AI auto-reply

AI auto-reply is a workspace-level receptionist for customer-initiated WhatsApp, Instagram, Messenger, and Telegram conversations. It uses your active services and approved knowledge to answer routine enquiries, ask a clarifying question, or hand the conversation to a teammate.

It does not create services, change prices, update opportunities, or make other CRM decisions by itself.

Before you enable it

Prepare the workspace first:

1. Add accurate, active services under Settings → Services.
2. Add service descriptions and matching aliases that reflect phrases customers actually use.
3. Decide who should receive handoffs.
4. Add approved facts and FAQs for information that is not already clear in the service catalogue.
5. Start in Draft for approval mode and review the first replies before considering automatic sending.

Caution

AI can only use the services and knowledge available in the selected workspace. Keep prices, descriptions, policies, availability guidance, and handoff instructions current.

Configure the Receptionist

Owners and Admins can open Settings → Agents, then select the Receptionist card to open its settings. Only agents with available settings appear as cards. Choose All agents to return to the overview. The Automation page links to the agent overview. Your existing settings, approved knowledge, and activity remain available here; moving to this page does not enable replies or change the selected mode.

1. Choose a mode

• Draft for approval: Qonvera prepares a suggested reply. A teammate reviews, edits, sends, discards, or hands it off.
• Automatic send: Qonvera sends a reply when it meets your safeguards, without waiting for approval. Unsafe, uncertain, or unsupported requests are left for a teammate.

Use Draft mode for the initial rollout and after major changes to services or approved knowledge.

2. Select channels

Choose one or more connected channels:

• WhatsApp;
• Instagram;
• Messenger;
• Telegram.

You can select all four channels together, then choose Save Receptionist settings. AI channel selection does not change your subscription’s connected-channel limits.

The provider account must remain connected. Before generating an Instagram reply, the Receptionist checks authorization. If authorization has expired, needs reconnection, or cannot be verified, it stops before generating a new answer. Owners and Admins can review the connection warning in Settings → Channels. Restoring the connection does not automatically resend earlier failed replies. Instagram and Messenger also require an open customer-initiated reply window. Qonvera does not use AI to bypass provider messaging rules.

3. Set the schedule

Choose when AI may respond:

• Always;
• During hours;
• Outside hours.

Then choose the Schedule timezone from the worldwide list and set active weekdays, start and end times, and any excluded dates. Choose the region and city that follow your operating hours; daylight-saving changes are applied automatically where observed. UTC is also available. This schedule timezone is independent of the organization timezone in General settings. Select Save Receptionist settings to apply changes. Exclusion dates are useful for public holidays or planned closures.

An overnight period is allowed. For example, an outside-hours schedule can begin in the evening and continue into the following morning.

4. Set reply safeguards

Configure:

• Response delay: how long Qonvera waits after the inbound message before processing it;
• Maximum consecutive replies: prevents an extended AI-only exchange;
• Pause after human reply: keeps AI quiet after a teammate responds;
• Confidence threshold: low-confidence results are not sent automatically;
• Daily and monthly budgets: cap AI usage for the workspace;
• Languages: languages the AI may handle;
• Brand tone: concise guidance for the wording and style of replies.

Higher confidence thresholds are more conservative. If the model cannot produce an answer that passes the configured safeguards, Qonvera asks for clarification, hands off, abstains, or leaves the conversation for a teammate.

Budget already used is retained when settings change. If a request’s cost cannot yet be confirmed, Qonvera keeps an allowance for it; that allowance can continue to count against a later day or month until the cost is resolved. Changing the response schedule timezone does not reset the budget or change the budget’s established day boundary.

5. Limit service knowledge

Under Allowed services, keep Use all active services selected to make your active catalogue available. To restrict the Receptionist, turn it off and select the services it may discuss. With that option off, an empty selection makes no services available.

The AI can discuss only the eligible services in your catalogue. It cannot add a new service.

6. Configure handoff

Choose a Primary handoff owner or at least one teammate in the Handoff team before enabling the feature.

Also configure:

• the message shown when a human will continue the conversation;
• topics that should always go to a person, such as complaints, refunds, legal matters, medical advice, or other business-specific risks.

Qonvera hands off when a configured topic matches, the AI cannot answer confidently or in a supported language, an unsupported message needs review, or the consecutive-reply limit is reached. You can also choose Handoff on a pending draft.

