What Nortinia AI Chat actually does for the visitor

Classic web chat was a clickable FAQ. Nortinia AI Chat is a different category. It does not answer the question — it acts on the visitor's behalf. Concretes are easier than abstractions, so here is the real traffic data.

What an average session looks like

We looked at 30 days of traffic across the full Nortinia and Netorigo storefront fleet. An average session triggers 4 actions, and 70% of those fall into these four:

1. Answers a product question — "Is there a warranty on X?", "What is delivery time on the weekend?", "What is the difference between Pro and Enterprise?". The answer is not static FAQ — it comes from a live catalog lookup.
2. Adds to cart — "Put both of them in M size into the cart." The chat invokes the cart.addItem MCP tool; the visitor only confirms.
3. Applies a coupon — "Is there any promo code I can use?" If the tenant has an active coupon, the chat surfaces and applies it.
4. Navigates to the right PDP or page — "Where do I find references?", "Take me to the B2B packages page." SITE_NAV_navigate MCP call, panel stays open, page changes.

On top of those, 22% of sessions book a demo, and 9% initiate a post-purchase return ("I want to send something back, what do I do?").

The boring stat that surprised a lot of people

In 38% of sessions, the visitor uses the chat purely for navigation. They ask one question, land where they wanted to land, leave. They buy nothing and they will not buy in the next hour either. But they are happy: they found what they needed in 3 seconds instead of 30 seconds of clicking around.

This matters because it is the most commonly misread metric. A chat system's KPI is not how many sessions end in a purchase. The chat is a search + assistant + sales hybrid. If you only measure purchases, you miss 60% of the value.

What it does not do

It is important to be explicit about what we do not do. The chat does not replace human support in complex cases. If the request crosses a risk threshold (refund > €500, legal question, formal complaint), the system hands off to a human — separate article on that.

The chat does not write marketing messages at the user. It is reactive, not proactive. It only speaks when the visitor asks.

The chat does not step outside the tenant's catalog + knowledge base. If the visitor asks something out of scope ("give me personal financial advice"), the chat politely declines.

What this means for the owner

Two concrete business effects are measurable:

• Assisted purchases — orders where there was at least one chat interaction in the 24 hours before checkout. On the fleet, that is 14-21% of total conversion depending on the tenant.
• Deflected support tickets — questions a human did not have to answer. A conservative estimate is 30-50 tickets / tenant / day.

The per-session value is not where marketing copy points. The reality is more complex and cleaner to measure.