AI answers first. People step in with full context.

1. The customer calls or chats. CustoCare answers with your workspace's AI agent (for Accra Mobile Care, Adwoa).
2. The AI helps. It searches your Knowledge and uses tools: built-in ones (find the contact, check identity, create a case, schedule a callback, send a follow-up) and your own systems through the Integrations hub. Actions need the customer's confirmation; protected tools need a verified caller.
3. Resolved, or handed over. If the AI cannot finish, it transfers the conversation to a queue (for example when the caller asks for a person, or an escalation rule matches).
4. Routing offers it to a person. The best available agent with every skill the queue needs gets an offer on the Agent desk, with the transcript, summary and customer 360.
5. The agent finishes and wraps up. A disposition, notes and, if needed, a follow-up case with an SLA due time.
6. Everyone learns from it. Every conversation is saved under Interactions, scored by the AI quality review, and counted on the Wallboard and in Analytics.

1. Open your login email. Subject: "Your <company> login on KasadaAI CustoCare Call Center". It gives the sign-in page, your username (your email) and a password. Any role
2. Go to the sign-in page callcenter.custocareai.com/login, type Email and Password, click Sign in.
3. Choose your own password. With a temporary password you land on Choose a new password: fill Temporary password, New password (at least 8 characters), Repeat the new password, then Save and continue.
4. You land on your home page. Agents land on Agent desk, integrators on Integrations, everyone else on the Wallboard.
5. Change your password later from your name at the top right: My profileChange password (Current password, New password).
6. Sign out with the door icon at the far right of the top bar when you finish. This frees your account for your next sign-in.

• 1 Your company's name (or its logo and display name from Settings → Branding).
• 2 Sidebar sections: Contact centre, Customers, AI & integrations, Manage. You only see the pages your role may open.
• 3 Service box: "Connected to CustoCare" or "Not linked to CustoCare yet"; "AI agents are paused by billing" when billing stops the AI.
• 4 A small Demo pill appears only on a demo console (practice data, never live customers).
• 5 Presence (desk users only): Available, Busy, On break, Offline. Only Available gets new offers.
• 6 The headset icon opens the Kasada Dialpad, the browser softphone. Its dot shows Ready, Connecting…, Not connected or Off.
• 7 Your name and role: click it for My profile (name, phone extension, softphone password, languages, password).
• 8 Sign out. On a phone, the menu button at the top left opens the sidebar.

1. Find your home page. After sign-in you land on it (Agent desk, Integrations or Wallboard). Any role
2. Set up My profile: click your name at the top right, check Name, Phone extension, Languages you speak, then Save profile.
3. Try the theme button (moon or sun icon) for light or dark mode.
4. Click through every page in your sidebar once, so you know where things are before your first shift.

• 1 Presence in the top bar. "Go Available (top bar) to be offered new work." appears while you are not Available.
• 2 Inbox groups: Offered to you (Accept / Decline), Active, Wrap-up, Waiting in your queues (Pick up).
• 3 With the AI now: live AI conversations you may join with Pick up from AI.
• 4 Actions for the open conversation: Accept, Decline, Take over call, Hand back to AI (chats), Transfer, Take payment, Wrap up.
• 5 Reply box (chats only). For a call it says "This is a call: talk to the caller on the phone."
• 6 Side panel tabs: Customer 360, Lookups (tools you may run), Copilot. On a phone or tablet the desk shows tabs Inbox, Conversation, Customer & copilot.

Get ready (once)

1. Check your softphone. Click your name → My profile. If your admin assigned you a pool extension it is already there. Otherwise fill Phone extension and the softphone password (given by your admin), and Languages you speak; click Save profile. AgentSupervisor
2. Open the Kasada Dialpad (headset icon). It should say Ready · extension 4512. If it shows a reason such as "switched off by your admin", see module 18.
3. Go Available. Set the presence menu in the top bar to Available. Waiting work is offered to you at once.

Take a chat

1. Accept the offer under Offered to you within 45 seconds, or it moves to the next agent. Agent deskOffered to youAccept
2. Read before you type: the summary, the "Handed over:" reason and the transcript.
3. Reply. Your first reply takes the chat from the AI. Press Enter to send.
4. Hand back to AI if the AI can finish (for example to take a top-up), or Transfer it.

Take a call

1. Accept the voice offer. Your extension (the Kasada Dialpad) rings and is answered automatically; the AI leaves when you answer.
2. Joining a call the AI is still on: Pick up from AI under With the AI now, or Take over call on a call you already hold ("Ring your extension now; the AI leaves when you answer").
3. Call a customer back from Customer 360Call <number>: it rings your extension, then calls the customer. Or type a number on the Dialpad and click Call.
4. During the call the call card at the bottom right offers Mute, Keypad and End.

