How Our AI Matches Transactions
Reconcilify uses a combination of rule-based matching and machine learning to automatically reconcile transactions across your systems. Here's how it works.
The Matching Engine
Our matching engine combines multiple approaches to achieve high accuracy:
Rule-Based Matching
Deterministic rules that match on exact amounts, dates, and identifiers. These rules handle straightforward 1:1 matches with 100% confidence.
Fuzzy Matching
Handles variations in names, dates within tolerance windows, and split transactions where amounts don't match exactly.
ML-Powered Scoring
XGBoost model trained on medical spa transaction patterns evaluates 28+ features to calculate match confidence scores.
Semantic Embeddings
BERT-based text embeddings compare transaction descriptions to identify matches even when wording differs between systems.
Confidence Scoring
Every suggested match includes a confidence score from 0-100%. Here's what the scores mean:
High Confidence
Multiple signals align. Recommended for auto-approval in your settings.
Medium Confidence
Some signals match but require human review. Flagged for your attention.
Low Confidence
Signals conflict or are missing. Manual matching required.
Matching Signals
Our model evaluates these key signals when matching transactions:
Transparency & Human Control
Rule Traces
Every match shows exactly which rules fired and their individual scores. You can see why a match was suggested.
Exception Handling
Transactions below your confidence threshold are always flagged for human review. You maintain final approval on all matches.
Continuous Learning
When you approve or reject matches, the system learns your preferences to improve future suggestions.
Important Note
AI-powered matching is a tool to assist your reconciliation workflow, not replace it. Our system is designed to handle the routine matches automatically while flagging anything unusual for your review. You always have final control over what gets approved.
Ready to see AI-powered reconciliation in action?
Try our interactive demo to see how the matching engine works with realistic sample data.