The average ATM contains between 1 and 3 million rubles inside. They are arranged in 4 cassettes, each of which contains banknotes of its own denomination. Replacing cartridges is similar to replacing cartridges in a printer: the module is removed (and it doesn’t matter how much money is left there - collectors don’t even know this due to security rules), and another module is inserted in its place exactly according to the arrow indicating which side to do it and how to do it.
Collection is an expensive and quite risky procedure from a security point of view, so any ATM maintenance is quite expensive. The natural desire of the bank is to reduce the number of collections. This results in a classic resource allocation problem: on the one hand, there should always be money in the ATM for clients, on the other hand, ideally the bank wants the last bill to leave the box just before collection.
Banks optimize all cash handling processes: this includes the aforementioned modularity of the cassettes (so as not to take out money and count with your hands), and a special arrangement of bags in the machine, exactly corresponding to their order of use along the route, and much more. At the same time, everyone understands perfectly well that if the ATM does not give money, the client will be upset: this is not a beer stall where you can choose a different type or simply cross the road and get the goods from another place. When the client does not receive money from his “native” ATM, he becomes very upset, indignant and begins to lose confidence in the bank.
Accordingly, the main priority is to always have money. At the same time, you want them to be, if not back to back, then with a minimum reserve - and this is where you need very good IT technologies, which can cost very good money.
In principle, analytics has two tasks:
Let's start with a breakdown by banknote . As I already said, there are usually 4 cassettes: of course, two of them immediately go under small ones: under hundred and five hundred. But there are nuances: for example, you don’t have to put in 500-ruble bills at all, but you can put in more than others. Or use two boxes at once for bills of the same denomination. Or don’t put in five thousand, and so on. Here everything depends on the consumption scenario and the issuing algorithm: if this is a student canteen, then collection may be necessary if there are 3 full cassettes with the hundred hundredths of money running out.
The second point is actually how much balance there should be. The scenario for using an ATM is important here: for example, there are exclusively salary ATMs that are located in offices, there are ATMs that are located in shopping centers, and there are ATMs that are located in squares at train stations. Obviously, payroll workers will have pronounced “critical days”; those in shopping centers will have seasonal peaks and troughs corresponding to the general peaks and troughs of retail; in squares and train stations, much depends on fluctuations in traffic density and its nature.
For example, buy yourself an ATM in a business center. On those days when there is a salary in the offices, it should be filled to capacity with large bills so that those who withdraw can withdraw as much as they need. Then, for example, it is filled by a third with bills of lower denomination, and for some holiday or weekend it is filled with more. Based on the analysis of transactions, you can understand the peaks in withdrawal amounts, understand the client profile of this ATM, and find out what breakdown needs to be done.
There are a lot of nuances, but in general, the task largely comes down to solving the problem of providing goods on the store shelf. With all this, there are a lot of security features, for example, a ban on systematically visiting an ATM (using the same routes).
About the ATM:
The first part of the solution is architectural. You don’t need to figure out how to transport money, but first, you need to design and build an ATM network so that it works optimally. This is done based on a bunch of data. Our analytics can plan the ATM network itself so that everything is optimal, taking into account many restrictions - the presence of nearby competitors, traffic on streets, shopping centers, train stations, taking into account seasonal changes, and so on.
Then we must decide how much money and in what denominations should be lying in the ATM. Plus - how often it must be collected.
ATMs can also be replenished by customers, and the money there is collected in a separate cassette, which can also overflow. Money is not dispensed from this cassette: input-output at the ATM in different processes. From an IT point of view, it seems logical to use this money for issuance, but according to the law, all this must first be passed through the accounting department, and for this the cassette must be brought to a bank branch. Therefore, it is beneficial for collection to immediately replace the cassette with money from replenishments (since they have already arrived anyway).
Next, you need to ensure that ATMs are properly filled. To do this, we take a system that allows us to optimize the route from the point of view of minimizing costs and at the same time set restrictions like “routes must change once a week” (except for security services, no one knows which route the collection vehicle will take tomorrow).
Returning to logistics, we have several blocks: first, we need to plan it strategically, second, we need to understand who to collect with what amounts, and so on, and third, we need to optimize the collection schedule, both for individual ATMs and for a chain of ATMs at once. If there are a lot of them, then for each we calculate how to optimally collect it, and then we group them according to routes.
