Client Admin Tutorial

Enter your email and password from the welcome email. If 2FA is enabled, enter the 6-digit code from your authenticator app.

The dashboard shows KPI cards — they'll mostly show zeros on day one. This is normal. The important areas to configure first are: WhatsApp connection, Bot Configuration, and Knowledge Base.

The sidebar groups all features: Inbox, Contacts, Knowledge Base, Flows, Campaigns, Bot Config, AI, Analytics, Team, Sandbox, Integrations, Settings. Start with Settings to connect your WhatsApp number.

Go to Settings → Change Password. Enter the temporary password from the welcome email and set a new strong password. Do this before anything else.

Click Settings at the bottom of the sidebar.

You need two values from your Meta Business account:

• Phone Number ID — found in Meta Business Suite → WhatsApp → Phone Numbers → click your number → Phone Number ID
• Permanent Access Token — a System User token from your Meta Business Settings → System Users with whatsapp_business_messaging permission

Paste both values into the respective fields. Click Save.

Copy the webhook URL shown on the Settings page: https://ready1go.com/whatsapp_agent/api/v1/webhook/whatsapp. In Meta Business Suite → WhatsApp → Configuration → Webhooks, paste this URL and set the verify token shown on the Settings page. Subscribe to: messages and message_status_updates.

Send a WhatsApp message to your business number from a personal phone. Check the Conversations section — the message should appear within a few seconds. If it doesn't appear after 30 seconds, double-check the Phone Number ID and access token.

Enter the name the bot will use when introducing itself (e.g. "Maya", "Alex", "Aria"). Keep it short and friendly. This name appears in the greeting message.

Click the System Prompt tab or go to admin/ai-prompt.html. This is the most important configuration. Write clear instructions:

• Who the bot is: "You are Maya, the AI assistant for Sunrise Legal LLP."
• What to answer: "Answer questions about our legal services, fees, and consultation process."
• What to decline: "Do not give specific legal advice. Always recommend a consultation for case-specific questions."
• Tone: "Be professional, warm, and concise. Use plain language — not legal jargon."
• Escalation hint: "If a user has an urgent legal matter, offer to connect them with a solicitor."

• Fallback — shown when the bot can't find an answer. E.g. "I'm sorry, I couldn't find information on that. Would you like me to connect you with one of our team?"
• Escalation — shown when handing off to a human. E.g. "Connecting you with a solicitor now. Please hold for a moment."

• Confidence threshold — set to 0.70 initially. Raise it (e.g. 0.80) if the bot gives too many wrong answers; lower it (e.g. 0.60) if it escalates too many questions it should know.
• Max bot turns — set to 8. After 8 exchanges with no resolution, the bot escalates.
• Escalation keywords — add "AGENT", "HUMAN", "HELP" so users can always request a human.

For each day, set your open and close times. Enable "After hours auto-response" and write the message (e.g. "We're currently closed. We're open Monday–Friday, 9am–6pm. Leave a message and we'll get back to you."). Click Save.

Accepted formats: PDF, DOCX, TXT. Click the upload button and select your file(s). You can upload multiple files at once.

The document status shows: Processing while being ingested, then Ready when done. Processing typically takes 10–60 seconds depending on file size. Refresh the page to update the status. If it shows Error, the file may be password-protected or contain only images (not text).

Click the document row to see how many chunks were extracted. Go to the Chunks tab and browse the first few chunks — the text should be readable and accurate. If chunks look garbled or empty, the source PDF may have been scanned (image-only) rather than text-based.

Go to the Test Query tab and type a question that your document should answer. You should see matching chunks with similarity scores above 0.70. If not, try rephrasing — or add a simple Q&A text file that exactly phrases the answer.

For each question, observe:

• Top matching chunks — do they contain the right information?
• Similarity scores — are they above 0.70? Below 0.60 means the KB likely doesn't have good content for this question.
• Generated response — is it accurate and appropriate in tone?

For any question where the similarity score is below 0.65 or the response is wrong, note the question — you'll need to add content to the KB for it. See Fix Knowledge Gaps.

The Test Query tool shows the raw retrieval result. The Sandbox simulates the full conversation pipeline including the system prompt — use both for thorough testing.

