AI customer support agent
Answers from your tables and policy docs, actions through appflows, humans when the case is not routine — customers never need a Maicos login.
Answers from your tables and policy docs, actions through appflows, humans when the case is not routine — customers never need a Maicos login.
A support bot that cannot query the order table is a delayed FAQ page with extra latency. This agent is only worth running if it is grounded.
Customer-facing mode, approved tables and docs, appflows for refunds or replacements, escalation with the transcript attached. Internal staff can watch volume from the same workspace.
Tables
Grounded lookup
Orders and policies you named
Flows
Actions with brakes
Refunds pause when you say so
Human
Handoff with context
No “please repeat your order number”
Traditional tools
Intercom bot + helpdesk + a spreadsheet of orders the bot cannot query.
With Maicos
One agent, the same tables your team edits, appflows for mutations, humans on exceptions.
Look up, act, escalate. In that order.
Order status and policy quotes come from records you point at. If the row is missing, the honest behavior is “I don’t see that order” plus escalation — not a invented tracking number.
Returns, replacements, password resets — as flows with optional manager approval. Chat should not UPDATE a table because the model felt confident.
The human gets the customer, the transcript, and the rows already fetched. That is the difference between escalation and starting over.
Web or WhatsApp can hit the same agent and the same tables. Do not run a different bot per channel with a different memory.
Source of truth, then actions, then the exception path. Personality last.
Orders, shipments, a policy doc. If it is not in that list, the agent should not “recall” it from training data.
Refunds, address changes, cancellations. Add amount or SKU brakes that force a human.
Unknown order, abuse, legal, medical, or “customer asked for a human.” Measure how often it fires; do not bury it.
Guests stay outside the workspace. Your team watches tickets from inside.
The only acceptable answer is from the shipments table.
Customer asks on the public channel
No Maicos account. Customer-facing agent.
Agent looks up 4821
Row says shipped yesterday, carrier + tracking in the table.
Optional appflow
Resend tracking email if that flow exists. No row? Say so and escalate.
If they want a return over the auto-limit
Start the refund flow, pause for a human, pass the transcript.
Customers forgive “I don’t see that order, I’m getting a person.” They do not forgive a fake tracking number.
Good: order status, policy, how to return. Bad: unrestricted legal or medical advice. Put those on the escalation list.
Support Agent
Live · customer-facing
The same apps you already work in — not a second chatbot that needs uploads.
Useful agents have a job description. These limits keep this page honest — and help you pick a different agent when you need one.
How people use this every day.
Cite the policy doc. If the doc is silent, do not improvise a discount — escalate or take a message.
Shipments table, then tracking via appflow. Same agent on web and WhatsApp.
Under $X automatic flow; over $X human. The transcript goes with the ticket.
If the table has prior tickets, the agent should not ask for the order number you already collected last time — unless you failed to store it.
A customer-facing Maicos agent that answers from your tables and docs, starts appflows for actions, and escalates to staff with context. Customers do not join the workspace.
It can start a refund appflow you designed. Put human approval on amounts or case types you do not want fully automatic.
No. That is the point of customer-facing mode.
Say you cannot find it and escalate. Do not guess. Fix the data pipeline if this happens often.
Support is for customers. Ops is for employees (IT, facilities). Different audience, often different tables, never the same permissions by accident.
Start free — Workspace, Create, Automate, and Integrations built for AI from day one.
No credit card required · Free during beta