Review alerts that arrive already understood

You name the things you care about. A tag called “late delivery”, another called “installation”, another called “damaged on arrival”. Mark the ones you want applied on their own and every new review is scored against them as it arrives.

A language model does the reading, not a keyword rule. It scores each of your tags from irrelevant up to “this review speaks directly to that”, keeps the ones that clearly apply and drops the rest. There is no word list to maintain and no synonym to forget.

Which is why it holds whatever the review was written in. A French review about la livraison and an English one about a parcel that never showed up land on the same tag, and land in the same count on Monday morning.

The tag then travels with the alert. On a Zendesk ticket it arrives as reviewflowz_late_delivery, an ordinary Zendesk tag, so a trigger can send the delivery complaints to logistics and the billing ones to billing. Tags are added to a review, never written over the ones already on it.

Turn translation on for an alert and pick the language your team reads. Reviews in another language arrive with the translation attached; reviews already in that language are left alone, and a star-only review is skipped because there is nothing in it to translate.

On a Zendesk ticket the body stays in the original and the translation lands as an internal note headed “Translated review”. Only agents see it, and it never posts back to the platform. In Slack and Microsoft Teams it arrives in the same message.

The reply goes the other way. Your AI agent has two language settings: mirror the reviewer, or always answer in one language you pick. On the default it writes back in the language the review was written in, drawn from the replies you have already published in that language.

Each review is translated once and kept, so a second alert on the same review reuses it rather than paying for it again. And an alert can be scoped to languages and countries, so the German and French reviews go to the people who speak German and French.

Tags come only from your list. A tag the model invents is resolved against your account and dropped, and if it applies every tag you have to one review, the whole result is thrown away.

The name match has to be shown. The model quotes the words it matched a person on, and Reviewflowz verifies those words appear in the review before anyone is credited. Anything short of certain sits in a queue with the phrase highlighted, waiting for a click.

Replies are written from replies you already published. For each review the agent retrieves five to ten of your own, gated to the same star band and the same language, widening only when there are fewer than five. Replies you wrote by hand rank first.

A reply is marked published only once the platform confirms it. A failed post stays failed, with your text intact and a notice on the card, instead of showing you a reply that never left the building. If the review itself is later deleted, Reviewflowz answers in the original Slack thread to say so, and when it was last seen.

One alert selects the review profiles it watches, a star rating with an at most, at least or exactly mode, and the languages and countries it covers. That is the panel, and it fits on a screen.

“Only reviews with text” is on by default for alerts and for help desk tickets, because a bare star has nothing in it to read or to answer. Turn it off on the Google Sheet, where the count is the point.

There is no keyword filter on an alert, and we would rather tell you than let you find out. If you want waking by the word “refund”, make it a tag. The tagging is semantic, so it catches the review that describes a refund without ever using the word, which is the half a keyword misses.

Replies are a switch on the alert rather than a permission per person. Turn it on in the channel where you trust people to answer, and leave it off in the one where the whole company is watching.