Every company I have looked at has a support function that is measured on closing tickets, and a product function that is measured on shipping things. Between those two facts sits the most valuable unread document the business produces, which is the accumulated record of every customer who took the trouble to explain what was wrong.
The complaint gets answered. That is what the system is designed to do, and it does it competently. What almost never happens is the second thing: somebody reading a hundred of them together and noticing that forty are the same complaint wearing different words.
Resolution and comprehension are not the same operation
A ticket is closed when the customer stops being unhappy. That is a legitimate goal and it is orthogonal to whether anything was learned. A refund closes a ticket. A workaround closes a ticket. A patient explanation of why the thing works that way closes a great many tickets, and each closure removes the evidence from view.
The categories make it worse. Support taxonomies are built for routing — billing, access, integration — because routing is what the team needs to function. They are useless for diagnosis, because the interesting pattern almost never respects them. Nobody has a category for the customers who all misread the same screen.
I have come to think the fix is smaller than it sounds and is mostly about who reads rather than what gets collected. A product manager reading twenty raw tickets a week — unsummarised, uncategorised, in the customer's own words — will find things no dashboard produces, because the dashboard reports the taxonomy and the taxonomy is the thing hiding the pattern. That is the general case of instrumentation nobody reads: the number was produced, and the decision it should have informed sits with someone who never sees it.
Automation is making this both better and considerably more dangerous. A system that resolves routine contacts removes exactly the high-volume, low-difficulty complaints that were the easiest signal to count — so a recurring irritation that generated four hundred tickets a month now generates four hundred silent resolutions, and the volume that would have made it visible never accumulates anywhere a human looks. The AI answered and the customer called anyway describes the cost side of this. There is an information side too, and it is quieter.
None of this argues for treating every complaint as a mandate. Most are individual, some are unreasonable, and a company that redesigned around each would ship nothing. It argues that the aggregate is a research asset the company is already paying to collect, and that filing it under resolved is a decision to throw it away.
The question I would ask any product leader is simply when they last read a complaint that had not been summarised for them. In my experience the answer is a length of time that surprises the person saying it.
Topics opinionmanagement