A handoff pauses AI for that conversation and assigns the primary owner if their workspace membership is active. Otherwise, Qonvera selects the first active member of the configured handoff team. This is not round-robin assignment or a check of who is online. Opportunity and follow-up owners do not change.

The assigned teammate receives an in-app notification linking to the conversation. If no configured recipient is active, the conversation still pauses and the Inbox shows that someone must assign it manually. Owners and Admins can also find open handoffs under Operations.

Customer handoff message

• In Automatic mode, Qonvera attempts to send the saved handoff message once, subject to the channel’s reply window and connection state.
• In Draft mode, an automatic handoff pauses and assigns the conversation, but does not send anything. Review the message in the Inbox and choose Send handoff message.
• Choosing Handoff yourself on a pending draft authorizes sending the saved handoff message. The AI-generated draft is not sent instead.

The Inbox shows whether the message is awaiting approval, pending, accepted by the channel, not sent, or has an unconfirmed outcome. Channel acceptance is not proof that the customer read it. An uncertain outcome is not automatically retried, to avoid duplicate messages; check the conversation before replying manually. A teammate’s intervening reply or resuming AI cancels an acknowledgment that has not started sending.

AI stays in human-handoff mode until a teammate chooses Resume AI. Resuming closes the handoff and discards old queued work and pending drafts; it does not replay the customer’s previous message. New inbound messages follow the saved schedule and safety rules.

7. Save and enable

Choose Save Receptionist settings. If the feature is enabled, at least one channel and one handoff recipient are required.

If Qonvera says AI replies are unavailable, you can prepare the settings, but replies will not run yet. Contact Qonvera support to check availability.

Add approved facts and FAQs

Use Approved facts and FAQs for trusted answers that are not suitable as service descriptions.

Examples include:

• opening hours;
• supported locations;
• booking requirements;
• standard onboarding steps;
• an approved answer to a frequent customer question.

Choose Frequently asked question or Approved fact. Give each entry a Reference key, such as opening-hours, and write its approved answer. FAQs also require a customer question; facts do not. The key starts with a letter or number and can contain letters, numbers, dots, hyphens, underscores, or colons (up to 120 characters). Questions allow up to 500 characters and answers up to 4,000. Avoid temporary information unless someone will keep it updated.

Changes to the AI configuration are marked Unsaved changes until you select Save Receptionist settings. Refreshing activity or adding knowledge does not discard those unsaved settings.

AI receives only relevant approved entries. Adding an entry does not guarantee it will be used in every reply.

Work with AI replies in the Inbox

Open a supported conversation in Inbox. The AI panel shows whether the conversation is monitored, paused, handed off, or waiting for draft review.

In Draft mode, a teammate can:

• review the proposed reply;
• edit and send it;
• send it unchanged;
• discard it;
• hand the conversation to a person.

Teammates can also pause or resume AI for the selected conversation. A human outbound reply automatically takes priority and prevents the AI from competing with the teammate.

While a draft action is running, competing actions are disabled. Clearing the draft leaves it empty and prevents sending. If another update changes the same draft while you are editing, Qonvera keeps your text and asks you to load the latest draft before continuing. Copy any edits you want to retain first.

Qonvera rechecks recent human activity before sending. If a teammate replied while the AI was preparing an answer, the AI reply is cancelled rather than sent into the same exchange.

What AI can and cannot do

AI auto-reply can:

• answer using active service information and approved knowledge;
• ask a clarifying question;
• identify relevant configured services;
• follow the configured tone and supported languages;
• hand the conversation to the configured owner or team.

It cannot:

• invent services, prices, discounts, availability, or policies;
• create or change contacts, accounts, opportunities, stages, tasks, or quotes;
• send outside Meta reply windows;
• reply through a disconnected channel;
• continue when cost limits or the maximum reply count have been reached;
• answer prohibited or configured high-risk topics automatically.

Monitor activity

Owners and Admins can review Operations at the bottom of the AI auto-reply settings. It summarizes pending replies, failures, handoffs, and sent replies. Use Refresh to load the latest state.

Some temporary connection problems are retried automatically. Replies stopped by messaging rules, a teammate’s reply, or an unclear send result are not automatically sent again. Check the conversation and final status before replying yourself.

Review interrupted attempts

Owners and Admins can open Settings → Agents → Receptionist → Interrupted attempts. This lists older work whose outcome still needs review.

