Effectively managing potential trade settlement fails—when either the buyer does not deliver funds or the seller does not deliver securities in an appropriate form on settlement date—is a key element in driving efficiency across the entire securities market ecosystem.

Even though most trading operations are driven by automatic STP processes today and many banks have made massive investments in improving the efficiency of information systems, there are often still large volumes of trades that require manual human intervention―which is time consuming and highly prone to error.

Luckily, emerging technologies such as artificial intelligence (AI) and machine learning, along with higher computing power, are allowing banks to leverage vast amounts of data and develop systems that make managing potential trade settlement fails much more efficient from a cost, time and quality perspective. This could be done in a process we like to define as “Analyze -> Diagnose -> Propose”.

The guide to developing a better settlement fail system


A first step in managing trade settlement fails is to analyze large quantities of historical data to identify complex patterns. These data patterns feed algorithms, which are then used to develop more accurate data-driven decisions. When developing a solution for our Accenture Post-Trade Processing business, we used multifaceted machine learning algorithms to develop different decision models. Data sets include key trade features such as trade date, settlement date, counterparty, currency, security, security category, quotation market, custodian, depository and historical fails settlement patterns (including rolling six-month historical data, which is now refreshed every month, and new data that has entered the system since the previous refresh).


Next, on Value Date -1 or Trade Date, all open trades are added to this machine learning model to identify potential problematic trades and the respective probability of trade failure. Once potentially rejected trades are identified, the model diagnoses the possible anomalies, indicates why a trade was rejected and pinpoints applicable gaps. Typical diagnoses could, for example, include a matching issue due to incorrect booking or reference data setup, a funds issue or an insufficient security position.


Lastly, when built out fully, the AI/machine learning models could not only predict a potential failure and identify the point of error, it could also leverage insights from the learning data sets to propose a solution.

Using such an approach, AI and machine learning can improve the accuracy of trade settlement fail prediction by continually learning while analyzing, capturing trends and identifying patterns. This is a key differentiator for using artificial intelligence for such an industry issue over traditional approaches, and self-learning can be accomplished with very minor to no new coding.

In our own examples, we have seen impressive results. Our prediction algorithms and data sampling techniques are returning very positive results: The accuracy of fail prediction is ranging from a modest 83% to a very accurate 97%, depending on the data samples and learning techniques used.

Back office operations could leverage such predictive tools multiple times during the day to gain extra validation and take appropriate actions on any prospective failed trades well in advance, accelerating fails management efficiency. To further streamline the workflow, we have added an interactive visualization tool that provides multiple views―including a summary view of critical markets, clients and asset class.

In short, AI and machine learning can analyze complex sets of data, identify failed trades and provide the reason for the fail―along with prescriptive action. They offer increased efficiency, insight and accuracy in managing potential failed trade settlements that could make a significant difference in preventing investment banks from suffering costly consequences such as market and reputation loss, financial penalties or regulatory violations.

Want to learn more? Leave a comment or contact me directly at dean.l.jayson@accenture.com.