In the , the potters of Delft had a problem with “sharding.” When a kiln failed or a plate cracked during the cooling process, the resulting waste wasn’t just trash; it was a structural hazard.
To handle the volume, they created specialized disposal pits that were graded by size. If a piece was too large for the pit, it was hammered down until it fit. The efficiency was remarkable. The streets remained clear, the workshops stayed tidy, and the system worked perfectly, provided you didn’t care that you were pulverizing valuable evidence of why the kilns were failing in the first place.
By forcing the output to fit the disposal method, they destroyed the data needed to fix the input. We are currently doing the exact same thing with the little glowing bubble in the bottom right-hand corner of our screens.
The Architecture of the Chat Bubble
Consider the “Chat Now” icon as a psychological system. It is designed to look like a friendly invitation-a soft-edged speech bubble, perhaps a smiling avatar of a person named “Sarah” who doesn’t actually exist.
But functionally, it is a valve. In any pressurized system, a valve’s primary job is not to facilitate flow, but to control it. Most modern support chatbots are “pressure-reducing valves.” They are installed at the front door of a business to ensure that the high-volume, low-complexity queries-“Where is my package?” or “What is your return policy?”-never reach a human being.
The problem is that a valve doesn’t know the difference between a high-pressure stream of water and a high-pressure stream of liquid gold. It just knows the volume. When a customer approaches a business with a question that contains nuance, they are immediately met with the hammer of the Delft potter.
The chatbot offers a menu of four options. If your problem is Option Five, the system effectively tells you that your problem does not exist.
After force-quitting a malfunctioning design application seventeen times this morning because it couldn’t recognize a simple SVG path, I realized that my frustration wasn’t with the software itself. It was with the “Help” menu that kept offering me tutorials on how to “Open a File.” The system had trained me to stop looking for a solution. It had trained me to stop asking.
The Silence in the Logs
When a company brags about “80% deflection rates,” they are usually celebrating a ghost. Deflection sounds like a victory-it implies the customer was satisfied and went away.
The “Deflection Rate” often measures the customer’s loss of hope, not the resolution of their problem.
But as an archaeological illustrator, I spend my life looking at the things people left behind. In the dirt, the most important thing isn’t the pot that survived; it’s the pattern of the breaks in the ones that didn’t.
“If you only look at the shards that fit into your catalog, you lose the history of the entire civilization.”
– Finley C., Archaeologist
When a customer types a complex, 40-word question about whether a specific multi-zone heat pump will handle the humidity of a converted basement in coastal Maine, and the chatbot responds with a link to “General Shipping Information,” a very specific type of “sharding” occurs.
The customer doesn’t usually get angry. They don’t demand a supervisor. They simply sigh, close the tab, and leave. Or, worse, they “guess.” They stop asking the hard questions because the interface has signaled, quite clearly, that hard questions are not welcome here.
They learn that nuance is a waste of their time. This is “learned helplessness” applied to the checkout cart. The valuable conversation-the one that would have prevented a $3,140 ordering mistake-quietly never happens. The company’s logs show a “successful interaction” because the user stopped typing. In reality, it was a catastrophic failure of communication.
The Great Signal Failure of
In the late 19th century, the railway systems began experimenting with automated signaling. In one famous instance, a “dead man’s bell” was installed to alert operators if a track switch hadn’t been fully engaged. The bell was binary: either the switch was locked, or the bell rang.
One evening, a switch was partially engaged-enough to satisfy the mechanical sensor but not enough to prevent a derailment. The operator heard no bell and assumed all was well. The automation had replaced his habit of walking out to physically inspect the iron.
Because the “instant answer” (the silence of the bell) suggested safety, he stopped doing the nuance-check. Modern e-commerce is currently derailing on “partially engaged” information. We have built systems that are excellent at answering the questions that don’t matter, which in turn makes customers feel foolish for asking the ones that do.
In the world of complex home improvements, like HVAC, this is a recipe for a decade of regret.
The Archaeology of a Bad Order
If you are buying a pair of socks, a chatbot is fine. If the socks are the wrong color, the stakes are the price of a latte. But if you are buying a ductless mini-split system, you are engaging in a technical engineering project for your own home.
I’ve seen how this plays out. A homeowner knows their bedroom gets afternoon sun and has 12-foot ceilings. They want to ask if the standard BTU calculation needs to be adjusted for that thermal load. They open the chat. The bot asks, “Are you looking for a single-zone or multi-zone system?”
The homeowner thinks: “It’s more complicated than that.”
The bot says: “I don’t understand. Please choose from the following options.”
The homeowner chooses “Single-zone” just to move forward. They’ve just been “sharded.” They’ve been hammered down to fit the pit. They eventually buy a 9,000 BTU unit that will never, ever cool that room properly because they were trained by the interface to stop trying to explain the 12-foot ceilings.
This is why the curator model is the only thing that actually survives the test of time. At
MiniSplitsforLess, the model isn’t built on “deflecting” the customer.
It’s built on the realization that the most expensive thing you can sell someone is the wrong system. When you remove the robotic gatekeeper and invite the “hard” questions-the ones about line set lengths, zone compatibility, and actual BTU loads-you aren’t just selling a box. You are preventing a failure.
The Liberty Ship of Home Comfort
During World War II, the United States built Liberty Ships at a record-breaking pace. To speed things up, they switched from riveting the hulls to welding them. It was faster, cheaper, and more “efficient.”
However, the welds were too rigid. In the frigid waters of the North Atlantic, cracks would form, and because the hull was one continuous welded piece rather than individual riveted plates, the cracks would zip through the entire ship. Some ships literally snapped in half while sitting at the pier.
Fast, brittle, prone to catastrophic failure when stress increases.
Flexible, resilient, stops cracks from spreading through “play.”
The “welded” interface of a chatbot is just as brittle. It creates a seamless, fast experience that works perfectly right up until the temperature drops-or, in our case, right up until the complexity of the project increases.
A human-centric support model is like a riveted hull. It has “play” in it. It can absorb the stress of a weird question. It can stop a crack before it splits the entire purchase.
Reclaiming the Gray Area
We have been sold the lie that friction is always bad. We are told that every “click” removed from a process is a victory for the consumer. But some friction is functional.
The “friction” of having to explain your specific room layout to a human expert is the only thing that ensures you don’t end up with an icing evaporator coil three months from now. When a company puts a human at the front door, they are making a radical statement: “We are not afraid of your complexity.”
It’s an invitation to bring your 12-foot ceilings, your coastal Maine humidity, and your weirdly shaped sunroom to the table. By refusing to shard the customer’s needs into a pre-set menu, you preserve the data. You keep the kilns from failing. You ensure that the system that arrives at the door is actually the one that belongs in the house.
The next time you see a chat bubble, remember the potters of Delft. Don’t let the interface hammer your question into a shape that fits their bin. If the system isn’t built to handle your nuance, it isn’t built to solve your problem. It’s just built to keep the workshop tidy while your house stays hot.
Seeking out the experts who still value the “rivets” over the “welds” isn’t just a better way to shop-it’s the only way to ensure that what you’re building actually holds together when the pressure hits.