You believe your support team is your first line of defense, but they are actually your most successful camouflage. We are taught to look at a high volume of repetitive tickets as a metric of “efficiency”-how fast can we close them, how many can one agent handle, what is the cost per interaction.
We treat the human being in the swivel chair like a biological filter, straining out the noise so the engineers can focus on the “real” problems. But in my world, the world of balancing variables so a system doesn’t collapse under its own weight, a repetitive question isn’t a ticket. It is a leak. And the better your agents are at plugging that leak with their own voices, the less likely you are to ever fix the pipe.
The Miracle of Translation
Consider the case of May. It is in a quiet corner of a digital service hub. The air conditioning is humming a low, mechanical B-flat, and the only other sound is the rhythmic, almost percussive tapping of May’s keyboard.
She is on the “Graveyard Shift,” a term that feels increasingly literal as the hours stretch toward dawn. May is not a technician, but she performs a daily miracle of technical translation. Over the last , she has answered the exact same three questions 142 times.
3
142
The repetition ratio: 3 core issues amplified into 142 distinct human interventions.
The first question is a ghost that haunts every digital transaction: “Where is my withdrawal?”
In May’s world, specifically within the fast-moving ecosystem of taobin 555, the system is actually working perfectly. The platform is an automated cashier; it doesn’t sleep, it doesn’t take lunch breaks, and it doesn’t “hold” money for fun.
When a user in Chiang Mai or Hat Yai hits that button, the API fires, the ledger updates, and the request is sent to the banking rails. But the human on the other side of the screen isn’t looking at an API log. They are looking at a static screen that says “Success,” while their bank balance remains stubbornly unchanged.
The Transaction Finality Paradox
In game design, we see this when a player hits a button and there’s a three-frame delay before the character jumps. Those three frames are the difference between a game that feels responsive and a game that feels “broken.” In finance, those three frames can be three minutes or three hours, depending on the recipient bank’s legacy middleware.
May spends her night bridging this gap. She explains the “why” that isn’t on the page. She tells them about the “interbank clearing lag”-a technical term she softens into “just a few more minutes for the bank to catch up.” She is a hero because she prevents a panic. But she is a symptom of a defect because the “Withdrawal Status” page doesn’t explain this lag.
To a developer, “Processing” is a factual state. To a user at , “Processing” is a void where their money has gone to die.
I recently tried to return a toaster to a big-box retailer without a receipt. The clerk was a young man who clearly knew the toaster was theirs-it had their exclusive brand name embossed on the side-but the system demanded a 12-digit transaction code that had long since faded from the thermal paper in my wallet.
He spent fifteen minutes “heroically” bypassing the system, calling a manager, and manually entering overrides. He was great. But the system was a failure. It was designed to forget me the moment I walked out the door, and his labor was the only thing correcting that institutional amnesia. We praise the clerk, but we should be firing the software architect who decided thermal paper was a permanent record.
The Mathematical Tax
The second question May faces is more mathematical: “Why is the amount different?”
This usually happens when a user doesn’t account for the tiny, granular frictions of the financial world. Perhaps it’s a small fee from the receiving bank, or a misunderstanding of a promotion’s turnover requirements. On a platform with 3,000 game titles and live dealers, the math moves fast. When you remove the “agent layer”-the middleman who used to manually handle these bets-you gain speed, but you lose the human who used to explain the “tax” of the game.
May explains the math. She does the subtraction for them. She points out the line item that the UI has tucked away in a “Details” dropdown that nobody ever clicks. She is essentially a human calculator, performing 80 sessions of basic arithmetic a night because the interface assumes everyone is a forensic accountant.
42
seconds per repeat
An agent repeats a known answer 80 times a night. By Tuesday morning, she has “burned” the equivalent of a full feature update’s worth of cognitive energy.
If we look at the statistics of this labor, the “Support Debt” becomes staggering. We think we are paying for “Support,” but we are actually paying a high-interest tax on poor documentation. We are essentially hiring people to read the manual out loud to people because we were too lazy to write the manual in a way that is readable.
The Existential Ping
The third question is the most existential: “Is the site down, or is it just me?”
Because the platform is browser-based-a brilliant move that bypasses the bloat of app stores and storage limits-it is subject to the whims of mobile data. A user on a bus in Bangkok might hit a dead zone. A player in a provincial district might have a flickering 4G signal. When the “Live Dealer” stream stutters, the user doesn’t blame the tower; they blame the site.
May checks the server status. She checks the latency. She asks the user to clear their cache-the digital equivalent of “have you tried turning it off and on again.” 99% of the time, the site is fine. But that 1% of doubt is where trust goes to bleed out. May’s job is to be the “Pulse Check.”
“The servers are still humming. It’s just your signal.”
Again, this is a documentation gap. A simple, real-time “System Health” light on the dashboard would save May of work a week. But because May is so good at being the “Health Light,” the product team never sees the need to build one.
They look at the “Down Time” reports and see 100% uptime. They don’t see the 400 people a week who thought it was down and had to be talked back from the ledge by a human being.
The Danger of “The Heroic Guard”
In game balancing, if I notice players are struggling with a specific level, I look at the data. If 80% of players are using a specific “exploit” or “cheese” to beat a boss, I don’t congratulate the players on their creativity. I realize that the boss’s mechanics are poorly telegraphed. The “exploit” is the players’ way of fixing my mistake.
Support agents are the “cheese” of the corporate world. They are the workarounds. They are the “heroic” patches on a system that is failing to communicate. When May logs off at , her eyes are dry and her wrists ache, but her “Customer Satisfaction Score” is likely 98%.
The management will see that 98% and think, “We are doing a great job.” They won’t see the 142 times she had to do the product’s job for it. The most expensive defects in any organization are the ones that someone has heroically absorbed.
Management View
98% CSAT
“The system works perfectly.”
The Reality
142 Leaks
“May is doing the UI’s job.”
When a problem is “solved” by a human, it disappears from the data. It doesn’t trigger an alert. It doesn’t show up as a bug. It just becomes a line item in the payroll-a recurring cost that we accept as the “cost of doing business.”
But the cost is higher than just May’s salary. The cost is the loss of clarity. By allowing May to bridge the gap between the “Automated Cashier” and the “Anxious User,” we are staying blind to how our users actually perceive our service.
If we want to build systems that actually scale, we have to stop being grateful for our “hero” agents and start being curious about why they need to be heroes in the first place. We need to look at the three questions that are asked all night long and realize that every time May types the answer, she is writing a sentence that should have already been on the screen.
In the end, May hands over her log to the morning shift. She mentions the “withdrawal lag” questions, the “missing 50 baht” math, and the “is it down” pings. The morning shift agent nods, sips her coffee, and prepares to do it all over again.
The cycle continues not because it is efficient, but because the human capacity to compensate for a system’s silence is almost infinite. We are so good at fixing things that we forget to wonder why they keep breaking in the same spot.
The next time you look at your support volume, don’t look at how fast the tickets are closed. Look at how many of them shouldn’t have existed if your product had the courage to speak for itself.