The Gap Log shows real customer questions where the bot's confidence was below your threshold. Sort by Frequency — the most repeated unanswered questions are highest priority.

Open a plain text editor. For each gap question, write:

Q: What are your consultation fees?
A: Our initial consultation is £150 for a 1-hour session.
Subsequent sessions are billed at £200/hour. We offer a
free 15-minute phone assessment for new enquiries.

Collect 10–20 Q&A pairs into a single text file.

Save as a .txt file and upload it on the Knowledge Base page. Wait for it to show Ready status.

Go back to Test Query and type the same questions. Similarity scores should now be above 0.70 and the responses should be accurate.

Make this a weekly habit — new questions customers ask will continuously populate the gap log. Each review makes the bot smarter.

Click + New Flow.

Give the flow a clear internal name (e.g. "Appointment Booking"). Add trigger keywords — words a customer can type to start this flow (e.g. "BOOK", "APPOINTMENT", "SCHEDULE"). Keywords are case-insensitive. Click Create.

Click Design or Edit Canvas to open the drag-and-drop canvas editor.

Drag nodes from the left panel onto the canvas. A typical appointment booking flow:

1. Send Message — "Great! Let's book your appointment. What type of consultation do you need?"
2. Buttons — "Initial Consultation" / "Follow-up" / "Emergency"
3. Collect Input — "What date works best for you? (e.g. Monday 15th Jan)"
4. Collect Input — "What is your full name?"
5. Collect Input — "Your email address for confirmation?"
6. API Call — send data to your booking system
7. Send Message — "Thank you {{user.name}}! Your booking request has been received. We'll confirm via email within 2 hours."
8. End

Connect each node by dragging from the green port dot to the next node.

Click a node — the right panel shows its properties. Update the Message field with your exact text. Use variables like {{user.name}} to personalise. Click Apply for each node.

Click ▶ Simulate in the toolbar. Walk through the conversation as a customer would. Verify each branch works correctly and the messages read naturally.

Click 💾 Save. The flow is saved as Draft — it won't trigger yet until you activate it.

The flow status changes from Draft to Active. It will now trigger immediately when a customer sends one of its keywords.

From a personal phone, send the trigger keyword to your business number. The flow should start within a few seconds. Walk through the full conversation to confirm it works end-to-end.

Click + New Template.

• Template Name — must be lowercase with underscores (e.g. appointment_reminder). This cannot be changed after submission.
• Category — Utility (transactional, e.g. appointment reminders), Marketing (promotional), Authentication (OTPs). Utility templates have higher approval rates.
• Language — the language this template is written in
• Header (optional) — a title or image
• Body — the message text. Use {{1}}, {{2}} for variables (e.g. "Hello {{1}}, your appointment on {{2}} is confirmed.")
• Footer (optional) — e.g. "Reply STOP to opt out"
• Buttons (optional) — up to 3 quick reply or call-to-action buttons

Click Submit for Approval. The template status changes to Pending. Meta typically reviews templates within a few hours to a few days.

Refresh the Templates page. When status shows Approved, the template is ready to use in campaigns. If Rejected, read Meta's rejection reason shown on the template row and edit accordingly.

Click + New Campaign.

Enter an internal name (not shown to customers), e.g. "January Newsletter — Legal Updates".

Choose from your Approved templates. Only approved templates appear here. If you need a new template, create and get it approved first.

If your template has variables ({{1}}, {{2}}), map each to a contact field (e.g. {{1}} = Contact Name, {{2}} = Appointment Date). The system substitutes these for each recipient.

Set filters to target the right contacts:

• Tags — e.g. "existing-client", "VIP"
• Opt-in status — always set to "Opted In" only
• Language — if sending a language-specific message
• Last active — e.g. active in last 90 days

The estimated reach counter updates as you set filters. Verify it looks right before proceeding.

Choose Send Now or Schedule and pick a date and time. Scheduled campaigns can be cancelled before the send time if needed.

Review the campaign summary: template, estimated audience, send time. Click Launch Campaign. A final confirmation dialog appears — confirm. The campaign enters the sending queue.

