after staging, re-open the chat conversation and check whether the importer logged an error on a row you missed. Common causes: a substrate weight outside the validation range, a press tag the importer could not match.
5. Apply Changes
Open Pending Changes in the sidebar and review the staged rows one more time. The list groups create, update, and delete operations so you can sanity-check the shape of the import.
Select Apply Changes to publish the staged rows to live setup. Confirm the dialog.
The Pending Changes count returns to zero and the imported substrates, presses, and finishing machines appear in their respective tables in Estimate Setup. Spot-check Print Machines and Substrates to confirm the key rows landed.
6. Validate with a quote
Create a quote that uses one of the imported substrates and presses. If the imported configuration is wired correctly, the quote calculates and the new rows appear in the route information. If the quote fails or the route lists the new press under Excluded machines with a "no machine matches tags" reason, the import is missing tags — open the press record and add the tag the substrate carries (see Configure tags for machine compatibility).
Things to know
The AI Configurator chat does not bypass review. Every imported row lands in Pending Changes, not live setup. If a row looks wrong, discard it from the queue before applying — applied rows can only be reversed by editing or deleting them after the fact.
The chat conversation persists per shop. If you start a second import the next day, the chat already has context from the first one. Use New conversation to start clean when you do not want that context.
Large files (full multi-vendor paper catalogs) are processed in batches. The progress strip names the current batch so you can tell which subset has finished extracting.
AI Configurator cannot invent values your source file does not contain. If a press lacks a click rate in the source, the imported row will lack it too — fix the row in the editor before applying, or the press will be excluded from digital quotes for missing click cost.