Six months ago we opened early access to 30 back-office teams. We thought we knew what they'd use Parseloom for. We were mostly right about the what - invoices and bills of lading were the dominant use case, as we expected. We were wrong about the how.
What surprised us: the integration comes last
We expected teams to start by connecting to their ERP. Nearly every team started by verifying extraction accuracy on their specific document types first, then worrying about integration. They wanted to see the extracted JSON before they'd invest time in plumbing.
This shaped how we ordered the product onboarding. The quickstart now puts extraction first, with a raw JSON response you can inspect. Integration configuration comes in step three.
Templates needed more flexibility than we built
Several teams had document types where the layout varied significantly across vendors - not different document types, but the same document type with different field positions depending on who sent it. Our first template model assumed one layout per template. That assumption held for about 60% of use cases.
We rebuilt the template matching layer to handle layout variants under a single template definition. That work shipped in May. The teams that were hardest to onboard in January are now running cleanly.
Month-end is a real spike, not a metaphor
We monitored extraction volume per account across the six months. The month-end spike is real and large - typically 3 to 4 times mid-month daily volume for the AP-heavy teams. We tuned our processing queue for that pattern.
The teams that got the most value in early access were the ones that used Parseloom to flatten that spike, not just to speed up the daily baseline. That's where the ROI math works out clearly.
What we're still working on
Handwritten annotations on printed forms remain a hard problem. About 8% of BOL documents in early access had handwritten notes in margin fields. Extraction confidence on those is lower than we'd like. We haven't shipped a fix for it yet - it's on the roadmap but not a Q3 commitment.
Custom output schemas also came up repeatedly. Teams want to define the exact field names and structure that matches their ERP's import format. We have a basic version now; a more complete schema builder is coming in the next major release.