Because the job of the agent was to take as many calls an hour as possible, not to do research.
A smart company would have a "trigger" that would send the call to an agent who might fix the issue. See BigA's insurance experience of an example of how that worked out positively.
Some are, some are not. Often a "boilerplate" survey is used to determine how to put together a deep dive into specific areas that might include, for example, "what would make you come back to American Express" with open ended responses triggering specific possibilities depending on the content of the open-ended answer.
The first is that you can't research something nobody has even seen or heard before. This is why, no matter how hard record labels and radio stations have tried to research brand new songs, there is no way to predict a hit.
You can easily research how iPhone users like and employ various features, screen sizes, cost-to-benefit perspectives and the like. But they have to have had the phone for some time to form opinions. And even more useful: find former iPhone users and dig to find why they left the brand.
Apple also produced the Newton. It took them a decade to figure that failure out. And then there was the "Lisa"...
Even the best consumer brand ever, P&G, fails with nearly 50% of all new products.
"Settlers" build farms and ranches and grow a family. "Pioneers" get shot or killed.