High deflection rates are the surest sign that your customer service department is currently dying, not succeeding. Most leadership teams look at a dashboard showing a chatbot handling 74% of incoming inquiries and they see a triumph of efficiency, a liberation of resources, and a justification for a leaner payroll.

They are wrong. What they are actually looking at is a map of where they have decided to stop being helpful. In the world of complex home infrastructure-the kind of world where a homeowner is trying to figure out if a 36,000 BTU condenser can actually support three 9,000 BTU air handlers and one 12,000 BTU unit in a high-ceiling living room-deflection is just a polite word for abandonment.

The Automated Sheet-Folder Problem

I spent nearly thirty minutes this morning attempting to fold a fitted sheet. It is a task that defies the laws of Euclidean geometry. No matter how you tuck the corners, there is always a lump of recalcitrant fabric that refuses to conform to a flat plane.

I mention this because a chatbot is essentially an automated sheet-folder that only works on flat linens. As soon as you give it the elastic, rounded, three-dimensional mess of a real-world technical problem, it begins to hallucinate a solution or, worse, it “deflects” the user toward a dead-end FAQ page.

And yet, the senior advisor who spent fifteen years learning the “elastic corners” of multi-zone HVAC compatibility is the first person to be reassigned or “optimized” out of a job once the bot goes live.

In my primary work as a court interpreter, I have learned that the most dangerous moments in a trial aren’t the long, flowing testimonies. They are the moments of silence where a witness uses a technical term that has no direct equivalent in the target language.

If I “deflect” that complexity by choosing a generic word, I have failed the court. Most businesses are currently failing their customers in exactly this way. They have automated the generic and, in doing so, they have convinced themselves that the technical no longer requires a human touch.

The Fallacy of the Waiting Room

We talk about a “70% deflection rate” as if it is a victory of logic, but in plain human terms, that is like a triage nurse announcing that they successfully ignored 70% of the waiting room because those patients only had scrapes, while the 30% with compound fractures were left in the hallway because the nurse was busy celebrating the empty chairs.

The 30% that the bot cannot handle are not “noise.” They are your highest-value customers. They are the ones about to spend $6,000 on a multi-zone system but who are terrified of buying a unit that won’t work with their existing electrical panel.

The tragedy of the modern rollout is the retirement of the person who handled the “hard cases.” I saw this recently in a leadership deck where a director beamed about how their new AI interface was fielding thousands of questions a day.

“The one guy who could untangle a six-zone compatibility puzzle-the guy who knew that a specific brand’s 24,000 BTU outdoor unit has a minimum circuit ampacity that trips a standard 20-amp breaker in older homes-was moved to ‘content strategy.'”

– Observations from the Floor

He was no longer allowed to talk to customers because the data said the bot was “handling support.” Efficiency is the ratio of output to input, which means that if you reduce the input (the human expert) while maintaining the output (the automated response), you have increased efficiency, even if the output itself has become entirely useless to the person receiving it.

The Maine in February Problem

This is the edge case that breaks the definition of success. If a customer asks, “Will this work in my garage in Maine during February?” and the bot replies with a link to the shipping policy, the “ticket” is marked as resolved because the customer didn’t reply within ten minutes.

The bot won. The company lost a sale. The expert who could have explained the performance curve of a hyper-heat pump at -13 degrees Fahrenheit was busy in a meeting about “optimizing the knowledge base.”

The problem with the curator-and-advisor model is that it is hard to scale in a spreadsheet. It’s much easier to buy a software license and tell the board that you’ve “solved” support. But in the HVAC world, specifically with ductless mini-splits, the “easy” questions are a vanishingly small part of the real value.

Anyone can read a spec sheet. Very few people can look at a floor plan and realize that the 12,000 BTU unit the customer wants to put in the bedroom is going to short-cycle constantly because the room is too small and well-insulated, eventually killing the compressor.

When you remove the human expert, you aren’t just saving money; you are removing the “check” in “compatibility check.” You are leaving the customer to decode the confusing specs on their own. This is where the most expensive mistakes happen.

A customer buys a system that is mismatched to their BTU load, or they forget the line sets, or they don’t realize they need a condensate pump for a specific wall-mount location. By the time they realize the mistake, the unit is on their driveway, the installer is charging them for a wasted day, and the chatbot is “deflecting” their anger with a polite “I’m sorry, I don’t understand that question.”

Putting the Expert on a Pedestal

Success in this industry isn’t about how many people you can stop from calling you. It’s about how many people you can guide toward the right decision before they hit the “buy” button. Organizations like

MiniSplitsforLess

understand that the “hard cases” are actually the only ones that matter.

The easy questions-“Do you ship to Ohio?”-don’t require an expert. But the moment a customer asks about the difference between a single-zone 12k and a multi-zone head unit’s modulation range, you need a person who has seen these units fail in the field.

I’ve spent a lot of time in courtrooms watching people try to simplify the truth to make it fit a legal narrative. It never works. The truth is always in the messy details, the “fitted sheet” corners of the testimony. When a company decides that their senior advisors are too expensive to keep on the front lines, they are essentially saying that they no longer care about the truth of their customers’ problems.

The irony is that the more “efficient” the bot becomes at answering the 70% of easy questions, the more visible and painful the 30% of hard questions become. The gap between the automated response and the human expertise grows until it becomes a canyon.

The customer who was handled perfectly by the bot for their tracking number is suddenly met with a wall of incompetence when their compressor fails to start. The “win” of the deflection rate becomes the “loss” of the brand’s reputation.

We need to stop treating expertise as a cost center that needs to be minimized. If your chatbot can answer every question a customer has, your product is probably too simple to be worth a premium price. If your product involves multi-zone configurations and BTU calculations, then your “deflection rate” is a metric of failure.

The Next Decade of Service

I think back to that fitted sheet. I eventually got it on the bed, but it’s not pretty. One corner is tucked under the mattress in a way that will probably pop off in the middle of the night. That is what automated support feels like. It looks okay from a distance, but the moment you actually try to use it, the whole thing snaps back and hits you in the face.

The companies that will survive the next decade aren’t the ones with the best bots. They are the ones that use the bots to handle the “where is my package” noise so that their smartest, most experienced humans can spend more time talking to the person who is trying to figure out how to heat a four-zone farmhouse in a blizzard.

Because when the “weird case” comes in-and it always does-the bot isn’t going to save the sale. The human who knows the voltage drop of a 50-foot line set will. The silence of a deflected customer is not the same as the satisfaction of a served one.

Every time we automate a human interaction, we should be asking what we are doing with the time we “saved.” If that time isn’t being reinvested into the complex, the difficult, and the high-stakes problems, then we haven’t actually saved anything. We’ve just outsourced our incompetence to a machine.

And eventually, the customer will notice that while the bot is very fast at replying, it is also very fast at being wrong about the things that actually cost money.