📥Inbox — everything that needs doing, in one list. The chips at the top split it up and show how many of each, in the order worth working: Waiting (someone wants a reply now), Unchecked answers, Missing answers, Picked follow-ups, Duplicates, Weak entries, Matched, unchecked, Untagged, Unchecked wording, Never tested. Every one of them ends the same way — an answer goes into the Knowledge Base. Work top to bottom; items disappear as you deal with them.
✅Unchecked answers — answers the AI already sent. This is the part people find surprising: these were already given to a customer. The AI was only partly confident, so it answered from an entry that only partly matched, and nobody has looked at it since. Each card shows the answer that went out and where it came from — the Knowledge Base entry the AI used. Nobody is waiting on you. If the answer was fine, click NO CHANGE NEEDED and it leaves the list — nothing is sent, nothing changes. If it was wrong, click FIX & SAVE TO KB and write the right answer, so the next person asking gets it directly. That customer is not messaged again either way.
🕑History — permanent record of every conversation. Queue items don't disappear — they move here once resolved. You can read the full thread, compare the AI's draft vs the agent's reply, and click + ADD TO KB on any answer — the bot's or the agent's — to add it to the Knowledge Base retroactively.
📚Knowledge Base — your permanent Q&A library. Entries stay until you DEL them. EDIT updates the answer in place — the row does not disappear. Use the Search box to find entries as the list grows. Aim for 50+ entries covering common questions; the more you add, the less the Queue fills up.
🏷️Topics — how entries are filed, and you can add to the list. An entry can carry several topics, not just one: a question about inviting a prospect can be Training and Invitation at the same time. The list is not fixed — any trainer can create a topic, and it is meant to grow as you write entries. Untagged in the Inbox lists entries carrying none. Renaming or archiving a topic is admin-only, because a rename relabels every entry using it.
⚡ANALYSE & SORT — fastest way to train. Paste a training script or upload a .docx/.pdf/.txt file. The AI extracts Q&A pairs for you to review, edit, and import in bulk. Use this to get many entries in quickly.
🔍TEST KB SEARCH — verify your entries work. Type a question a user might ask. ≥58% = AI will answer confidently. 35–58% = borderline, answered from a partial match. <35% = AI can't answer, goes to Queue. Test before adding (checks for duplicates) and after (confirms the entry is findable).
⭐H / M / L Quality Votes — optional but helpful. Anyone can vote on any entry. H (High) = accurate and well-written, gets search priority. M (Medium) = decent. L (Low) = weak, flags it for improvement. You don't need to vote on everything — just when you notice an entry is especially good or poor.
🔁Duplicate guard. If you add a question close to one already saved, the system shows the saved ones with OPEN THIS ENTRY, so you can read or fix the existing entry instead. If yours asks something different, save it anyway. This prevents 5 entries all answering "What is GWT?" in slightly different ways.
👤Trainer vs Admin. Everyone's saves go live immediately. A trainer's new entry or edit also appears in admins' Inbox under Trainer changes, showing before and after, where an admin clicks LOOKS FINE or REVERT whenever they have time. Revert puts back the version from before the trainer's change, or deletes an entry the trainer created. Any entry marked Admin only — compliance and other regulated wording — can be edited by admins alone.
🎯Never tested — prove an entry can actually be found. An entry matching its own question proves very little. Open Never tested and type the question the way a member would really ask it — different words, not the entry's own wording. If the search finds that entry confidently, it is marked as covered. If a different entry wins, the card tells you which one — that usually means two entries are splitting the same question and should be merged. This never touches the entry's usage count.