For example, if one ATM needs to be collected once every 2 days, and another - once every 3 days, this will be two trips. If you collect both at once, then it will be one trip, which is more profitable and safer. At this level, the solution is to put less money into those ATMs that need to be collected less frequently, so as not to create unnecessary idle reserves.
It is clear that all these blocks can be made on anything. For example, logistics is Oracle Transportation Management, making forecasts on IBM SPSS, network placement on something else. Almost every major vendor has its own solutions.
The main obstacle to good optimization is security requirements. To put it simply, optimization itself implies a clear system for visiting points, and security requirements say that such a system will not last long. There are further options:
Considering the level of work of predictive software using the example of this reading of the thoughts of call center clients , you can do the following: when authorizing a card, you can create individual work scenarios and proposals. For example - “top up my phone”, “give me money for lunch”, “me as usual”, all sorts of “I feel lucky”, in combination with the purpose of payments - reminders about paying for the Internet (with an instant opportunity to resolve the issue), payment of fines (even dialed 20 minutes ago) and so on.
The next obvious step is to understand which clients to offer what. You can react to actions with an ATM and analyze behavior patterns. For example, launch personalized marketing campaigns, make “tasty” offers to customers who may leave, and so on.
In fact, there are no such complex projects in Russia. At best, individual blocks have been worked out. Banks practically do not deal with this: not everyone has the idea that they can live even better and at what expense.
From the standard scheme “when there are 200 thousand left in the ATM, you need to panic and go” to a smart ATM without taking into account space solutions, the difference is from 40 to 80 thousand rubles in savings per year per ATM. Considering the number of ATMs of one bank, this difference completely covers the cost of solving the problem and implementing the system.
Collection is an expensive and quite risky procedure from a security point of view, so any ATM maintenance is quite expensive. The natural desire of the bank is to reduce the number of collections. This results in a classic resource allocation problem: on the one hand, there should always be money in the ATM for clients, on the other hand, ideally the bank wants the last bill to leave the box just before collection.
Banks optimize all cash handling processes: this includes the aforementioned modularity of the cassettes (so as not to take out money and count with your hands), and a special arrangement of bags in the machine, exactly corresponding to their order of use along the route, and much more. At the same time, everyone understands perfectly well that if the ATM does not give money, the client will be upset: this is not a beer stall where you can choose a different type or simply cross the road and get the goods from another place. When the client does not receive money from his “native” ATM, he becomes very upset, indignant and begins to lose confidence in the bank.
Accordingly, the main priority is to always have money. At the same time, you want them to be, if not back to back, then with a minimum reserve - and this is where you need very good IT technologies, which can cost very good money.
Tasks
In principle, analytics has two tasks:
- It is optimal to create a banknote breakdown: what bills, in what denomination and how many pieces should be in the ATM.
- Know the collection schedule and, depending on the specific situation, calculate such a reserve of funds so that it will last until the next time without unscheduled trips.
Let's start with a breakdown by banknote . As I already said, there are usually 4 cassettes: of course, two of them immediately go under small ones: under hundred and five hundred. But there are nuances: for example, you don’t have to put in 500-ruble bills at all, but you can put in more than others. Or use two boxes at once for bills of the same denomination. Or don’t put in five thousand, and so on. Here everything depends on the consumption scenario and the issuing algorithm: if this is a student canteen, then collection may be necessary if there are 3 full cassettes with the hundred hundredths of money running out.
The second point is actually how much balance there should be. The scenario for using an ATM is important here: for example, there are exclusively salary ATMs that are located in offices, there are ATMs that are located in shopping centers, and there are ATMs that are located in squares at train stations. Obviously, payroll workers will have pronounced “critical days”; those in shopping centers will have seasonal peaks and troughs corresponding to the general peaks and troughs of retail; in squares and train stations, much depends on fluctuations in traffic density and its nature.
For example, buy yourself an ATM in a business center. On those days when there is a salary in the offices, it should be filled to capacity with large bills so that those who withdraw can withdraw as much as they need. Then, for example, it is filled by a third with bills of lower denomination, and for some holiday or weekend it is filled with more. Based on the analysis of transactions, you can understand the peaks in withdrawal amounts, understand the client profile of this ATM, and find out what breakdown needs to be done.
There are a lot of nuances, but in general, the task largely comes down to solving the problem of providing goods on the store shelf. With all this, there are a lot of security features, for example, a ban on systematically visiting an ATM (using the same routes).
To optimize we need to know
About the ATM:
- Number of cassettes
- Issuing algorithm (is it possible to select large/small?)
