Every chat the assistant has is kept in your dashboard in full, including the point where it handed over to a person. Filter by feedback, date, unanswered questions, leads or source — or search the lot.
Each conversation is there to read as it happened: what the customer asked, what the assistant answered, and the moment it handed over to a person. Because the assistant answers 24/7 and in the customer's own language, that includes the chats nobody on your team was awake for.
Customers can rate an answer in the chat window, and the rating stays attached to that conversation. Filtering to the badly rated ones puts a review session straight in front of the answers your customers have already told you were weak.
Filter by rating, by date, by the questions the assistant could not answer, by the conversations that produced a lead, or by the source a chat came from. Full-text search finds a phrase — a product name, a policy word — across conversations when you are chasing something specific.
Questions the assistant could not answer surface as knowledge gaps, and reading the conversation is how you decide which are worth writing up. You change your own content; the assistant answers from the new version next time, with nothing to retrain or script.
Point Adoomi at your site so the assistant answers from your own pages, FAQs and approved content. That is also what makes review worth doing later: you are checking answers against content you approved, not against general knowledge.
You read what the assistant proposes to say and change anything you dislike, before it goes live. The conversations inbox is that same job repeated afterwards, on real questions from real customers instead of on your own test ones.
The chat window goes live and conversations are kept from the first one onwards. There is nothing to switch on for review, no tracking to configure and no sampling to set. Connecting the site, reviewing the answers and adding that line takes ~10 min in total.
Set the email alerts so flagged conversations reach the right person — an enquiry to whoever prices work, anything sensitive to a named person.
Questions the assistant could not answer surface as knowledge gaps, so a review session ends with a list of content to write rather than a vague worry. Limit: a gap tells you what was missing, not what to say about it; the new wording is yours, and the assistant only improves once your content does.
Ratings customers leave in the chat window roll up into helpfulness signals, and the same ratings filter the inbox so you can open the poor answers first. Limit: only customers who choose to rate leave one, so ratings cover a slice of your conversations and never all of them.
Routes the conversations that need a person — an enquiry — to the right teammate, so they are dealt with before anyone sits down to review. Limit: an alert is a notification, not a ticketing system, and there are no queues, assignments or reminders behind it.
Conversations that produced an enquiry are marked, so you can filter to them and read what the assistant said before the customer handed over their details. Limit: the mark tells you a lead came out of the chat; working that lead happens in your leads list and your own tools and your own tools.
Your teamMonday review: filtered to conversations customers rated badly over the weekend, opened the first one.
CustomerWhat do you charge for a same-day call-out on a Sunday?
AdoomiSame-day call-outs are covered on the site, but weekend pricing isn't in the content I've been given, so I won't guess at a figure. I can pass this to the office with your number and have someone confirm the cost — would that help?
CustomerI just wanted a price.
Your teamVerdict: the assistant was right to refuse — weekend pricing was never in the approved content, so inventing a figure was the only alternative. The gap is ours. Weekend and bank-holiday call-out charges added to the pricing page.
Your teamThe following weekend, same question from a different visitor:
AdoomiWeekend and bank-holiday call-outs carry a different call-out charge to weekdays, and the current figures are on our pricing page. If you'd like it confirmed for your postcode, I can pass your details to the office.
Nothing here rewrote the answer the first customer received: review is read-only, and the fix landed on the pricing page rather than in the transcript. The refusal to invent a weekend price was the guarantee working as intended; the review is what turned that refusal into a page that answers the question.
There is no editing a reply the assistant has already given, and no correction sent to the customer who received it. You fix the source content — a missing price, an out-of-date policy — and the next person asking gets the better answer. The conversation stays as the record of what happened.
You cannot watch a conversation in progress and jump in. When a customer needs a person, the assistant hands over instead: an enquiry with their details reaches your team by email, or the conversation carries on with them on WhatsApp. Review is what you do afterwards.
Rating an answer is optional and most people will not do it, so the rating filter surfaces a sample rather than a verdict on your assistant. An answer with no rating is not an answer that went well. The unanswered-question filter and full-text search are what cover the silent majority.
Review shows what happened, conversation by conversation. There is no automated quality scoring, no deflection reporting and no charts of resolution rate. Customer ratings roll up under helpfulness signals; everything else is you reading.
GDPR export and deletion apply to conversation data, so honouring a right-to-be-forgotten request removes that conversation from the inbox too. The record you were part-way through reviewing goes with it, which is the correct behaviour rather than a bug.
Every conversation the assistant has is reviewable in full: the customer's questions, the assistant's answers and the point where it handed over to a person. Nothing is sampled or trimmed, so you read the exchange as the customer had it.
Filters cover the rating a customer left, the date, questions the assistant could not answer, conversations that produced a lead, and the source a chat came from. Full-text search runs across conversations when you are hunting a specific word or phrase.
Editing a past answer is not possible; review is read-only oversight. You fix the underlying content — a missing price, an out-of-date policy — and the assistant answers from the new version next time. The old conversation stays as the record of what happened.
Adoomi's conversations inbox is not a live-chat console, so you cannot jump into a chat in progress. When a customer needs a person the assistant hands over: an enquiry with their details reaches your team by email, or the conversation continues with them on WhatsApp.
Conversation data is hosted in the EU, and GDPR export and deletion apply to it, so a right-to-be-forgotten request can be honoured. Search is built so query text never lands in a web address, conversation content is never written to logs, and the chat window sets no tracking cookies.
Reading costs nothing: one credit is one AI answer, so credits are spent when the assistant replies, never when you review. Plans are Free at £0 a month for 50 credits, Local £8 for 200, Growth £20 for 500 and Scale £75 for 4,000.
Adoomi starts with your public business knowledge and routes the moments that need a person to your team.