A visitor raises three concerns in one message. You answer the third, because it was the one you knew the answer to, and the chat moves on. Twenty minutes later you close it feeling good about the exchange, and the first two concerns are still sitting unanswered in the transcript.
This is not carelessness. It is what happens when a conversation is fast, when you are handling more than one at a time, and when nothing in front of you is keeping a list.
The concern you forgot you were told
The blocker panel in the agent console keeps that list. It reads what the visitor writes, lifts out the sentences that read like a doubt, a complaint or a hard obstacle, and shows them on the right rail as rows you tick off. The collapsed pill shows how many are still open.
It is worth being clear about what it does not overlap with. Explicit questions, the ones ending in a question mark, belong to a different panel and are skipped here entirely. Overall mood is a separate gauge. This panel is only interested in the specific, addressable statements that stand between the visitor and a yes.
Each visitor message is stripped of HTML and split into sentences on full stops, exclamation marks, question marks and line breaks. Anything shorter than eight characters is dropped as noise.
What survives is matched against a list of patterns, each carrying a weight. A strong pattern qualifies on its own. A soft pattern does not: it needs to combine with another before the sentence counts. That two-tier design is the difference between a list you trust and a list you learn to ignore, because generic frustration words like annoyed or disappointed are soft, and only become a row when they arrive alongside something concrete.
Rows are truncated to a couple of hundred characters, deduplicated against each other and against what is already on the board, and capped per visitor. You get the sentence in the visitor's own words, not a paraphrase, which matters when you go back to answer it.
The eight kinds of blocker
Every row is tagged with a category, which is mostly there to help you triage a long list at a glance. There are eight.
Price and value covers too expensive, cannot afford, over budget, not worth it, cheaper elsewhere. Time covers waiting, slowness, and the promised thing that never arrived. Broken covers does not work, error, damaged, missing or wrong items. Doubt covers not sure, not convinced, worried about, and does not add up.
Refund and cancellation covers wanting money back, closing an account, or returning something. Trust covers the words people use when they are suspicious: scam, sketchy, unsafe, a privacy concern. Compare covers a named alternative being better or cheaper. And vent covers the frustration markers, which as noted are mostly soft on their own.
The categories are English phrase lists, not a language model. They are good at the common shapes and they will miss an unusually phrased objection. Treat the list as a safety net under your own attention, never as a replacement for it.
The two rules that stop false rows
Every wrong row costs you attention, so two rules exist purely to suppress them.
The first is the question rule. A sentence that ends in a question mark is never a blocker here, whatever else it matches, because the questions panel owns that surface and two panels showing the same sentence is worse than either alone.
The second is a list of sentences that look like objections and are not. No problem at all. I do not mind waiting. It is working now. All good. That makes sense now. A match against any of these removes the sentence outright, no matter how many objection patterns it also hit. These are the phrases people use when something has just been resolved, and a resolution should not create a blocker.
Open, addressed, resolved
Rows start open. When you send a substantial reply, meaning one longer than a brief acknowledgement rather than a one-word yes or one sec, every row raised before that reply flips to addressed.
Addressed is not resolved, and the distinction is the whole point. It means something was said after the concern was raised. It does not mean the concern was answered, and the panel is careful never to claim otherwise: nothing ever moves itself to resolved. Marking a blocker resolved is a deliberate click, and so is skipping one that was extracted in error. Anything you marked can be reopened.
A green tick that appeared by itself on a real blocker would be worse than no panel at all, because you would trust it. So the automation stops exactly where certainty stops.
Why it can hold up your wrap-up
The panel feeds the wrap-up card. Open and addressed rows both appear in the recap you can copy or save, and while any row is neither resolved nor skipped, the wrap-up readiness indicator sits at warn.
That is the one place this panel applies real pressure, and it is the reason it exists: closing a chat with an unanswered blocker is the specific mistake it was built to prevent. Clearing the warning takes one click per row, and if a row was never a real objection, skipping it is a legitimate way to clear it.
Like the other panels, the list is per-visitor and held in the browser only. When a chat is transferred or accepted by someone else, the list is dropped rather than inherited, because another agent's judgement about what counts as addressed is not something you should be handed silently.
What it is not
It is not a report. Nothing is written to a database, nothing reaches a CRM, and the list disappears when the chat does. It is not automation either: it never replies, never routes, and never tells the visitor anything.
It is also not a scorecard for the agent. A chat with six blockers may be a difficult customer handled well, and a chat with none may be one where the visitor never opened up. The count says something about the conversation, not about the person handling it.
What to measure
Since the panel stores nothing, anything you want to measure has to come from what you record deliberately. The useful habit is to note the real blocker in your chat notes or as a tag before you close, and then look at those tags monthly in your chat reports.
Two patterns are worth watching for. A blocker that keeps appearing across many chats is usually a website or pricing-page problem rather than a chat problem, and it is cheaper to fix at the source. And chats that closed with a blocker only ever marked addressed, never resolved, are the ones most likely to come back as a second contact.
Put it into practice
- Open a live chat and press Ctrl+Shift+B, or click the pill on the right rail, to see the current list.
- Before you close any chat, read the open count. Zero is the target, and skipping a bad row counts.
- Mark a row resolved only when you have actually answered it, not when you have merely replied since.
- Skip rows that were extracted in error. It keeps the list trustworthy for the next chat.
- Record the real blocker as a tag or note before wrapping up, since the panel itself keeps nothing.