The hex key is already beginning to strip. It is the third time I’ve tried to tighten bracket 48, and the metal is giving way before the tension does. I am sitting on a floor of cold laminate, surrounded by 88 pieces of particle board that supposedly constitute a ‘modular shelving unit,’ but currently look more like a graveyard for a Swedish forest. There is a missing cam lock. Just one. Without it, the entire vertical structure has the structural integrity of wet cardboard. I reached for my phone, squinting at the screen to find the manufacturer’s support page, only to be met with a ‘Self-Service Portal’ that feels like it was designed by someone who fundamentally hates people. This is the physical manifestation of a digital lie: the promise that you can solve your own problems if you just try hard enough, provided the tools you’re given aren’t intentionally broken.

I spent 18 minutes navigating a menu that seemed designed to hide the very possibility of a missing part. It’s a specific kind of architectural cruelty. You click ‘Orders,’ then ‘Returns and Missing Items,’ and instead of a form or a chat, you are redirected to an article titled ‘How to Assemble Your Unit.’ I know how to assemble the unit. I am doing it right now. What I don’t have is bracket 48. But the system doesn’t care about my specific reality. It cares about its own internal logic, a logic built to ensure I never, under any circumstances, speak to a human being who might cost the company $28 in labor.

As someone who spends 48 hours a week as an AI training data curator, I see the bones of these systems every day. My name is Alex V.K., and my job is essentially to teach machines how to understand the mess of human intent. I spend my time staring at 108-line spreadsheets of customer queries, trying to categorize the frustration of people who are just trying to pay a bill or find a missing screw. What I’ve realized is that most self-service portals are not designed to empower the customer. They are defensive perimeters. They are built on the philosophy of ‘Deflection.’

In the industry, we talk about ‘Deflection Rates’ like they are a metric of success. If 88% of people who visit the help page don’t end up calling the support center, the project is considered a victory. But nobody asks if those 88% actually solved their problem, or if they just walked away in a state of quiet, burning resentment.

Take the invoice scenario. You’re looking at your screen, and there’s a charge for $228 that you don’t recognize. You click the button that says ‘Help with this invoice.’ In a world that respected your time, that click would trigger a context-aware system that knows exactly which invoice you’re looking at, which line items are disputed, and perhaps offers an immediate explanation or a way to flag it. Instead, you are dumped into a generic FAQ search bar. You type ‘unrecognized charge,’ and you get 18 articles about how to update your credit card or how to cancel a subscription. It is a profound disrespect for the customer’s intelligence. It assumes that your problem is generic, and therefore, the solution must be generic.

This morning, while I was still trying to find that missing cam lock, I found myself thinking about the 58 different steps it takes to actually reach a ‘Contact Us’ form on most major retail sites. You have to pass through the ‘Help Center,’ then the ‘Community Forum’ (which is usually a ghost town of unanswered complaints), then the ‘Virtual Assistant’ (which is just a glorified search bar), and finally, if you’re persistent enough, you might find a hidden ‘Email Us’ link.

There’s a deep irony in my work. We are using massive amounts of data to make machines sound more human, while the companies using these machines are using them to act less human. They want the efficiency of a robot with the optics of a friend. But you can’t have it both ways. If your ‘Self-Service Portal’ requires a user to have a PhD in your company’s internal hierarchy just to find a shipping update, you haven’t built a service; you’ve built an obstacle course.

“I’ve seen data sets where 78% of the ‘successful’ self-service interactions were actually just users timed out or closing the tab in frustration. In the database, that looks like a ‘resolved’ ticket. In the real world, that’s a customer who is currently looking at your competitor’s website.”

“

We need to move past the idea that self-service is about saving the company money. It should be about saving the customer’s sanity. This requires a shift from static repositories of text to dynamic, conversational interfaces. That’s why systems like Aissist are actually terrifying to the old guard-they remove the friction that companies traditionally used as a filter.

I remember a specific data set I worked on for a telecommunications firm. They had 388 different FAQ pages. When we analyzed the search logs, we found that 68% of users were searching for terms that didn’t appear in any of the titles. The users were using human language-‘Why is my bill so high?’-while the articles were titled ‘Understanding Pro-Rated Billing Cycles and Regulatory Surcharges.’ There was a fundamental disconnect between the language of the provider and the language of the seeker. This isn’t just a technical failure; it’s a failure of empathy.

[The death of the FAQ is long overdue.]

Back on my living room floor, I finally gave up on the portal. I went to a third-party hardware forum where a guy named ‘NutBolt88’ explained that this specific shelving unit frequently ships without the cam locks, and that you can buy a pack of 8 at any local store for about $8. Why couldn’t the manufacturer tell me that? Why couldn’t their self-service portal say, ‘We’ve noticed a packing error in Batch 48; if you’re missing a piece, click here and we’ll overnight it’? They knew. The data exists. But the portal wasn’t designed to share information; it was designed to contain it.

As a curator, my task is often to label ‘sentiment.’ I see the ‘Angry’ tag more than any other. People aren’t angry because they have a problem; they’re angry because they’re being ignored. They’re angry because they are being forced to do the company’s work for them. When you make a customer navigate an ineffective portal, you are essentially asking them to act as an unpaid intern for your support department. You’re saying, ‘Our time is worth $128 an hour, and your time is worth zero. So you go ahead and find the answer yourself.’

I’ve noticed that the companies that actually succeed in the long term are the ones that treat support as a product, not a cost center. They invest in the 28% of interactions that require deep empathy, and they automate the rest with tools that actually work. They don’t hide their ‘Contact Us’ button behind 8 layers of menus. They understand that a customer who gets a quick, effortless answer is a customer who stays for 8 years. A customer who spends 48 minutes fighting a chatbot is a customer who tells 18 people how much they hate your brand.

[Effort is the enemy of loyalty.]

There is a specific kind of exhaustion that comes from digital friction. It’s a low-grade hum of frustration that follows you throughout the day. You start with a missing screw, and you end with a feeling of general helplessness against the machines. But it doesn’t have to be this way. We have the technology to make support invisible. We have the data to predict what a customer needs before they even ask. The only thing missing is the will to stop treating customers like a line item to be ‘deflected.’