Transfer and wrap up

1. Transfer: TransferTo a queue → Queue, a Reason (shown to whoever picks it up), Transfer. Supervisors and admins also get To an agent.
2. Wrap up: click Wrap up. Choose the Disposition, write Notes, tap a Customer satisfaction (optional) score 1–5.
3. Need follow-up? Tick Create a follow-up case: Subject, Description, Type, Category, Priority.
4. Finish wrap-up. The conversation closes and you are free for the next one. When the customer leaves a chat it waits in Wrap-up: "The customer has left. Wrap up to finish."

1. Log a case CasesNew case: Subject, Description, Type, Category ("Sets the default priority and queue."), Priority ("Empty: from the category. Sets the due time."), Queue, Contact, Assignee → Create case. AgentSupervisorAdmin
2. Find your work: filter by status, priority, type, queue, SLA, or tick Assigned to me. Switch to Board for status columns.
3. Open a case to see Details, Timeline and Handling. Click Assign to me to take it.
4. Talk to the customer: add a comment on the Timeline. Tick Internal note (not shared with the customer) for team-only notes. Email the contact sends from your company's email and notes it on the timeline.
5. Change the status with Mark pending, Mark resolved, Mark closed or Reopen. Write the Resolution first: it is saved when you resolve or close.
6. Link another system in Reference in another system (for example a helpdesk ticket number).

1. Search Contacts: name, phone, email or account number; or filter by Tag.
2. Add one ContactsNew contact: Name, Phones, Emails, Tags, Tier (e.g. vip), Language, IDs in your systems (key=value, e.g. crm=AC-100482), Attributes (used for identity checks), Consent switches → Save. AgentSupervisorAdmin
3. Open a contact for Profile and Timeline. Buttons: Send email, Edit, Block/Unblock and Delete (Block and Delete: supervisors and admins).
4. Send email: choose To, optionally Note it on a case, write Subject and Message → Send email. It needs email set up in Settings → Email.

Automatic abuse blocking

CustoCare also protects your team without anyone clicking Block. Mild swearing from a frustrated customer is never acted on; insults aimed at the agent, harassment and threats are.

• On calls the AI warns once, politely, and ends the call if the abuse goes on. Blocked numbers never reach the AI.
• A block you set yourself is lifted with Contactsthe contactUnblock SupervisorAdmin.
• An automatic block that was a mistake: contact Kasada AI support with the customer's number.

1. Create a queue Queues & routingNew queue. Fill Name, Code (cannot change later), Description, Skills needed, Channels, Priority, SLA, Overflow, Active → Save. SupervisorAdmin
2. Choose overflow: Callback case, back to the AI, Voicemail (callback case) or Back to the AI.
3. Tune the Routing rules card (same page): Assign automatically, Offer timeout (s) (10–600), Overflow after (s) (empty = automatic), Wrap-up time (s), Calls at a time per agent (1–3). Admin (read only for supervisors)
4. Give agents skills Staff & teamsStaffEdit → Skills and levels (5 is expert), Chats at a time (1–20), Team. Supervisors may change skills, chats, presence and team. SupervisorAdmin
5. Make teams Staff & teamsTeamsNew team (Name, Description) for reporting and supervision. Admin
6. Point the AI at the queue: an AI agent's Default queue and its Escalation rules (module 8).

• 1 Greeting: name the company and the agent, nothing more.
• 2 Instructions (up to 20 000 characters): how the agent behaves, what it may and may not do.
• 3 Status Draft / Live / Paused, Languages (the first is the main one), Voice, Brand voice, Channels, Default queue, Phone numbers.
• 4 Save first, then Sync to CustoCare so the live agent gets your changes.

1. Create AI agentsNew AI agent: Name ("What the agent calls itself"), Slug (optional), Role → Create. It starts as a draft. Admin
2. Persona & channels tab: write the Greeting and Instructions; choose languages, voice, channels and default queue.
3. Phone numbers: add the Number and an optional Label → Add. "Calls to these numbers reach this agent after Save and Sync." Test numbers 7100–7199 work for trying it out.
4. Brand voice (optional): on Persona & channels, choose a cloned voice under Brand voice ("None (use the voice below)" keeps the normal Voice). To add one, click Clone a new voice…, fill Voice name and Whose voice it is, add the recordings, tick The speaker agreed to have their voice cloned for this business → Clone. A voice marked "(not ready)" cannot be chosen yet. Admin
5. Tools tab: leave All tools on, or switch it off and tick the tools this agent may use. AdminIntegrator
6. Escalation tab: Add rule. When: Intent is, Sentiment is, Caller says, Caller is VIP, Identity check fails, Caller asks for a person; then the Queue (empty: the default queue).
7. Prompt preview tab: read the Greeting and the Prompt CustoCare gets (persona, instructions, compliance, tool guidance, queues and escalation). Save first to see changes.
8. Go live: set Status to Live, click Save, then Sync to CustoCare. Check the Provisioning tab says "Synced".