1. Choose Open conversation and review the customer’s channel before sending anything new.
2. Select Check saved evidence. Qonvera records a confirmed delivery when matching saved evidence exists. If an attempt is known never to have received permission to send, it can be closed without a message. An expired job can also be closed without rerunning it.
3. If delivery remains uncertain, you may choose Close without resending and confirm the warning. This permanently prevents replay of that attempt. It does not mean the message was delivered or that nothing was sent.

Recovery never sends a message or releases reserved AI costs. Closing an attempt does not resume AI or change the assigned handoff teammate. Recent attempts cannot be closed while they may still be running; refresh after they finish. A message already in transit may still arrive.

Receptionist messages cannot use the Inbox’s general failed-message retry action. The Inbox shows Review Receptionist to Owners and Admins when Receptionist settings are available; other teammates should ask an Owner or Admin to review the attempt. Use this review workflow instead of resending. If a previously opened page still shows Retry and the request is refused, refresh the page and open Receptionist settings. A failed label alone does not prove that the provider received nothing. If you need to follow up manually, check the conversation in the customer’s channel first to avoid duplicating a message.

Test safely

Use this rollout sequence:

1. Enable one connected channel in Draft mode.
2. Send a fresh customer-initiated message from a test account.
3. Wait for the configured response delay.
4. Open the conversation and review the draft.
5. Confirm that the answer uses only the correct service and approved facts.
6. Edit and send, then confirm the customer receives one message.
7. Test a vague question and confirm Qonvera asks for clarification or hands off.
8. Send a human reply before the delay ends and confirm the pending AI action stops.
9. Review Operations for the resulting sent, handoff, or failed status.
10. Consider Automatic mode only after the workspace passes the draft review.

Use a fresh customer message for each new test and check that no earlier reply is still pending.

Troubleshooting

The AI panel does not appear

• Check with your Owner or Admin that AI auto-reply is available and enabled for the workspace.
• Confirm that the conversation uses WhatsApp, Instagram, Messenger, or Telegram.
• Refresh the page after the feature has been enabled.

Saving settings returns an error

• Select at least one channel before enabling.
• Choose a primary handoff owner or at least one handoff teammate.
• Choose the correct Schedule timezone from the list, such as Asia/Dubai, and save your settings.
• Check start and end times, language entries, dates, confidence, and budget values.
• Read the validation message displayed beside Save Receptionist settings.
• If the message says setup is incomplete, contact your administrator or Qonvera support. Reloading settings alone will not complete setup.
• If valid settings still cannot be saved, share the visible error with Qonvera support.

No reply or draft is created

• Confirm the workspace configuration is enabled.
• Confirm the current day and time match the schedule.
• Check that the channel is active and the provider reply window is open.
• Check whether the conversation was paused or recently handled by a teammate.
• Confirm the customer used a supported language and the workspace has relevant active services or approved knowledge.
• Review Operations for a skipped reply, handoff, or failure.

A reply failed temporarily

Qonvera may retry temporary connection problems. Do not send repeated test messages while a retry is pending. Refresh Operations and review the final result before testing again.

A draft’s delivery is not confirmed

Check the conversation and the customer’s channel before sending another message. Qonvera does not automatically send the same reviewed draft again when the result is unclear. Refresh the Inbox and ask an Owner or Admin to review Interrupted attempts. If the result remains unclear, share the channel, approximate time, and visible status with support. Do not share access tokens or private customer messages.

Settings or approved knowledge changed while editing

Another teammate may have saved changes. Keep a copy of any edits you need, then reload the saved settings and review the latest values before saving again. Reload saved settings replaces your unsaved settings edits. When replacing approved knowledge with the same reference key, refresh first so you review the current entry.

The answer is incorrect or too vague

• Correct the service name, description, price, aliases, or workspace availability.
• Add or update an approved FAQ.
• Raise the confidence threshold for more conservative behavior.
• Add the topic to Always handoff topics when a human should handle it.
• Return to Draft mode while correcting the workspace knowledge.

Permissions

• Owner and Admin: configure AI auto-reply, manage approved knowledge, and review operations.
• Manager and Agent: work with eligible conversation state and drafts from the Inbox, subject to their normal workspace and conversation permissions.

Every material draft decision records who acted. AI-generated messages are also marked separately from human replies so audit and pause behavior remain reliable.