On the Realities of the Field Limiting Optimal ATM Operations
Advances in Data Science enable software solutions in almost all angles of ATM operations; from location selection to cash replenishment, from monitoring to route planning, even predicting failures before they happen. The machine learning algorithms excel in prediction and optimization.
However, the people in the field know. When you get out to the real world, there are so many parameters and requirements that needs to be addressed before one can utilize the results taken in the simulation domain and make something of it. Here are a few examples of some of the many real-life limitations that needs to be tackled:
1 - Fleet Size
When ATM Cash replenishment needs to be optimized, it is a trade of between idle-cash left in the ATM machines vs. the number of replenishments. High frequency of replenishments increase the CIT costs, whereas low frequency of replenishments increase the idle cash cost.
The aim of cash replenishment optimization solutions is to find the optimal point. However, the idle cash cost depends usually on the interest rate in the country. In cases where there is a sudden interest rate increase, the equilibrium point will shift towards a higher number of CIT stops. However, the CIT company may not have additional 20 armored trucks ready at hand. Therefore, a sub-optimal point should be utilized for the operation at least until the CIT company increases the fleet size.
2 - ATM Capacity
The opposite can be observed in countries where the interest rate is very low, or the real value of banknote is very low.
In this case, the optimal point can be depositing much more than the available cassette size inside the ATM.
The sub-optimal solution will be filling the ATM in each case, or locating multiple ATMs side by side, and not replenishing them until all of them becomes critical.
3 - Location Feasibility
Selecting an ATM location with high number of transactions or high number of DCC transactions is always preferable. However, there are various criteria in order to enable the location as a potential installation site.
In some countries road access, and electricity grid availability may be an issue. In some others, an IAD might be looking for a supermarket, or a gas station to place an ATM. Therefore, the optimal locations are limited to a selected few.
4 - ATM Security
Perhaps, the hardest part of the CIT operations is the unplanned visits. Consider the case, where there are many trucks on the field and suddenly an unplanned order arrives. If everything was inside a simulation toolbox, it would be a rather easy problem to re-optimize the routes. However most of the ATMs today do not have electronic locks. They have keys at the cash management centers (CC).
Therefore, a truck from the field needs to go back to CC, get the keys and then go to the ATM to be serviced, without missing the original SLAs for the route. In order to overcome this, usually CIT companies keep a spare vehicle at the cash center, or may give the keys of the close-by ATMs to the truck's original morning route. Both cases are either cost or security inefficient.
These are just few examples of what happens in the field. But there are many more, and operations people know. When the ivory towers of software and algorithm world are left for a few days and time is spent in the field, one appreciates. The outcomes and the experience though, strengthens the analytical solutions to a great extend.
Please do comment and share as well if you had interesting experiences in the field and methods that you have overcome to handle the daily needs of the ATM operation.