Cash Forecasting is the science of precise forecasting of transactions for cashpoints such as ATMs and Branches. Arute Solutions is the leading cash optimization company in Turkey, with its Cash+ product in charge of more than 8000 ATMs. Naturally, forecasting ATM transactions constitutes a crucial part of Cash+ solution. The forecasting task presents a unique set of challenges. Below, we will examine three of them in-depth to give you a glimpse into these challenges.
1- Multiple seasonalities in cash forecasting:
Cash Forecasting requires the handling of various seasonalities of different frequencies. ATM transactions are sensitive to hour of the day, day of the week, day of the month, and month of the year! Moreover, these sensitivities change drastically from one ATM to another. For example, whereas one ATM can be located in a business district and thus can be very sensitive to hour of the day and day of the week, another ATM could be located in a holiday-favorite town and might only be sensitive to the month of the year, but not so much to the day of the week! At the figures below you can see the two examples of such ATM’s:
Additionally for cash forecasting, these seasonalities can appear and disappear without warning! For example, below is an example from an ATM we manage, where suddenly a new monthly seasonality appears due to a new wage payment agreement between the bank in question and a large company:
Our algorithms are on the watch as seasonalities appear, disappear, and interact, to provide a seamless service to ATM customers.
2- Predicting trends in cash forecasting:
How people use ATM’s to obtain cash is very sensitive to trends at levels varying from bank policies to macroeconomic markers. Cash forecasting is the art and science of predicting these trends. For example, below you can see the effects of a bank implementing a drastic new policy of decreasing clerks’ workload by increasing the maximum withdrawable cash from the ATM’s:
On the other hand, below you can see the sudden trend appearing in cash transactions due to the volatility of USD/TRY parity, but as opposed to the previous example, returning to previous levels over time:
Handling trends of different nature is an important part of cash forecasting!
3- Peak days and outliers in cash forecasting:
A crucial part of cash forecasting for ATM transactions is handling extreme values. Handling such values require extreme care. A value observed as such can be due to a one-off event, for example a large concert happening in a venue close by. On the other hand, however extreme, it could be part of a pattern, such as the pay day of a nearby large company.
To make matters more difficult, the payday could be shifting from month to month due to company’s weekend / weekday payment policy, or national holidays. Therefore handling extreme values in ATMs to improve cash forecasting requires much more than a naive outlier analysis or seasonality detection!
Let us examine such shifting of such an extreme value at the ATM below:
Here is an ATM near a company that has the 5th of each month as the pay day. However, since in the May 5th, 2019 is a Sunday, the company has shifted its payday to the following Monday. In such conditions, some other companies move the payment to Friday! We detect such patterns automatically, and make sure with our cash forecasting methods that the customers of the ATMs we manage are served flawlessly.
Above we have touched upon some of the challenges presented by cash forecasting. Others include, but are not limited to, the effects of exogenous variables such as the weather and ATM position, handling erroneous data due to system failure on the banks’ side, or regularly shifting seasonalities due to non-Gregorian calendar holidays.
If you would like more information as to how Arute Solutions handles these challenges (and more!) in cash forecasting, successfully by combining machine learning methods and domain expertise, please contact our sales team at email@example.com.
Any questions? Contact us to learn more!
Arute Solutions Data Science Team