Review queue and AI classification

The review queue is where you stay in control of AI. After AI reads feedback, it suggests where each customer problem belongs. You confirm or correct it, so the opportunity list stays accurate.

Who can do this: people who can review feedback (Member, Admin, Owner). Viewers can read suggestions but not act on them.

What AI suggests

AI looks at each signal and compares it with your existing opportunities. It produces one suggestion per theme:

• Link to an existing opportunity, when the same problem is already tracked.
• Create a new opportunity, when nothing matches.

Signals that describe the same problem are grouped into a single card, so you decide once, not for every quote. Only pains, requests and churn risks are suggested. Praise, questions and observations never appear here.

Work through the queue

Open Feedback, then Review. For each card you see the suggestion, the supporting quotes and why it was suggested. Your choices:

• Accept. Creates the opportunity or adds the evidence to the one suggested.
• Reject. Dismisses the suggestion and the signals behind it.
• Link elsewhere. Connect the signals to a different opportunity.
• Split signal. Break a mixed signal into two, both keeping a link to the original feedback.

The header shows your position, such as "3 of 12 to review". A short recap shows how many were accepted, rejected and auto-linked recently.

Automatic linking

By default, AI only proposes. Under Settings › AI linking you can allow high-confidence matches to link on their own. The threshold can be set from 72% to 99% and defaults to 86%. Values at or below 82% show a warning, since more wrong links will slip through. Automatic links are recorded so you can see what happened.

Who can do this: people with the AI linking permission, normally the Owner.

Fixing mistakes later

You can unlink a signal from an opportunity at any time. It goes back to being unlinked. You can also archive a signal; if nothing else remains in that feedback item, the item is archived too.

Tips

• Review a little every day. Small batches are faster than a big backlog.
• Use reject for noise, and link elsewhere when the topic is right but the match is wrong.
• Turn on automatic linking only after you trust the suggestions.