I made a mistake in . I worked for a payment company. I was a junior analyst. A customer withdrew forty-two dollars from a digital wallet. The customer found a bug in the code. The customer withdrew the same forty-two dollars again.
The system allowed the second withdrawal. I saw the error on a Tuesday morning. I felt shame. I told my manager about the error. I proposed a manual review step for all withdrawals. I wanted to prevent the error from happening again. My manager agreed to the step.
We added the check to the process. That check stayed in the process for seven years. I left the company. The manager left the company. The bug in the code was fixed in . The manual check remained. The manual check cost the company time and money every day. I was wrong to add the check. I should have fixed the code and moved on.
Phase I: Request and Verification
The first step on the map is the withdrawal request. The customer clicks a button. The system receives the request. The system records the time. The system records the date. The system checks the balance in the account. This step is fast. This step is necessary.
The second step is a balance verification. A staff member looks at the request. The staff member looks at the balance. The staff member confirms the numbers match. The numbers always match because the system already checked the numbers. The staff member spends thirty seconds on the check. The staff member does this one hundred times a day.
The third step is a history check. The staff member looks at the history of the account. The staff member looks for patterns. The staff member does not have a list of patterns to look for. The staff member uses their intuition. Intuition is not a reliable tool for security. The staff member spends two minutes on the history check.
The fourth step is a geographic check. The staff member looks at the IP address of the customer. The staff member looks at the city. The staff member looks at the country. If the country is different from the home country, the staff member flags the request. Many customers travel. Many customers use virtual private networks. The flags are usually false. The flags require more work.
Phase II: The Legacy of Step Five
The fifth step is a duplicate check. This is the step I created in . The staff member looks for other requests for the same amount. The staff member looks for requests made at the same time. The software already blocks duplicate requests. The staff member checks the work of the software. The staff member finds no errors. The staff member spends one minute on the duplicate check.
Redundancy rate of Step 5 since the software patch of 2018.
The sixth step is a compliance check. The staff member checks the name on the account. The staff member checks the name on the bank account. The names must match. This step is a legal requirement. It is a necessary step. But the staff member does not have the tools to verify the names quickly. The staff member opens a different window. The staff member types the name. The staff member waits for the window to load.
Phase III: The Hierarchy of Observation
The seventh step is a level one review. A supervisor looks at the work of the staff member. The supervisor checks the boxes. The supervisor does not look at the data. The supervisor looks at the marks made by the staff member. The supervisor trusts the staff member. The review is a formality. The review takes one minute.
The eighth step is a level two review. This step happens for withdrawals over a certain amount. The amount is small. The amount is five hundred dollars. A manager looks at the request. The manager is busy. The manager has meetings. The request sits in a queue. The customer waits for the manager to finish the meetings. The manager clicks the button. The manager does not see an error. The manager has not seen an error in three months.
The ninth step is the incident check. This step exists because of a problem in . A fraudster used a specific method to take money. The method involved a specific type of bank. The business no longer works with that bank. The risk is gone. But the step remains.
The staff member looks for the name of the bank. The staff member never finds the name of the bank. The staff member spends ninety hours a month looking for a bank that is not there.
The tenth step is the payment queue. The request enters a list. The list is sent to the bank. The bank processes the list in batches. The bank does not work on weekends. The bank does not work on holidays. The customer waits for the bank.
The eleventh step is the final clearance. The system updates the balance. The system sends an email to the customer. The email says the money is on the way. The money arrived in the bank account two hours ago. The email is late. The process is finished.
A business grows by ten percent. The review queue grows by twenty percent. Every new problem creates a new check. No one removes the old checks. Removing a check is a risk. If a manager removes a check and an error occurs, the manager is blamed.
If the manager keeps the check and the business wastes money, no one is blamed. The waste is hidden in the operational budget. The waste is called security. The waste is called diligence.
The Hidden Calculus of Delay
The math of the manual review is simple. A staff member earns twenty-two dollars an hour. The staff member spends ninety hours a month on step nine. The cost is one thousand nine hundred and eighty dollars.
Step nine caught six errors in one year. The errors totaled one hundred and forty dollars. The business spent over twenty thousand dollars to save one hundred and forty dollars. This is not a good trade. This is a bad trade.