• 1 Built-in tools carry a violet Built-in badge instead of a connector name, and cannot be deleted.
• 2 Record the outcome is off unless your industry template switched it on (for example politics).
• 3 The three payment skills stay off until an admin switches payments on in Settings → Payments (module 21).
• 4 Open a built-in tool to change only its Policies, its response and Enabled.

Identity check by SMS code

1. The AI starts the check before it shares or changes anything personal, or when a tool needs a verified caller. The code goes by SMS to the number on the contact. AI
2. No SMS possible? It asks the Identity questions instead: contact attributes such as dateOfBirth or ghanaCardLast4 (fill them in on the contact, module 6).
3. Set the rules SettingsCompliance: Identity questions, SMS code digits (4–8), Verification attempts (1–10), and Verify the caller before these tools (for Accra Mobile Care, get_case_status) → Save. Admin
4. Agents see the result on the desk: Agent deskCustomer 360Identity check shows Verified, Not verified with the number of attempts, or Verification locked (too many attempts). Agent
5. Send failed checks to a person: add the escalation rule Identity check fails on the AI agent's Escalation tab (module 8).

Follow-ups, outcomes and opt-outs

Switch a skill on or off

1. Open IntegrationsTools and click the built-in tool, for example record_outcome. AdminIntegrator
2. Tick Enabled (or untick it) and review the Policies (Needs a verified caller, Runs per conversation…).
3. Save tool. Then check the AI agent may use it: AI agentsAdwoaTools (All tools, or tick it).
4. Try it in IntegrationsTest console before a real customer meets it.

1. Write an article KnowledgeNew article: Title, Body ("Write facts the AI can quote: prices, hours, steps, policies."), Category, Tags, Status (Published / Draft / Archived), Audience → Save. SupervisorAdmin
2. Choose the audience: Everyone (AI and agents), or Agents only ("the AI never uses it") for internal procedures.
3. Send it to the AI: click Sync to AI. The number in brackets is how many articles changed since the last sync.
4. Retire old facts: open the article → Archive. Never leave out-of-date prices published.

Autopilot set-up (fast start)

1. Open Autopilot set-up Admin
2. Fill Company website, optionally API description and Anything else the AI should know. Click Read my website.
3. Review the draft: the company card, the AI agent greeting ("Update the AI agent"), Knowledge articles (tick the ones you want) and Tools from your API ("Create these tools").
4. Apply to my workspace. Articles are published and sent to the AI; API tools are created switched off: review them in Integrations and switch them on.

AI Lab

Apply an improvement

1. Open AI LabImprovements. Click Analyse now if you do not want to wait for Monday's review. Admin (supervisors and analysts read)
2. Read each suggestion: a knowledge article, an instruction change or a tool. Edit the Answer: "Replace every [fill in: …] before adding it." with your real facts.
3. Apply it (articles are published and sent to the AI; instruction changes are synced to the agent) or Dismiss it.

Run the Test lab

1. Open AI LabTest lab (digital twin) and choose the AI agent (Adwoa). AdminSupervisor
2. Create test customers (admins): six appear under Simulated customers, each with a persona, a mood and a goal.
3. Run all tests (supervisors and admins): "Test run started: simulated customers are talking to the agent".
4. Read the result under Test runs: how many are done, passed and failed, and the average Score.
5. Make it automatic (admins, card at the top of AI Lab): tick Run every test overnight, pick the hour in At, keep Weekly improvement review (Mondays) ticked → Save.

1. Add a connector IntegrationsConnectorsAdd from template (ready connectors with suggested tools; "Suggested for your industry" is marked) or Custom connector. Secrets are encrypted and never shown again. IntegratorAdmin
2. Test it: Test connection on the connector card shows Connected or Failed. For REST, Import API description drafts tools; for an MCP server, Import MCP tools.
3. Build or edit a tool IntegrationsToolsNew tool: cards What the AI sees (Name, Title, Description for the AI, Kind, Connector), the arguments, Operation, Response (Items path, Keep first, Pick, Max characters) and Policies.
4. Set the guardrails in Policies: Needs the customer's confirmation (on by default for actions), Needs a verified caller, Agents may run it from the desk, Confirmation read-back, Channels, Callers, Runs per conversation, Runs per hour, Timeout, Redact fields.
5. Try it safely IntegrationsTest console: pick the Tool, set the Conversation context (tick Caller verified if needed), fill the arguments → Run. Actions show Confirm and run. Disabled drafts can be tested too.
6. Switch it on with Enabled → Save tool, then give it to an agent: AI agentsAdwoaTools.
7. Watch it in IntegrationsRuns & stats: Tool performance (runs, success, p50, p95) and the Runs log (inputs redacted).

