About Transaction Matching Assistance

Transaction Matching Assistance predicts potential matches for unmatched transactions in Transaction Matching.

Machine Learning algorithms analyse patterns in your historical manual match data and use these patterns to train a prediction model for a match type. The prediction model can then be used to predict potential matches for the match type. Both the training and predictions are done for each match process within the match type. The prediction model can be retrained at regular intervals to ensure that it adapts to changing patterns and behaviors. Detailed log files provide insights into the training and prediction processes. This information enables you to understand the system’s behavior and use this to monitor and optimize performance.

Note:

It is recommended that you retrain the prediction model periodically, especially after significant data changes.

Considerations When Using Transaction Matching Assistance

• To train match predictions, there must be a minimum of 2500 manual match 1 on 1 transactions for the match type.

  When there are more than 2500 manual match 1 on 1 transactions for the match type, the most recent 2500 are considered during training.

• Governor limits are used to ensure optimal performance for large data sets.

  If a profile has more than 500,000 unmatched transactions, either in the source system or subsystem, prediction is skipped for that profile.

• Only one Train Match Prediction job can be run at a time. If multiple jobs are submitted at the same time, the remaining jobs are queued until the current job completes.