Reconciliation is an accounting process that uses sets of records to ensure financial figures are correct and in agreement between multiple parties. In the financial services domain, trade reconciliation refers to a set of post-trade activities (typically occurring T+0 or T+1) related to identifying and resolving trade breaks which can occur between a buyer and seller of a security (or between a borrower and lender). The vast majority of trades are executed, confirmed, and settled automatically with no issues.
However, for various reasons, trades can fail, meaning the values in the trade don’t match between the counterparties. Failed trades don’t complete the journey through the straight-through processing system and are grouped together, waiting to be settled manually. Reasons for trade breaks include mismatched trade values (price, rates, amounts, ids), incorrect accounts, insufficient funds, etc.
The trade operations and support team (middle/back office) is responsible for reviewing all the executed trades and addressing any trade break errors, or exceptions. They then send that list to the trading execution desk to gather information on the trade breaks to assist with resolution. Once the breaks are resolved, the trade data is updated to correct the breaks, and all systems are updated to reflect these fixes. This process is predominantly manual, involving staff at both firms on each side of the trade.
For a large financial services firm that does hundreds of thousands of trades a day, even a small percentage of trade breaks will involve hours of manual effort to reconcile. The vast majority of the costs associated with trade settlement involve exception processing of rejected trades by middle office and back-office staff. Artificial Intelligence (AI) can be of enormous benefit in trade reconciliation.
Working as an AI partner with a client in the post trade space, nova IQ created a deep neural network which was trained on historical counterparty breaks and matches. That AI model was able to identify patterns, gaps, tendencies, and behavior from the historical data which would never be apparent to a human. Not only did the model “learn” from the data, it was constructed in a manner to be able to output a prediction of ongoing breaks with a high degree of accuracy. This model was wrapped in a RESTful API and integrated into the client’s product, allowing all of their clients’ access to break resolution predictions generated by the AI model. This alone doesn’t resolve the break, but it does point the middle and back office staff in the right direction for resolution.
The business impact of the integration of the AI model to their product is profound:
The future of AI in reconciliation will enable further automation of the post-trade process. Very soon, firms will accept the AI trade break recommendations automatically, making it an integral part of the straight-through process.