MCP & API keys

Your workspace is an MCP server: any MCP client with a key can use the same tools as the AI agents, through the same guardrails.

1. Open MCP & API keys. In Workspace MCP server choose Limit tools to an AI agent (or All enabled tools) and Copy URL. IntegratorAdmin
2. Create a key Service keysNew key: Name, tick the MCP scope → Create key.
3. Copy the key now: "This is the only time the secret is shown." Click Copy key, store it, then I have stored it.
4. Revoke a key you no longer need: anything using it stops working straight away.

• 1 The five types: REST / HTTP, GraphQL, Database (read-only SQL), MCP server, Webhook. The type cannot change after you create the connector.
• 2 Auth: how the connector signs in. Databases sign in with the user and password below, so Auth stays on No auth.
• 3 Engine: PostgreSQL, MySQL / MariaDB, Microsoft SQL Server, Oracle, MongoDB, Elasticsearch / OpenSearch. The usual port fills in for the engine.
• 4 Secrets: typed once, shown as set afterwards, never displayed again. The cross next to a secret marks it will be removed.
• 5 Create connector, then Test connection on its card.

Sign-in methods (Auth)

Pick the method your system's administrator gives you. Type the values only into the connector's own fields, in the console, never into a chat, an email or this course.

1. Open IntegrationsConnectorsCustom connector, choose the type, fill Name and pick the Auth. IntegratorAdmin
2. Fill the address fields (Base URL, GraphQL endpoint or Webhook URL), Timeout (seconds, ≤ 30) and, if needed, Allowed hosts and Headers (not secret).
3. Type the secret values in the Secrets box. They are encrypted and never shown again.
4. Create connector → Test connection. To change a secret later: Edit, type the new value, Save; leaving it empty keeps the old one.

GraphQL

1. Connector: type GraphQL, the GraphQL endpoint, and the sign-in method. IntegratorAdmin
2. Tool: IntegrationsToolsNew tool, choose the connector; the Operation card asks for the Query, Variables, Operation name (optional) and Data path (optional) ("Dotted path inside data, e.g. customer.orders.").
3. Shape the answer in Response (Items path, Pick, Summary (the spoken answer)) and test it in the Test console.

Databases: SQL Server, Oracle, PostgreSQL, MySQL

1. Connector: type Database (read-only SQL), Engine Microsoft SQL Server or Oracle, Host, Port, Database (Oracle: Service name, "as in host:port/service"), User, Max rows (≤ 1000), Use SSL; secret Database password. IntegratorAdmin
2. Tool operation: SQL (one SELECT), and Parameters for the values the AI fills in (for example the caller's number). Ask your database administrator to write the query.
3. Test in the Test console, then switch the tool on.

MongoDB and Elasticsearch / OpenSearch

Webhooks

1. Connector: type Webhook, the Webhook URL of your system, Auth HMAC signature with a Signing secret your system also knows. IntegratorAdmin
2. Tool: an Action with an Event name (e.g. callback.requested) and an optional Body (default: the tool's arguments).
3. Your system checks the signature on each call, then acts (for example creates a ticket). Keep Needs the customer's confirmation on.

1. Create CampaignsNew campaign. Start from a template if your industry has some, or Blank campaign. Fill Name, Purpose, AI agent, Starts at, What the AI should achieve, Calling from / until, Days, Time zone, Calls at the same time, Retry after (minutes), Attempts per number → Save. SupervisorAdmin
2. Import targets: open the campaign → Import targets → Upload .csv or .json or paste rows (a phone column; other columns such as {name} fill the script).
3. Start it (scheduled if Starts at is in the future). Pause stops calling; Run now places the next calls at once.
4. Follow progress: Targets, To call, Done, Failed, Skipped ("No consent"), the Targets table and Results.

Predictive outreach

Your systems send an event (a delivery is late, a bill is due) and the AI calls, messages or emails the customer at once. Case notices email customers when their case is created, updated or resolved.