📥收件匣 — 所有待辦事項,集中一個清單。 上方的分類按鈕會顯示每類的數量,次序就係建議處理嘅次序:Waiting 等候中(有人正在等回覆)、Unchecked answers 未覆核答案、Missing answers 未有答案、Picked follow-ups 揀選咗嘅跟進建議、Duplicates 重複、Weak entries 質素欠佳、Matched, unchecked 搜得到但未覆核、Untagged 未分類、Unchecked wording 措辭未覆核、Never tested 未測試。它們最終都指向同一件事——把答案寫入知識庫。由上而下處理,完成的項目會自動消失。
✅Unchecked answers 未覆核答案 — AI 已經發送咗嘅答案。 呢部分最容易誤會:呢啲答案已經俾咗客人。當時 AI 信心一般,只係參考咗一條部分吻合嘅記錄作答,之後冇人檢查過。每張卡會顯示發出去嘅答案,以及答案來源——即係 AI 當時參考咗邊條知識庫記錄。冇人喺度等你。若果答案冇問題,撳「NO CHANGE NEEDED」,佢就會離開清單,唔會再發任何嘢,知識庫亦唔會改動。若果答錯咗,撳「FIX & SAVE TO KB」寫低正確答案,令下一個問同樣問題嘅人直接得到正解。無論點揀,都唔會再次聯絡嗰位客人。
🕑歷史記錄 — 所有對話的永久存檔。 排隊項目處理後不會消失,而是移至此處。可查看完整對話、AI 草稿與客服回覆的對比,並可點擊「+ 儲存至知識庫」將好的答案補錄入庫。
📚知識庫 — 永久問答資料庫。 記錄會一直保留,除非你按「DEL」刪除。「EDIT」會直接更新內容,記錄不會消失。隨著記錄增多,可用搜尋框快速定位。目標是建立 50 條以上常見問題,記錄越多,排隊越少。
🏷️Topics 主題 — 記錄點樣分類,而且你可以自己加。 一條記錄可以有幾個主題,唔係淨係一個:一條關於邀約嘅問題,可以同時屬於 Training 同 Invitation。個清單唔係固定嘅——任何訓練員都可以新增主題,本身就係設計成隨住你寫記錄而不斷增加。收件匣入面嘅 Untagged 未分類 就會列出冇任何主題嘅記錄。改名同封存主題就只有管理員先做得,因為改一次名會影響晒所有用緊嗰個主題嘅記錄。
⚡分析及整理 — 最快捷的訓練方式。 貼上培訓文稿或上傳 .docx/.pdf/.txt 文件,AI 自動提取問答對供你審閱後批量匯入。
🔍測試知識庫搜尋 — 驗證記錄是否有效。 輸入用戶可能問的問題。≥58% = AI 會有信心咁答。35–58% = 邊緣,會參考部分吻合嘅記錄作答。<35% = AI 答唔到,會排隊等人手回覆。新增前可測試是否已有重複,新增後可確認記錄能被找到。
⭐H / M / L 質量投票 — 非強制但有用。 任何人均可投票。H(高) = 準確且表達清晰,搜尋優先。M(中) = 尚可。L(低) = 需改善。不必對每條記錄投票,發現特別好或特別差的才需標記。
🔁重複問題保護。 若新增的問題與現有記錄相似度超過 88%,系統會提示並顯示現有記錄,供你選擇保留兩者、更新現有答案或取消,避免重複記錄堆積。
👤訓練員與管理員的分別。 所有人儲存後都即時生效。訓練員新增或修改嘅記錄,會同時出現喺管理員收件匣嘅「訓練員修改」,顯示修改前後,管理員得閒時撳「LOOKS FINE」或「REVERT」。REVERT 會還原到訓練員修改之前嘅版本;如果係訓練員新增嘅記錄,就會刪除。任何標示為「Admin only」嘅記錄——合規同其他受規管字眼——只有管理員先可以修改。
🎯Never tested 未測試 — 證明條記錄真係搵得到。 一條記錄配對到自己條問題,其實證明唔到乜。撳「Never tested」,用member 真係會咁問嘅講法打一次——要用唔同嘅字眼,唔好抄返記錄本身嗰句。如果搜尋有信心咁搵到嗰條記錄,就會標記為已覆蓋。如果贏嘅係另一條記錄,張卡會話你知係邊一條——咁通常即係兩條記錄分薄咗同一條問題,應該合併。呢個測試唔會影響記錄嘅使用次數。
