A data-driven approach to finding the truth
Azimo exists to make international money transfer cheaper, safer and more reliable. Over the last five years we’ve built a vast network that can move much-needed money into the bank accounts, cash pick-up locations and mobile wallets of people in almost 200 countries.
The scale and complexity of this system brings several challenges. Every bank has a different tech stack, team and view of how payments should work. Some banks’ systems are more up-to-date than others. That makes it difficult to give users one of the things they want most: a reliable idea of when their hard-earned money will be delivered.
What was the problem?
In the old world, we calculated delivery times through a limited data set and some averages. Despite 80% of our transfers taking less than 30 minutes, the delivery times we stated didn’t take into account a bunch of complicating factors, such as:
- Public holidays
- Business hours
- Weekends
- Timezones and daylight saving time
- Speed of local payment settlement systems (most countries have more than one)
- Operating hours of clearing
- In some destination countries, even the opening hours of the local regulator whose system is required to review each transaction
We were also blind to system failures and maintenance downtime. Sometimes a partner’s maintenance window would be known and predictable, but often it was not. With so many partners all over the world we struggled to find out what was causing the issues.
The result? Unhappy customers and complaints about delays. In addition, several customers politely informed us that we were far faster than we claimed to be. We knew we had to improve the quality of our information.
How did we solve the problem?
Thanks to sophisticated tracking, Azimo has a vast number of data points about how fast money travels around our systems and those of our partners. We realised that if we could apply statistical analysis to all that data over time, we’d have far greater insight into the anomalies and seasonal factors that affect our performance.
So began the DTI (Delivery Times Intelligence) statistical analysis project. After launch, DTI immediately started to spot patterns that it would have taken a human being years to identify manually.
In Poland, for example, DTI identified that Friday afternoons are a particularly bad time to send money. This allowed us to remove the static, “usually delivered in under 3 hours”, message for Polish customers and replace it with dynamic times based on the latest data.
DTI crunches the numbers at regular intervals throughout the day, giving us 24/7 coverage. As I write this post, I can tell you that Azimo is sending money to Nigerian bank accounts in under 14 minutes — no wahala!
But DTI isn’t only for client side information, it’s a part of our monitoring system. Thanks to DTI’s pattern detection, we are able to identify the source of an anomaly, whether it’s our own infrastructure, a partner or the receiving bank. The Azimo operations team can then proactively communicate with our partners to let them know about the issues.
What next
We built DTI for better anomaly detection. So far we have used statistical methods with fixed boundaries, but the next step is real-time monitoring. Our goal is to react to anomalies before the customer even notices that there’s a problem.
The final stage of the project will be to use machine learning or deep learning models to predict which transactions could become an anomaly. Thanks to DTI, Azimo will be able to take its customers back to the future.
Azimo is hiring! Product Manager, Apps and ML: https://azimo.workable.com/j/B02EEAD5DD