1. Open Predictive outreach. Tick Outreach is on. Admin (supervisors and analysts can view)
2. Add a rule: Event, Title, How (AI phone call, SMS, WhatsApp, Email), Not again for (h), the message or "What the AI should do on the call", Active → Save.
3. Or start from a suggestion: next to Add a rule, one button per suggested rule your workspace does not have yet. Click one to add it, ready to edit. For Accra Mobile Care (telecom): Network outage in the area (SMS), Order or service delayed (AI phone call), and the case notices by email Case received, Case updated, Case resolved. Then Save.
4. Test it: type Your number for tests, save, then click the send icon on the rule.
5. Optional: tick Also call back customers who contacted us repeatedly without a fix.
6. Connect your system: an integrator or admin creates a service key with the Outreach events scope (MCP & API keys). See "How your systems send events" on the page.
7. Check Recent events: AI call queued, Message sent, Email queued, Case for the team, No rule, Skipped, Failed.

1. Watch the Wallboard Wallboard: Waiting now, AI containment today, Interactions today, Abandoned today, Average wait, CSAT today; each queue's SLA today (Breaching, Nobody signed in); Agents by presence; Live now with sentiment. SupervisorAnalystAdmin
2. Open the Live monitor Live monitor and select a live interaction. Its transcript, emotion badge and controls appear. SupervisorAdmin
3. Listen (silent) or Barge in (you join the call): your extension rings. Stop listening / Stop barging to leave.
4. Whisper to the AI: type guidance and Send. The AI follows it; the caller hears nothing.
5. Take over rings you, the AI leaves and the interaction opens on your desk. For chats use Take over chat: your first reply takes over.
6. Hang up only for abusive or stuck calls ("Hang up this call?"). It is recorded in the audit log.

Always up to date

Emotion levels

1. Spot it: on the Live monitor and the desk, a badge shows Annoyed, Frustrated or Angry. Hover over it to see the signals that were picked up. SupervisorAgent
2. Act yourself: Listen, then Whisper or Take over (above).
3. Or let the system act the first time a conversation reaches the level set in SettingsEmotion & escalation (module 19). Admin

Quality review on an interaction

1. Open Interactionsthe conversation and scroll to Quality review ("Reviewed automatically after the conversation ends."). SupervisorAnalystAdmin
2. Read the score out of 100: Good (80 and up), Fair (60 and up) or Needs attention, the summary, and each item marked Passed or Missed.
3. Re-run QA (top right of the card) after a correction, for example when a transcript was completed late. Only closed conversations can be reviewed: an open one says "The review runs when the interaction is closed."
4. Add supervisor notes: in Supervisor notes fill Tags, CSAT and Notes → Save. SupervisorAdmin
5. Check what the AI did: Tool runs lists each lookup and action with Source, Status, Result and Latency.

Coaching

• 1 Team this week (supervisors, analysts, admins): click a name to open that agent's report; "back to mine" returns.
• 2 The numbers: Quality score, Conversations, Customer rating, Average handling.
• 3 Your plan this week: praise, one Focus, and a role-play to Practise with a colleague.
• 4 Skills: 0–10, from the AI quality review of each conversation; strongest and weakest.
• 5 Moments: Your best conversation and One to learn from, each with a tip and Open the conversation.

1. Agents: open Coaching once a week and read Your plan this week. Agent
2. Open the conversation under Moments to hear or read what went well and what to change.
3. Practise the role-play with a colleague before your next shift.
4. Supervisors: scan Team this week for falling scores (Change) and the Work on column; open each agent and agree the focus in your one-to-one. Supervisor

The Compliance page

1. Open Compliance (/compliance) and choose the Period: 7 days, 30 days or 90 days. SupervisorAnalystAdmin
2. Read the four numbers: Compliance score (out of 10, with how many were reviewed), Customers verified, Stopped before verifying ("The AI tried a protected action too early and was blocked") and Sensitive numbers masked.
3. Check Masked in transcripts (what kinds of numbers were hidden) and Your rules (from Settings → Compliance; module 19).
4. Work through Conversations to review ("Compliance scored below 5 by the AI quality review."): When, Channel, Score, Verified. Open each, add supervisor notes, and coach the agent or fix the AI's instructions.
5. Export CSV for your auditors or regulator.

1. Open Analytics and pick 7 days, 30 days, 90 days or Custom (From / To, up to a year). SupervisorAnalystAdmin
2. Read the headline numbers (table below), then the charts: Volume by day, Volume by channel, Outcomes, Containment, Sentiment, Top intents, Handoff reasons, Dispositions, Cases by category.
3. Check people and tools: Per agent (Handled, AHT, Avg wait, CSAT, QA) and Tool success (Calls, OK, Errors, Success, Avg latency).

