Open any customer and you will see what Arbyn has learned about them. This article explains what is there, how to read it, and how to remove any single detail without deleting the customer.

Things the customer told you. Stated in their own words in a chat or an email. These are marked as stated, and Arbyn speaks about them plainly.

Things Arbyn worked out. Inferred from orders, browsing or context. These are marked as inferred, and Arbyn speaks about them as guesses: "past orders suggest", "seems to".

Things Arbyn computed. Worked out overnight across a customer's whole history: how they tend to respond, what categories they lean toward, whether they look at risk of drifting away. Each one shows the job that produced it and when it last ran.

Every detail also shows a confidence level and how many times Arbyn has seen the same thing. A detail mentioned once with low confidence and one confirmed three times are not the same, and the page does not pretend they are.

Next to any detail is Forget this. It asks you to confirm, then removes that single row. The rest of the customer's record is untouched.

Two things worth knowing before you use it:

It cannot be undone from this page. There is no restore. That is why it asks first.

If the customer mentions it again, Arbyn may learn it again. Forgetting is not a permanent instruction never to know something; it removes what is stored today. If a customer has asked you never to keep a particular detail, the durable answer is to tell them Arbyn will only remember it if they bring it up, or to remove the whole customer record.

Computed signals do not have a Forget control, deliberately. They are recalculated every night from the underlying conversations and orders, so removing one would bring it straight back tomorrow. A control that appears to work and quietly does not is worse than no control. To change a computed signal, change what it is computed from.

Arbyn will not record special-category personal data, even when a customer volunteers it, and customers volunteer it constantly and innocently:

If a customer says they need something before surgery, Arbyn helps with the request and records nothing about the surgery. This is enforced when the record is written, not left to the AI to remember.

If your store falls into a category Arbyn recognises, you will also see details that only make sense there: a shade match and skin type for beauty, fit and usual sizes for apparel, ring size and metal preference for jewellery.

Some of these are only ever recorded when the customer states them in their own words. A shade, an undertone, a body measurement or an ingredient someone avoids can stand in for information about race or health, so Arbyn refuses to infer them from a photo, a name, or anything else, however confident it is. When a customer does state one, their exact sentence is kept beside it so you can check the reading rather than trust a parsed value.

If your store does not fall into a recognised category, this section says so rather than showing you an empty box.

If a customer asks to be removed entirely, everything described here goes with them, including what Arbyn worked out and computed rather than only what they said. Those inferences are personal data too, arguably the most sensitive kind, because they are judgements formed about someone rather than statements they made.

The same applies to a data request: the answer covers all three kinds, not only the ones the customer would have thought to ask about.