Success Rate per 14,800 Checks
1 Error Found
The human is not a good filter for high-volume data.
Institutional memory is stored in procedures. Procedures are the tracks left by old mistakes. When a company experiences a failure, the company reacts. The company adds a layer of human oversight. The oversight feels like safety. The oversight is actually a delay.
The customer experiences the delay. The customer does not feel safe. The customer feels frustrated. The customer asks why the withdrawal takes so long. The customer is told that security is the priority. The customer knows this is a lie. The customer knows the technology exists to move money fast.
The Direct-Operator Alternative
Modern platforms handle this differently. A direct-operator platform like
does not use agent layers. An agent layer is a person who stands between the business and the customer. Manual review steps are internal agent layers.
When you remove the layers, you use automation. Automation does not get tired. Automation does not have intuition. Automation follows the rules of the code. If the code is good, the process is fast. If the code is bad, you fix the code. You do not add a human to watch the bad code.
I teach people about financial literacy. I tell people to look at the costs they cannot see. A fee is a visible cost. A delay is a hidden cost. A delay is the cost of a business that does not trust its own systems.
A business that trusts its systems can settle a withdrawal in seconds. A business that trusts its systems does not need an A3 map on the wall. It does not need step nine.
Guarding the Dead Monument
The queue is a monument to past fear. The staff members are the guards of the monument. They watch the monument every day. They watch the monument every night. They wait for a ghost to appear.
The ghost is the incident from . The ghost does not come back. The staff members keep watching. They click the buttons. They check the names. They look at the IP addresses.
I spoke to a staff member last week. The staff member was tired. The staff member had three hundred withdrawals in the queue. I asked why the queue was long. The staff member said the manager was in a meeting.
“The manager earns sixty dollars an hour. The manager will spend one hour approving the withdrawals. The business will pay the manager sixty dollars to approve six thousand dollars of requests. The manager will find zero errors.”
– Internal Audit Reflection
For every fourteen thousand eight hundred manual checks, a human finds one error. The error is usually small. The cost of the labor for those checks is higher than the value of the error. The human is not a good filter for high-volume data. The human is slow. The human gets distracted. The human wants to go home.
When a business removes the manual layers, the business changes.
The staff can do better work. The staff can help customers with real problems. The staff can answer questions about game rules. The staff can help a customer set a budget. The staff can be human. They do not have to be a step on a map. They do not have to be a check in a queue.
I look at my own bank account. I see the time it takes for money to move. If the money moves fast, I trust the bank. If the money moves slow, I wonder what is wrong. I wonder if the bank has an A3 map. I wonder if the bank has a Step 9. I wonder if someone is looking at my forty-two dollars and clicking a box.
Breaking the Cycle
The goal of a system should be the removal of the human from the routine. The human should be reserved for the exception. Most withdrawals are not exceptions. Most withdrawals are routine. They are the result of a customer having fun. They are the result of a win.
A win should be celebrated with speed. A win should not be punished with a manual review. The person who added Step 9 is gone. The person who added Step 5 is gone. I am gone. The steps remain.
They are the tracks of our fear. They are the reason for the delay. We must learn to delete the steps. We must learn to trust the automation. We must learn that a process that grows forever is a process that eventually fails. The queue will grow until the business stops. We must stop the queue first.
A customer wants their money. They want it now. They do not care about . They do not care about the incident. They care about their balance. They care about the truth of the numbers on the screen.
The numbers are only true when the money arrives. Until the money arrives, the numbers are just light on a display. Automation makes the light real. Manual reviews keep the light as a promise. A promise is not a payout. A payout is a payout.
I will not add another manual step. I will fix the code. I will trust the system. I will let the humans do human things. I will let the machines do the math.
The math is faster. The math is cheaper. The math is the truth. The manual review is a lie we tell ourselves to feel safe. We are not safe. We are just slow. It is time to be fast. It is time to remove the tape from the map. It is time to take the paper off the wall.
The wall should be empty. The queue should be empty. The customer should be satisfied. That is the only goal that matters. Any step that does not lead to that goal is a step that does not belong on the map.
I learned this the hard way. I learned this from forty-two dollars. I hope you learn it for less.