How to Un-Fixate an Algorithm

I accidentally bought sugar-free Heinz beans on Amazon. The product listing looks almost identical to the regular ones — close enough that I didn’t notice until the box arrived. I noticed my mistake too late, shrugged, and ate them.

Then I bought them again.

Then, in a moment of discount-induced madness, I ordered a case of the things. They were 40% off. I reasoned: maybe they’re actually fine. They are not. They taste like someone described beans to a chemist.

The algorithm took notes

Amazon’s recommendation engine treats every purchase as a revealed preference. Buy sugar-free beans twice and you clearly love sugar-free beans. Buy them in bulk and you are a sugar-free bean person. It is now very keen to tell me about reduced-sugar condiments, low-calorie canned goods, and something called BeanProtein which I will not be purchasing.

There is no button that says “this was a mistake.” No way to explain context, flag an accidental purchase, or tell the system your preferences have been corrupted by a series of bad decisions. Every purchase is treated as signal. Signal is permanent.

Heinz did not help

Part of this is on Heinz. The regular and sugar-free versions look almost identical online — similar enough that you can click the wrong one without realising. This is not an accident. It’s a design choice that trades customer clarity for search proximity. At scale that’s probably fine. For the individual it’s mildly annoying, and it means Amazon can’t distinguish “bought beans deliberately” from “clicked the wrong listing.”

Both look like demand.

The lesson isn’t really about beans. Recommendation systems optimise for a model of you built from your history, and there is no user-facing correction mechanism when that history is wrong. You’re not navigating the system. You’re inside it, and it only learns in one direction.

How I’m getting out

I haven’t. I still have approximately fifteen tins of sugar-free Heinz beans to work through. Madmen will tell you there is a reason beans are inherently funny — I am living that reason, one joyless tin at a time.

Once I’m done, I’ll probably switch to Bold Beans.