1. Invite Staff & teamsStaffInvite → Invite a staff user: Name, Email, Role, Team, Chats at a time, Extension, Languages, Skills and levels → Invite. Admin
2. Login email: the new person is emailed the sign-in page, their username and password. The console says "Login details emailed to <email>".
3. If it could not be sent the dialog says "We couldn't email the login. Share the password below with them." Copy the one-time password and give it to them privately (in person or by phone). It is shown only once.
4. First sign-in: they choose their own password (module 2).
5. Row actions on the Staff tab: Edit, Unlock (when Locked), Sign out, Reset password (type one, or leave empty for a temporary one), Deactivate. At the top: Sign out everyone (all but you).
6. Give a softphone: StaffEditAssign extension takes one from Kasada's pool, no password typing. Remove softphone frees it.
7. See where each person is signed in now: on the Staff tab, under each person's presence, a green line reads for example Signed in · Accra, GH · since 09:12, or Not signed in. Hover over it for the browser, IP address and when they were last seen. Sign out appears only for people who are signed in. Admin
8. See who signed in before: Staff & teamsSign-in activity: Time, User, Result, IP address, Location, Device. Admin

Open Settings (/settings). Sections are listed on the left (a drop-down on phones). Admins change them; supervisors read them. Changes apply to the AI agents and routing at once.

Softphone

1. Open SettingsSoftphone Admin
2. Tick Softphone on for our agents → Save. That is the only setting here: the phone system is set up for you.
3. If you see "Switched off by Kasada AI" or "Switched off by your partner", your agents have no softphone until it is switched back on, whatever you choose. Agents see the same reason on the Kasada Dialpad. Contact Kasada AI or your partner.
4. "The phone system is not set up yet." means Kasada AI has not finished it: contact Kasada support.
5. Give people extensions in Staff & teams (Assign extension) or they type theirs in My profile.

Email

1. Open SettingsEmail. Only admins see this section. Admin
2. Outgoing email: tick Send emails with these settings; fill Mail server, Port, Security (the recommended choice is marked; your email provider tells you which to use), Username, Password, From address (required), From name, Reply-to address → Save.
3. The password shows as Set afterwards, with Replace and Clear. It is never shown again.
4. Send a test email: type an address → Send test email. Unsaved changes are tested too; the server's error is shown (for example "Authentication failed"). At most 5 a minute.
5. Email log: every email, newest first; filter by Queued, Sent, Failed, Skipped; Resend failed ones.

Business hours and holidays

1. Open SettingsBusiness hours (/settings/hours). "Outside these hours (and on holidays) queues overflow straight away."
2. Time zone: leave empty to use the workspace's (for Accra Mobile Care, Africa/Accra).
3. Each day: switch the day on (open) or off (Closed), then set the opening and closing times.
4. Holidays: pick the date → Add. Remove one with the bin icon on its chip.
5. Save. Outside these hours the queue's overflow runs (usually a callback case, and the AI gives the caller the reference).

Channels and missed-call callbacks

1. Open SettingsChannels. Switch on the channels you serve: Voice, Web chat, WhatsApp, SMS, Email, USSD → Save.
2. Missed calls: on the card Call back customers who flash ("Many customers ring and hang up so you call them back. The AI calls them back within seconds."), tick Call back missed calls (flashes).
3. Set Call back after (seconds) (5–600, usually 20) and At most per number per day (1–10, usually 2) → Save.

Local languages

• 1 Tap each language your customers speak; your country's languages come first. Every one is spoken by the AI itself on calls and in chats (Voice & chat).
• 2 Two switches: code-switching (English mixed with a local language in one sentence) and landmark addresses.
• 3 Always say these in English: things that must stay exact. Empty means amounts, prices, dates, phone numbers and references.
• 4 Local words and names: the word, its meaning and its kind (Word, Phrase, Product / service, Place / landmark, Name). Below: Other words to listen for and Notes for the AI (600 characters).
• 5 Preview shows exactly what the AI is told, from the saved settings, and the words it listens for.

1. Open SettingsLanguages (/settings/languages) and tap your customers' languages.
2. Tick the code-switching and landmark switches if they fit your customers.
3. Keep in English: in Always say these in English add phrases such as amounts or account numbers (suggestions appear as you type).
4. Add local words: Add a word → word, meaning, kind (e.g. "momo" = mobile money). Add names and places to Other words to listen for, and short Notes for the AI.
5. Save, then click Preview on What the AI is told. "(nothing yet: choose languages or add words)" means nothing is saved yet.
6. Each AI agent's own Languages (AI agents → Persona & channels) decide which languages that agent answers in; the first is its main one.

Emotion & escalation

1. Open SettingsEmotion & escalation. Tick Watch live conversations for frustration.
2. Act when the customer is: Frustrated (recommended) or Angry (the levels are in module 14).
3. Choose the action: Tell the AI at once on calls (apologise, slow down, offer a person) and/or Send the customer to a person automatically.
4. Hand over to: a queue, or Default queue ("Empty: the AI agent's default queue.") → Save. The system acts once per conversation.

