Reduced double-counting errors in AI extraction
We've significantly reduced double counting issues by adding an additional AI review layer to validate extractions before totals are calculated.
We identified an issue where, in some cases, the AI would count the same costs twice. This typically occurred when vendor quotes varied heavily in structure, which sometimes caused levelled totals to calculate incorrectly.
To address this, we introduced an additional AI validation step that reviews extractions before totals are finalised. This secondary check has reduced double counting occurrences to single-digit rates, and we are continuing to monitor and refine the remaining edge cases.
Excited about this release?
Share it with someone who'd be interested, or reach out to us.
