The problem

FX Trade Processing Transformed with Machine Learning

ANZ Bank (Australia and New Zealand Banking Group) is a multinational financial services institution and one of Australia’s largest banks.They needed help to improve the manual, error-prone process to reconcile Foreign Exchange (FX) trades each day.Reason created a secure automated workflow, applying Artificial Intelligence (AI) to evaluate the emailed trade requests and Machine Learning (ML) to review and match requests to executions. In addition, chat bots flagged any unmatched trades for the relevant person to review and resolve.

Reason did

Product & workflow design

System architecture

Data strategy

AI automation

DATA MANAGEMENT

ML training

Full-stack engineering

Symphony chatbot integration

PHASE ZERO

Establishing the workflow

Reason validated the workflow concept for the robotic process automation with an interactive prototype in just 4 weeks. During this time, we designed the approach, target architecture and project delivery timeframe.

Phase One

Automating Workflows with Chat Bots:

The early focus created the environments, internal tooling and processes to allow the team to rapidly iterate over future releases.The initial release delivered basic matching functionality using Artificial Intelligence (AI), and a newly built Web App displayed the status of each trade.

50% auto-match rate on FX trades, with links to supporting documentation
TradeButler web application and chatbot fully integrated into daily workflows to flag the relevant people and resolve unmatched trades

Phase Two

Ingesting and Interpreting multiple data sources

Applying regression analysis to new data sets helped increase the automatic trade match rate to 70%, reducing the need for manual reconciliation.The reconciliation process for non-matched trades was aided by new chatbot and WebApp functionality, enabling inter-departmental communication and feedback.

70% auto-matching of FX trades 
75% faster resolution of non-matched trades using Web App and Chatbot to flag, review and resolve

Phase Three

Driving Match Rates with Machine Learning

To truly deliver on the promise of automation, new ML models (prodi.gy) used entity recognition and prediction based recommendations (regression analysis), to automatically and reliably ‘decide’ whether or not to auto-match a trade, improving match rates above 90%.Over time, the match rate increased as the models continued to ‘learn’ how to match trades using the available data.

90% of FX trades automatically matched 
Manual processing time reduced by 75%

The outcome

Secure and Effective Solutions

Our architectural approach balanced a need to adhere to internal policies (using an internal hybrid cloud platform) whilst allowing stakeholders the flexibility to experiment and adapt.
Docker containerisation solutions split operational from developmental concerns, while an event-driven micro-services architecture decoupled new feature developments (e.g: adding new data sources and model training) for smoother deployments.

Symphony’s Customer Innovation Award 2018

90%

automated trade match rate through AI & ML

4 wks

to deliver prototype and prove automation process

75%

time saved on manual reconciliations

case study - anz

Automating Trade Reconciliation Using Artificial Intelligence