How AI helps — concrete outcomes

This page is about outcomes, not features. Here's what the AI in NShop actually changes for the people using it.

Instead of clicking through filters, a shopper types "something warm for hiking trips in October, around €100" and gets three good matches with images, prices, and the right variant pre-selected. Semantic search catches intent that prefix search would miss.

Get real answers about real orders

"Where's my last order?" goes through the authenticated get_orders tool, pulls the customer's actual order, and reads back the status, carrier, and ETA — no support ticket, no waiting.

Buy without leaving the conversation

The shopping agent renders an inline product card with an Add-to-cart button. The shopper acts in one tap, then keeps chatting.

Be remembered across visits

"I usually buy Medium" — saved as a preference. Next time the agent filters for Medium automatically. No login walls, no preference forms.

Get help in their language

Both assistants follow the locale prefix in the URL. A Dutch shopper talks to a Dutch assistant; an English shopper talks to an English one. The persona stays the same — only the language changes.

For admins

See growth opportunities they'd miss otherwise

The Growth Insights panel reads 30 days of data and surfaces things that'd take a spreadsheet to find. "Ten of your top sellers are out of stock" is a one-sentence sales lift; you didn't have to query for it.

Get product photography on the same day

Instead of waiting on a photo shoot or stock-photo licensing, the catalog team writes a prompt and approves a result. Time-to-list drops from days to minutes.

Get faster support resolution

When a shopper asks a question, the storefront chat answers without opening a ticket. The admin team handles fewer routine questions and focuses on the ones that actually need human judgment.

Tune the brand voice once

Set tone, industry, and custom rules once. Every AI surface adopts the voice — no per-channel configuration, no copy-paste prompts.

For the platform

Tenant isolation that survives an LLM

The agent can call any tool, but every tool re-checks the principal and the tenant. Even if the model hallucinated a tool call for another tenant, the tool would refuse. The model can't leak data it doesn't have access to.

EU data residency

Every model call goes to europe-west4. Prompts, responses, and embeddings stay in the EU. For GDPR-conscious shops, that's a prerequisite, not a nice-to-have.

Costs that scale with usage, not with seats

Vertex AI is pay-per-call. There's no per-seat AI fee — small shops pay for what they use; big shops scale linearly.

What's on the roadmap

Agent-side checkout assist (walk the shopper through Stripe step by step).
Agent-side return / refund initiation for the customer's own orders.
Admin-facing agent that can edit catalog and run promotions on command — fully authenticated, fully audited.