Sunday Natural sells supplements and natural-health products from Berlin. Its customers do not write in about delivery windows nearly as often as they write in about the product itself: what is in it, how it is made, and whether they should be taking it at all.
The Questions a Supplements Retailer Cannot Answer
Roughly one contact in seven is a product question, and they go deep — molecule sizes, raw material sourcing, formulation specifics. Answering those well requires the catalogue, not a script.
A meaningful share go further than that, into pregnancy, breastfeeding, and blood test results. Those are not hard questions to answer. They are questions a retailer is not permitted to answer: under German rules Sunday Natural cannot give medical advice, and the correct response is a redirection to a doctor.
So the constraint is not capability. It is knowing precisely which questions must never be answered automatically, however confidently the system could phrase a reply.
Confident on Products, Deliberately Silent on Health
Open runs on email and web chat in front of Zendesk, where Sunday Natural’s agents continue to work their tickets. Nothing about their workflow moved.
On orders, the agent acts rather than advises: an order cancellation is executed directly against Sunday Natural’s Magento backend, authenticated and completed inside the conversation.
Health questions take the opposite path by design. Anything touching pregnancy, breastfeeding, or medical results is routed to a person, with no automated attempt at an answer. Account-access requests follow their own keyword-triggered handoff.
Product, feedback, and general enquiries sit in between: the agent drafts, and the reply is held so a human can intervene before it goes out. Refunds and vouchers stay manual until the underlying APIs exist.
The Results
Measured across the channels integrated at the time of the snapshot:
| Metric | Impact |
|---|---|
| Automation rate | 44% |
| Tickets AI-resolved (90 days) | 30,309 of 98,772 |
| Channels covered | Email and web chat |
| Handoff destination | Sunday agents in Zendesk |
The 31% is scoped: it covers the channels integrated at the time — principally email — against 98,772 tickets, with 30,309 resolved by AI over a trailing ninety days. Automation on chat runs lower than email, and the ceiling is a known one: refunds, vouchers, and newsletter changes are all waiting on APIs that do not exist yet.
The Impact
31% automation across email and chat lets a support team of seven carry a catalogue this technical without scaling headcount proportionally with order volume. Zendesk stayed; nothing was migrated.
What the team keeps is the work that genuinely needs them — the health question that has to reach a human, the partial refund that has to be calculated by hand, and the product query deep enough that a customer deserves an expert rather than a fast answer.