Settings → Compliance

1. Open SettingsCompliance (/settings/compliance).
2. Write the Rules, one per line. Your industry template already added some (module 22): read them and add your own.
3. Set the identity check: Identity questions, SMS code digits, Verification attempts, and the tools that need a verified caller.
4. Set retention: Keep data (days), and tick Legal hold only when your lawyers ask.
5. Save, then sync each AI agent (AI agents → Sync to CustoCare) so the rules reach the live agents.

• 1 Healthy, Needs attention or Not run: the last provisioning result.
• 2 The Widget step: "Created the web chat widget for …", "Added … to the web chat widget.", or "No website origin given." (skipped).
• 3 Web chat widget appears with Copy snippet once the widget exists: this is the paste code.
• 4 Website origin (for the chat widget): your site's address, starting with https://, without a page path.
• 5 Re-run provisioning (admins) creates the widget, or adds a new website to the one you have.

1. Open SettingsCustoCare (/settings/custocare). Admin (supervisors see the code but cannot re-run: "Only admins can re-run provisioning.")
2. Type your website origin in Website origin (for the chat widget), for example https://www.accramobilecare.example.
3. Click Re-run provisioning. Wait for "CustoCare provisioning finished" and check the Widget step is green.
4. Click Copy snippet next to Web chat widget.
5. Paste it into your website just before the closing </body> tag, on every page where the chat bubble should appear (or once in your site's shared footer or template). Your web designer can do this in two minutes.
6. Test it: open your website, click the chat bubble and send a message. The AI answers as "Chat with Adwoa" with "Hi! I'm Adwoa from Accra Mobile Care. How can I help?". The chat appears under Interactions with channel Web chat; ask for a person and it reaches the Agent desk.
7. Another website? Type its origin and Re-run provisioning again: the same widget is allowed there too, with the same snippet.

1. Read-back first: the AI repeats the amount, what it is for and the number, and waits for "yes". AI
2. The PIN never leaves the phone: the customer types it on their own handset.
3. The code step: some networks then send the customer a one-time code or voucher. The AI asks for it and passes it on with Give the payment code (at most 3 tries; the code is masked in logs).
4. Confirmed before goodbye: when the payment arrives during a live call, the AI is told at once; it checks with Check a payment and confirms the amount and reference before the call ends.

Payment settings

• 1 Let the AI and agents take payments: switches on the payment skills and the desk's Take payment button.
• 2 Payment provider: Sandbox (test, no real money), Paystack or Hubtel. The currency is the workspace's (Settings → Workspace → Currency).
• 3 Verify the customer's identity before taking a payment: the identity check (module 9) runs first.
• 4 Sandbox only: prompts approve themselves after these seconds, or wait for the approve button.
• 5 Sandbox only: practise the code step; the code is always 123456.

1. Open SettingsPayments (/settings/payments). Admin
2. Start with the sandbox: Payment provider Sandbox (test, no real money), tick Let the AI and agents take payments, set Smallest amount and Largest amount, keep Verify the customer's identity before taking a payment ticked → Save.
3. Practise the code step: tick Sandbox: simulate a network that sends the customer a code (it is 123456) → Save, and make a test call or desk payment.
4. Going live: choose your provider. A card asks for its key (Paystack secret key or Hubtel client secret, plus the Hubtel merchant account and client id): type it there → Save key. It shows Set and is "Stored encrypted and never shown again." Register the callback URL shown under it in your provider's account.
5. Save, then make one small real payment yourself before customers do.

On the desk: Take payment and Submit code

1. Click Take payment on the open conversation. Fill Amount, What it is for, Mobile money number, Network ("Detected from the number when empty.") → Send prompt. AgentSupervisor
2. The customer approves with their PIN while you talk.
3. If you see "The network sent the customer a code — ask them for it", the dialog Enter the customer's code opens: ask them to read it, type it in Code → Submit code.
4. "Payment PAY-… received" confirms it. Tell the customer the reference before you wrap up.

The Payments page

1. Open Payments (Customers section): Collected (30 days), Collected by the AI, Waiting for approval, Approved, and the list (Reference, Customer, Amount, Status, For, When). Filter by Status. AdminSupervisorAnalyst (agents see their own)
2. Check a pending one with the refresh icon on its row.
3. Sandbox rows are marked sandbox: supervisors and admins approve or decline a waiting one with the tick or cross icon.