📥收件箱 — 所有待办事项,集中在一个列表。 上方的分类按钮会显示每类的数量,顺序就是建议处理的顺序:Waiting 等待中(有人正在等回复)、Unchecked answers 未复核答案、Missing answers 未有答案、Picked follow-ups 选中的跟进建议、Duplicates 重复、Weak entries 质量欠佳、Matched, unchecked 搜得到但未复核、Untagged 未分类、Unchecked wording 措辞未复核、Never tested 未测试。它们最终都指向同一件事——把答案写入知识库。由上而下处理,完成的项目会自动消失。
✅Unchecked answers 未复核答案 — AI 已经发送过的答案。 这部分最容易误会:这些答案已经给了客人。当时 AI 信心一般,只参考了一条部分吻合的记录作答,之后没有人检查过。每张卡会显示发出去的答案,以及答案来源——即 AI 当时参考了哪一条知识库记录。没有人在等你。如果答案没问题,点击「NO CHANGE NEEDED」,它就会离开列表,不会再发送任何内容,知识库也不会改动。如果答错了,点击「FIX & SAVE TO KB」写下正确答案,让下一个问同样问题的人直接得到正解。无论怎么选,都不会再次联系那位客人。
🕑历史记录 — 所有对话的永久存档。 队列项目处理后不会消失,而是移至此处。可查看完整对话、AI 草稿与客服回复的对比,并可点击「+ 保存至知识库」将好的答案补录入库。
📚知识库 — 永久问答数据库。 记录会一直保留,除非你按「DEL」删除。「EDIT」直接更新内容,记录不会消失。随着记录增多,可用搜索框快速定位。目标是建立 50 条以上常见问题,记录越多,队列越少。
🏷️Topics 主题 — 记录如何分类,而且你可以自己添加。 一条记录可以有多个主题,不只是一个:一条关于邀约的问题,可以同时属于 Training 和 Invitation。列表不是固定的——任何培训员都可以新增主题,本来就是设计成随着你写记录而不断增加。收件箱里的 Untagged 未分类 会列出没有任何主题的记录。重命名和归档主题只有管理员可以做,因为改一次名会影响所有使用该主题的记录。
⚡分析及整理 — 最快捷的训练方式。 粘贴培训文稿或上传 .docx/.pdf/.txt 文件,AI 自动提取问答对供你审阅后批量导入。
🔍测试知识库搜索 — 验证记录是否有效。 输入用户可能问的问题。≥58% = AI 能自信回答。35–58% = 临界,会参考部分吻合的记录作答。<35% = AI 无法回答,会排队等人工回复。新增前可测试是否已有重复,新增后可确认记录能被找到。
⭐H / M / L 质量投票 — 非强制但有用。 任何人均可投票。H(高) = 准确且表达清晰,搜索优先。M(中) = 尚可。L(低) = 需改善。不必对每条记录投票,发现特别好或特别差的才需标记。
🔁重复问题保护。 若新增的问题与现有记录相似度超过 88%,系统会提示并显示现有记录,供你选择保留两者、更新现有答案或取消,避免重复记录堆积。
👤培训员与管理员的区别。 所有人保存后都即时生效。培训员新增或修改的记录,会同时出现在管理员收件箱的「培训员修改」,显示修改前后,管理员有空时点击「LOOKS FINE」或「REVERT」。REVERT 会还原到培训员修改之前的版本;如果是培训员新增的记录,则会删除。任何标示为「Admin only」的记录——合规及其他受监管字眼——只有管理员才可以修改。
🎯Never tested 未测试 — 证明记录真的能被找到。 一条记录匹配到自己的问题,其实说明不了什么。点击「Never tested」,用会员真正会问的说法输入一次——要用不同的字眼,不要照抄记录本身那一句。如果搜索有信心地找到该记录,就会标记为已覆盖。如果赢的是另一条记录,卡片会告诉你是哪一条——这通常表示两条记录分薄了同一个问题,应该合并。此测试不会影响记录的使用次数。