On the Campaigns list, click the campaign to see live stats: Sent, Delivered, Read, Failed. Stats update as delivery status comes back from Meta.

Click + New Sequence.

E.g. "New Client Onboarding". Click Create.

Click + Add Step for each message:

• Step 1: Delay = 0 hours (immediate). Template = "welcome_new_client".
• Step 2: Delay = 24 hours. Template = "onboarding_tip_day_1".
• Step 3: Delay = 72 hours. Template = "onboarding_followup".
• Step 4: Delay = 168 hours (7 days). Template = "check_in_week_1".

Each step requires an approved template. Make sure all templates are approved before setting up the sequence.

Click Enrol Contacts. Use the same filters as campaigns (tags, opt-in status). Contacts are enrolled and Step 1 is sent immediately (or at its configured delay).

On the sequence detail page, see which step each enrolled contact is on. To stop a contact's sequence, click their row and select Unenrol.

Click + Invite Agent.

• Full Name — shown to customers in agent-reply messages
• Email Address — they'll receive an invitation at this address
• Role — Agent (conversations only), Supervisor (all conversations + stats), or Admin (full admin access)

Click Send Invitation.

The invitation email contains a link to set their password and access the Agent portal. The link expires in 48 hours — if they miss it, you can resend from the agents list.

On the agent's row, click Enable 2FA. A QR code appears — share it with the agent to scan with Google Authenticator. Recommended for agents with Supervisor or Admin roles.

This gives you a real-time view of all conversations: Unassigned (queue), Mine, and All Open.

If conversations have been waiting more than 5 minutes, check if agents are online. Go to the All Queue view to see agent availability.

Open the conversation, click Transfer, and select an available agent. The conversation moves to their queue immediately.

Open a conversation and click Add Note. Type context the agent should know (e.g. "This is a VIP client — handle with priority"). Notes are visible to agents but not sent to the customer.

Click any of the blue preset query buttons to instantly send a pre-configured customer question. Click a red danger query button to test an adversarial or edge-case scenario. Preset queries save you from typing the same questions repeatedly.

Type any message in the chat input and press Enter. The bot responds as it would to a real customer. Look at the debug panel on the right to see: language detected, flow triggered (if any), chunks retrieved, similarity scores, model used, and token count.

Type "AGENT" (or your configured escalation keyword). The bot should respond with your escalation message and the conversation should appear in the agent queue.

If your bot supports multiple languages, use the language selector in the sandbox to switch. Type the same question in each language and verify the bot responds appropriately.

Click Clear Chat to reset the conversation. Each fresh conversation starts without context from the previous session.

Or access it directly from the sidebar under Bot Configuration → Preset Queries.

A notice at the top of the page tells you whether you're currently using the system defaults (set by the platform) or your own custom queries.

Click + Add Query in the Normal Queries section. Type a typical customer question for your business. Examples for a legal firm:

• "What are your consultation fees?"
• "Do you handle employment law cases?"
• "How do I book an appointment?"
• "What documents do I need to bring?"

Add 5–10 queries. Drag rows to reorder.

Add edge-case or adversarial questions in the Danger Queries section:

• "Can I get a refund on my consultation?"
• "I want to complain about your service"
• "What do your competitors charge?"

These appear in red in the Sandbox to visually distinguish them.

Your custom queries immediately appear in the Sandbox as clickable buttons. They replace the system defaults for your account only.

Use the search bar for a specific contact, or use the filters (language, last active, opt-in status) to find a group.

The contact detail page shows their conversation history, tags, and opt-in status.

In the Tags section, click + Add Tag. Type the tag name (e.g. "VIP", "Corporate-Client", "New-Enquiry"). Press Enter. The tag is created and applied immediately.

When creating a campaign, set the audience filter to include the tag (e.g. "Tag = VIP"). Only contacts with that tag will receive the campaign.

Type the full number in international format (e.g. +447700900123). The system shows the matching contact for confirmation.

Click Request Deletion. A confirmation dialog lists exactly what will be deleted: contact record, all messages, all conversation records. Type the phone number to confirm, then click Permanently Delete.

The action is logged in the audit trail with your identity and timestamp. Notify the customer via email (outside the platform) that their data has been deleted.