Automation

Autonomy on your terms.

Suvenna gives you a dial, not a switch. Choose whether the AI drafts or sends, set the refund amount it can approve on its own, and keep live money movement off until it earns your trust. Every setting is per store, and every action is logged.

Suvenna automation settings: Autopilot toggle, refund auto-approval ceiling, automation rules, and live-action controls

Two modes

Drafts first, sends when earned

Every store starts with a choice: does the AI send, or does it draft? You can change the answer per store, at any time.

Approve-mode

The AI reads the conversation, looks up the real order, and writes a complete reply. Then it stops. The draft waits for your team to approve, edit, or reject it. Nothing reaches a customer without a human's click. This is where every rollout should start.

Autopilot

The AI sends grounded replies on its own: order status, policy answers, routine resolutions. It still abstains and escalates whenever your policies don't cover the question, and money actions still follow your ceiling and confirmation rules. Autopilot changes who clicks send, not what the AI is allowed to do.

A dial, not a switch

The rollout we recommend

Trust is built in steps, and the settings are designed to follow them. Here is the journey most stores should take.

Week 1
Start in approve-mode
Every AI reply lands as a draft. Your team approves, edits, or rejects each one. The AI does the writing; humans keep the send button.
Then
Review the QA scores
Grade replies with per-reply QA as you approve them. The reports dashboard shows exactly where the AI is strong and where it still escalates.
When ready
Enable Autopilot
Flip AI sends on for the conversation types it has earned. Anything it isn't sure about still escalates to your team with full context.
Next
Set your refund ceiling
Choose the amount the AI can auto-approve under. Anything above it queues for one-click human approval. Start low; raise it as trust grows.
Finally
Enable live actions
Per-store live-action toggles ship off. Turn on refunds, cancellations, and returns when you're ready. Every execution stays confirm-gated and audit-logged.

Turn the dial back down whenever you want. Every step is reversible, per store, in one setting.

The refund ceiling

One number decides what the AI can approve alone.

Set a ceiling, say $20. Refunds under it can be auto-approved once the customer explicitly confirms. Anything above it queues for one-click human approval. And at execution time, every refund is bounded by the order's actual refundable amount, so the AI cannot refund more than the order allows, whatever the conversation said.

The customer explicitly confirms before any money moves
Refunds above the ceiling queue for human approval
Amounts are bounded by the order's real refundable balance at execution
Every action lands in the append-only audit log

Automation rules

Route, tag, and escalate on your conditions

Rules run on every incoming conversation, before the AI ever replies. Use them to decide who, or what, handles each customer.

Channel

Route chat-widget conversations differently from email. Tag by source so your views stay clean.

Sentiment

Send frustrated customers straight to a human. The AI handles the routine; your team handles the heat.

Business hours

Let Autopilot cover nights and weekends, or hold drafts for the morning shift. Your call.

VIP lifetime spend

Flag your highest-value customers by lifetime spend and route them to a person, every time.

Order age

Treat a day-old order differently from a ninety-day-old one. Return windows become routing logic.

Escalation

Any rule can route, tag, or escalate. Combine conditions to match how your team actually works.

The support layer

Everything the automation feeds into

Automation is only useful if the human side keeps up. These are the pieces that keep the desk moving.

Approval queue

Every gated action waits here with the order, the amount, and the full customer context. One click executes it against Shopify. Agent presence detection means two people can't double-approve the same refund.

Least-busy routing

New conversations are assigned to whoever has the lightest load among available agents. No cherry-picking, no pileups, and away agents don't get assigned work they can't see.

Spam suppression

Spam filtering keeps junk out of the queue entirely, and a sender blocklist ends repeat offenders for good. Your SLA timers only run on conversations that deserve them.

Get priority access

Suvenna onboards in limited waves. Join the list and you're first in line for the next one, with founding-member pricing locked in.

Request early access See pricing