AI e-commerce customer service
AI that knows what it knows, and hands over the rest.
The AI reads every message, finds the order and the right policy, and drafts a reply with its sources. It answers alone only what you allow, and never moves money.
Three modes
You decide how much the AI does.
Switch modes at any time from your portal.
- 01
Off
Every conversation goes to an agent. The AI still sorts the queue by topic, language and urgency.
- 02
Copilot
The AI drafts every reply with its sources; an agent reads, edits and sends. Recommended to start.
- 03
Autopilot
The AI answers alone the routine topics you allow, when every safety check passes. Everything else is drafted for an agent.
Before anything is sent
Four checks for an automatic reply.
The topic is on your allowed list. The answer is grounded in your order data and policies. Confidence is above your threshold. The customer is not upset. If one fails, a person answers.
Sent by your brand · approved by an agent
AI Active Support
- Customer identifiedEmail matched to the customer record
- Order found#3842 · Shopify · paid, shipped
- Tracking retrievedIn transit · delivery expected Thursday
- Policy checkedShipping policy · 3 to 5 business days
AI recommendation
Reply with the delivery date and tracking link. Sources: order #3842, shipping policy.
Verified order data Verified policy
- Damaged item 12 minPerson
- Missing item 41 minDraft ready for the agent
- Refund request 1 h 05Approval
Copilot draft · for the agent
The carrier marked it delivered yesterday at 14:10. Could you check with neighbours or your building’s mailroom? If it has not turned up by tomorrow, we will open an investigation with the carrier.
- Verified order data · delivered, carrier scan
- Verified knowledge · delivery FAQ
- Customer-provided · “I have nothing”
Illustration of the agent inbox with sample data.
Sources
Every statement says where it comes from.
- Verified order data and knowledgeRead from your store and your published knowledge base.
- Customer-providedWhat the customer said, not yet confirmed.
- AI inferenceWhat the AI concluded on its own: the part to double-check.
Guardrails
What the AI never does.
- Move money or change orders
- No refunds, cancellations, order edits or discounts. The AI has no access to do them.
- Answer sensitive topics alone
- Refund, return, exchange, damaged-item, payment and complaint conversations always go to a person.
- Invent facts
- Tracking numbers, links, amounts and delays that are not in your sources are caught before sending.
- Decide permissions
- Approvals and access rules are enforced by the platform, never by the AI. It can recommend, not authorize.
- Share internal procedures
- Internal procedures help agents but are never used in replies to customers.
- Spend without limit
- Costs are tracked per call; a monthly budget pauses the AI when it is reached.
The AI recognizes 20 topics; 12 of them are never answered automatically.
Pricing
AI Support: €49 / month
AI Starter · 50 AI resolutions included, then €0.89 each · unlimited team seats. A conversation reopened within 72 hours is billed once.
Is our data used to train AI models?
Requests are only routed to AI providers that do not collect prompt data, and personal details are masked in the text sent to the model.
Which languages does the AI work in?
It is tested in Français, English, Español, العربية, Deutsch, Italiano, Nederlands. Agents’ languages are agreed with you.
What if a customer tries to trick the AI?
Messages and knowledge articles that try to give the AI instructions are detected; they are not used as instructions and the conversation goes to a person.
Can we cap AI costs?
Yes. Set a monthly AI budget; at the limit the AI pauses and every conversation goes to your team.
Book a free demo
Let’s look at your real tickets.
In 30 minutes we go through a sample of your conversations, show which ones the AI could answer and which need a person, and give you a price.