When Kasada AI creates your workspace, it starts from one of 14 industry templates (healthcare, banking, insurance, telecom, utilities, ecommerce, logistics, hospitality, education, government, real estate, automotive, politics, or generic). Your industry shows in SettingsWorkspace under the card title. Everything the template made is yours to change.

1. Walk through what you got: open each place in the right-hand column above once. Admin
2. Fix the starter knowledge first: open each template article in Knowledge. Replace every text in [square brackets] (prices, times, codes) with your real facts, or Archive it → Sync to AI.
3. Read the compliance rules and instructions and change anything that is not true for your company.
4. Adjust queues, skills, wrap-up codes, case categories and SLA hours to how your team really works. Accra Mobile Care added its own SIM swap queue (module 7) next to the template's four.
5. Go live with the agent only after this: Status Live, Save, Sync to CustoCare (module 8).

1. Open Billing (/billing): Account, Status, Credit balance, Overdue, your plan with usage meters, This period's estimate, Top-up packages, Invoices, Payments. Admin
2. Pay or top up with Open billing portal or Top up; quick links Invoices, Usage, Payments open the same portal signed in.

Payment reminders

When a CustoCare invoice is past its due date, CustoCare Billing sends friendly reminders by SMS and WhatsApp, on top of its emails: usually a friendly SMS on day 1, a firmer WhatsApp on day 4 and a final SMS on day 8, only between 08:00 and 19:00 and once per step. The final one says what pauses next.

1. Watch for the reminder on the account's phone. Some are sent from your own workspace's number, so they look like your own messages. Admin
2. See which were sent: SettingsAudit log, type reminder in Filter action: "Billing payment reminder" (and "Billing final reminder") with the invoice number.
3. Pay from BillingOpen billing portal before the account moves to Suspended.

1. Note the exact words on screen and the page address. Any role
2. Find the problem below and do the check.
3. Still stuck? Tell your admin what you saw, when, and the interaction or case reference (for example CS-000214).

Glossary

AI agent

The virtual agent that answers calls and chats on CustoCare (e.g. Adwoa).

AI containment

Share of closed conversations the AI resolved without a person.

Barge in

A supervisor joins a live call so everyone hears them.

Built-in skill

A tool every workspace has (identity check, case status, follow-up, payments…), marked Built-in under Integrations → Tools.

Connector

A connection to one of the company's systems (REST, GraphQL, database, MCP server, webhook).

Customer 360

The desk panel with the contact, identity check, open cases, history and tool runs.

Code step

A one-time code some mobile money networks send the customer to finish a payment; never a PIN.

Disposition

The outcome an agent picks at wrap-up.

Emotion level

Calm, Annoyed, Frustrated or Angry, worked out live from what the customer says.

Guardrails (policies)

Rules the gateway enforces on every tool run: confirmation, verified caller, channels, limits, redaction.

Handover

The AI transferring a conversation to a queue for a person.

Kasada Dialpad

The browser softphone in the top bar.

MCP server

Your workspace's tools offered to MCP clients with a service key.

Overflow

What a queue does when closed, unstaffed or waiting too long (callback case or back to the AI).

Presence

Available, Busy, On break or Offline.

Queue

Waiting work grouped by the skills needed.

Sandbox

The test payment provider: behaves like a network, moves no money.

SLA

Answer time for a queue; resolve time for a case.

Sync

Sending AI agent or knowledge changes to CustoCare.

Web chat widget

The chat bubble on your website; its paste code is in Settings → CustoCare.

Whisper

Guidance a supervisor sends the AI during a call; the caller hears nothing.

Wrap-up

Finishing an interaction: disposition, notes, CSAT and an optional case.

Go-live checklist (admin)

• Settings: Workspace, Business hours and holidays, Channels, Compliance rules (including "answer only from our information").
• Knowledge: current prices, hours and policies published; template [placeholders] replaced; old ones archived; Sync to AI done.
• Settings → Languages: local languages, Always say these in English, local words; Preview checked.
• AI agent: greeting, instructions, languages, voice, default queue, escalation rules, phone numbers; Live, Saved and Synced.
• Queues with skills, SLA and overflow; Routing rules checked.
• Staff invited with the right roles, skills and Chats at a time; extensions assigned.
• Settings → Softphone: Softphone on for our agents; a test call rings an agent's Dialpad.
• Settings → Email set up; test email received.
• Integrations: tools tested in the Test console, enabled, and allowed on the AI agent.
• Web chat: snippet from Settings → CustoCare pasted on the website; a test chat answered.
• Payments (if used): tested in the sandbox, including the code step; one small real payment made.
• AI Lab: Test lab run passes (70 or more).
• A test call and a test chat: AI answers, hands over, an agent accepts and wraps up, a case is created.
• Billing page shows the account as active.