The AI-driven custom fields are the standout capability. I defined a classification field that determines whether an invoice represents plan spend or overage spend, and it has returned the correct value on every document processed. Combined with the auto-rename templates, each processed invoice lands with the date, vendor, total and spend type already in the filename, so the document library itself becomes a usable report without any downstream tooling.
Implementation was fast. I went from sample document to mapped fields to a live Gmail integration pulling matching invoices automatically inside an hour, with no developer involvement. The Gmail integration with subject filtering does the collection work I would otherwise do by hand every month.
Reliability has been excellent — zero extraction errors to date across every document I have run. I use it daily.
Support deserves specific mention. I raised a genuine edge case around re-importing previously deleted documents; it was diagnosed accurately, fixed in product over a weekend, and returned to me with clear reproduction steps. Review collected by and hosted on G2.com.
My reservations are about the interface, not the engine.
Reprocess is exposed in at least three separate places (document view, top right, and on AI pick-list fields) and does not always trigger on the first click. Bulk actions are inconsistent: bulk reprocess is available under All Documents but absent from the Extracted Data panel, which offers export, move and delete instead — I would consolidate bulk actions onto Documents only.
Navigation is the weakest area. There is no clear route back to the parser and no top-level menu; deleting a page while viewing a document leaves the view in a confused state rather than advancing to the next document, and Back does not behave as expected.
There is no way to abort a running Gmail import or halt processing once started, which means a misconfigured run has to be allowed to complete while consuming credits. Related: there is no reset for email imports after deleting previously imported records, short of reconfiguring the integration.
Auto-rename templates accept any text without validation, so a mistyped variable is only discovered later in the output filenames; highlighting recognised variables as valid would remove that risk entirely.
Finally, the settings taxonomy is not intuitive. Document renaming sits under Parser Settings > Post-processing rather than the Post-processing section itself, and the additional email settings are not under Integrations where you would look for them. Prompt guidance for AI fields is also thin — a few worked examples and a confidence score on pick-list selections would shorten the learning curve considerably.
None of this prevented me from getting to a working, valuable configuration. It is friction in the first few days rather than a limitation of the product. Review collected by and hosted on G2.com.