- Does he accept money or just give?
- What additional services does it provide?
- How and when the money was withdrawn (history of withdrawals and balances for Data Mining)
- Where is it located (in a business center, in a shopping center, in a bank branch)
- Operating mode of the place where it is located
- How can you combine ATMs into groups?
- Restrictions on the breakdown of banknotes (are you ready to put 50 banknotes? What about 100 bills?)
- Multi-currency ATM or not?
- Cost of finding money at an ATM
- Collection costs
- Goal (i.e. what level of service is needed)
Solution
The first part of the solution is architectural. You don’t need to figure out how to transport money, but first, you need to design and build an ATM network so that it works optimally. This is done based on a bunch of data. Our analytics can plan the ATM network itself so that everything is optimal, taking into account many restrictions - the presence of nearby competitors, traffic on streets, shopping centers, train stations, taking into account seasonal changes, and so on.
Then we must decide how much money and in what denominations should be lying in the ATM. Plus - how often it must be collected.
ATMs can also be replenished by customers, and the money there is collected in a separate cassette, which can also overflow. Money is not dispensed from this cassette: input-output at the ATM in different processes. From an IT point of view, it seems logical to use this money for issuance, but according to the law, all this must first be passed through the accounting department, and for this the cassette must be brought to a bank branch. Therefore, it is beneficial for collection to immediately replace the cassette with money from replenishments (since they have already arrived anyway).
Next, you need to ensure that ATMs are properly filled. To do this, we take a system that allows us to optimize the route from the point of view of minimizing costs and at the same time set restrictions like “routes must change once a week” (except for security services, no one knows which route the collection vehicle will take tomorrow).
Returning to logistics, we have several blocks: first, we need to plan it strategically, second, we need to understand who to collect with what amounts, and so on, and third, we need to optimize the collection schedule, both for individual ATMs and for a chain of ATMs at once. If there are a lot of them, then for each we calculate how to optimally collect it, and then we group them according to routes.
For example, if one ATM needs to be collected once every 2 days, and another - once every 3 days, this will be two trips. If you collect both at once, then it will be one trip, which is more profitable and safer. At this level, the solution is to put less money into those ATMs that need to be collected less frequently, so as not to create unnecessary idle reserves.
It is clear that all these blocks can be made on anything. For example, logistics is Oracle Transportation Management, making forecasts on IBM SPSS, network placement on something else. Almost every major vendor has its own solutions.
Reality
The main obstacle to good optimization is security requirements. To put it simply, optimization itself implies a clear system for visiting points, and security requirements say that such a system will not last long. There are further options:
- The security service itself works out the routes. We can simply issue recommendations and make monitors of how much money is in which ATM: communication between ATM software and the central office is a common thing. But we are just solving another problem - we don’t want to act reactively: “Ahh, our red light is on, the balance is zero, panic!”, We want to either help this red light light up today and notify that there is only enough money for tomorrow - or we must make a competent collection plan so that the lighting of this red light is something urgent for us. Roughly speaking, it is much more economical to spend energy on solving a problem at the architectural level than to try to correct the situation at the operational level every time.
- The security service is ready for innovations. In this case, an environment usually develops where routes can be built taking into account the optimal traffic pattern.
- The bank is ready for strategic planning. Here we are talking about calculating the optimal location network, setting optimal routes and alerts about the supply of money.
A little bit of the future
Considering the level of work of predictive software using the example of this reading of the thoughts of call center clients , you can do the following: when authorizing a card, you can create individual work scenarios and proposals. For example - “top up my phone”, “give me money for lunch”, “me as usual”, all sorts of “I feel lucky”, in combination with the purpose of payments - reminders about paying for the Internet (with an instant opportunity to resolve the issue), payment of fines (even dialed 20 minutes ago) and so on.
The next obvious step is to understand which clients to offer what. You can react to actions with an ATM and analyze behavior patterns. For example, launch personalized marketing campaigns, make “tasty” offers to customers who may leave, and so on.
In fact, there are no such complex projects in Russia. At best, individual blocks have been worked out. Banks practically do not deal with this: not everyone has the idea that they can live even better and at what expense.
Saving
From the standard scheme “when there are 200 thousand left in the ATM, you need to panic and go” to a smart ATM without taking into account space solutions, the difference is from 40 to 80 thousand rubles in savings per year per ATM. Considering the number of ATMs of one bank, this difference completely covers the cost of solving the problem and implementing the system.
