How can a recommendation experience learn from being wrong?
The anticipatory experiences panel talked about relevance that feels invited. A practical test of that: whether the customer can see why a suggestion showed up and correct the reason behind it.
Say what the suggestion is based on, in plain language, wherever the service can support it. If the recommendation uses a stated budget or a previous category choice, let the person see that input and change it.
Let people turn a suggestion down without explaining a private circumstance. Someone should be able to say it isn't useful without describing a change in health, family or finances.
A gift is a good example of a one time exception. Buying one doesn't necessarily mean the customer wants that category in every future recommendation, so treat an exception differently from a preference update.
Anticipation should leave room for the person to tell the service it has misunderstood, and that means nobody has to work around a persistent assumption.
What to try this week
Go through the correction path as a customer. Change an input, decline a suggestion with no reason given, and see whether the next result respects it and whether the preference can be removed later.