Platform
AI customer care with catalogue-backed order capture, appointment requests, phone calls, social DMs, website chat photos and owner-approved lessons. See reporting and practical limits.
What the AI actually does
- Instant, on-brand replies. Trained on your services, policies and voice, so it sounds like you, not a bot.
- Speaks their language. It replies in whatever language the customer writes in, and handles code-switched writing where two languages are mixed in one sentence, such as Spanglish, Hinglish, Franglais or Taglish.
- Collects appointment requests. In chat or on a live call, it collects the customer's details, service and preferred time, then records a request for your team to check and confirm.
- Reads photos and voice notes. Website and shareable chats accept customer photos. Voice notes on supported messaging channels are transcribed into the same conversation.
- Answers the phone. Give your AI a phone number and it holds a real conversation, interruptions and all, takes voicemail out of hours, and writes every call down as plain text.
- Sounds like your team. Give it a name, a job title and a manner, then train it in plain English. Separate instructions for calls coming in and calls going out.
- Captures and follows up leads. Every contact becomes a lead. The AI follows the playbook you wrote in plain English to re-engage them.
- Answers social DMs. Replies to customers in Instagram Direct and Facebook Messenger, then brings their conversations into your dashboard.
- Takes the order and files it. It confirms the order in the conversation and records it, with every line price and the total recalculated on our servers from your own catalogue.
- Knows when to stop. If it cannot help, or somebody asks for a person, it escalates with the whole conversation attached instead of guessing.
- Self-improving: it gets better every week. When your team corrects the AI on an escalated chat, that correction becomes a proposed lesson. You approve it or you do not, and only approved lessons ever reach a customer. The rule is enforced by the database, not by a prompt, so nothing a customer types can change how your assistant behaves for anyone else. The dashboard shows you the questions it could not answer, and your escalation rate before and after.
- A CRM that fills itself in. Contacts, cases, opportunities and the promises it made are written as it works, not typed up afterwards. Someone who messages on WhatsApp and phones two months later is one person with one history. A problem escalated on one channel and chased on another stays one case instead of two nobody owns.
- Fewer tickets, not faster ones. Most requests are finished in the conversation before a ticket exists. The rest escalate with the whole thread, the reason and the customer's history attached, and can open the matching ticket in your helpdesk at the same time.
- Guides refund requests. It explains your refund policy and can file a request for review with the order and reason attached. Payment refunds still require the appropriate approval and payment-system action.
- Analytics you can inspect. Conversation volume, active chats, current handoffs, orders, channels, topics and open questions, with a CSV export.
- Enterprise security. Row level data isolation per business enforced in the database, three tier role based access, API keys stored as hashes, OAuth 2.1 with PKCE for MCP, and encryption in transit. Your conversations are never training data for anyone else.
Reporting and quality review
See the work, then improve the answers
The analytics dashboard breaks conversations down by day, channel, topic and local hour. It shows active conversations, current handoffs, recorded orders and questions the assistant could not answer. Choose a date range, inspect the underlying conversations and export the chart data as CSV.
Illustrative example: if a selected period has 120 conversations and 18 are currently in the handoff queue, 18 divided by 120 is 15% currently escalated. This is a queue snapshot, not a resolution or satisfaction rate. A conversation resolved by a person is no longer in that queue.
Review real transcripts and open knowledge gaps, then approve or reject AI-proposed lessons before they affect future conversations. Recorded order value does not prove payment revenue. CSAT surveys, measured response-time reports and automatic answer-quality grading are not currently provided.
- Volume and demand. Daily conversations, channels, topics and busiest local hours.
- Work waiting for people. Active conversations and the handoff queue, with the original thread available for review.
- Knowledge gaps. Open questions grouped by how often customers asked them.
- Order activity. Recorded order count and value for the selected period.
- Export. A CSV of daily conversations, current escalations and order value.
Where it works
One assistant supports connected customer channels and sends requests that need a person to your team. The directory lists 96 platforms across messaging, stores, CRM and helpdesk, payments and couriers, of which 32 are available today and the rest say coming soon rather than failing the first time you need them. Nothing that works is held back for a higher plan. Full directory, with the status of every entry, at https://octadezx.com/integrations.
Run it from your own AI assistant
OctaDezx ships a native Model Context Protocol server at https://octadezx.com/mcp. Any MCP capable client connects to it, so you can run customer care from whichever assistant you already use rather than learning another dashboard.
Explore the work behind each conversation
Use these focused guides to check what the assistant reads, which action it can take and where a teammate still owns the outcome. A restaurant answering menu questions, an agency qualifying enquiries, a clinic collecting appointment requests and a retailer explaining products each need their own knowledge and handoff rules.
For larger teams, plan a staged rollout around business data isolation, teammate permissions and controlled developer access. Review API and SDK scopes, assign escalation owners and validate the workflows your operation depends on before expanding. A catalogue answer, a recorded order and a paid purchase are different events; test each transition.
Knowledge and discovery
Catalogue-backed answers and supported website photo analysis help shoppers find a relevant listed product.
Capture and follow-through
Confirmed orders and appointment requests give your team a record to work from. Checkout, fulfilment and appointment availability still depend on your business setup.
Read more on this
- What AI customer care actually automates, and what it does not
Most claims about AI support collapse the moment a real customer asks something specific. Here is an honest split of the work an AI agent takes off your plate, and the work it should never touch. - How to train an AI on your product catalogue
A practical guide to importing product data, writing clear policies, reviewing wrong answers and fixing the five failure modes of catalogue-trained support. - When should an AI hand a customer to a human?
Escalation is the most underrated part of AI customer care. Get it wrong and you either annoy customers or bury your team. Here are the triggers worth setting from day one.