Case Study

Seven weeks to agentic checkout on the Swiss rail network

How Swisper and SBB built, tested, and shipped the Mobility Agent

07.09.2026 · 5 min read

Swisper and Swiss Federal Railways (SBB) built an agent that plans a multi-leg Swiss rail journey and buys the ticket inside the chat. The contract was signed at the end of March 2026, the proof of concept was accepted seven weeks later, and the agent went live in production in August.

7 weeks

From contract to accepted proof of concept

2 people

Build team on the Swisper multi-agent orchestration platform

6 min

One prompt for a multi-day trip that took 29 minutes by hand

AT A GLANCE

Partner
SBB official sales partner
Scope
Multi-leg Swiss rail planning and ticket purchase in the chat, with Halbtax (the Swiss half-fare travelcard) and regional travelcards applied
Testing
70 SBB employees from mid-May to mid-July, then selected end users from the SBB community
Live
August 2026

The starting point

Planning and booking a multi-day business trip in Switzerland is manual work. The traveler copies each destination out of the calendar into the booking website, buys a separate ticket for every day, and compares the price offers day by day to land on the right combination of fares and travelcard discounts. In our own measurement, a trip with several destinations across several days took 29 minutes of continuous attention.

Everything needed to do this well already exists. The information sits in the traveler's calendar, in the timetable, and in the fare logic. What was missing was something able to read all three at once and then act on them.

What we built

The Swisper Mobility Agent takes a single instruction in natural language, for example a request to book the travel for next week's appointments. It reads the destinations and times, plans the connections across every leg, applies the traveler's Halbtax and regional travelcards, compares the available fares, presents the itinerary for review, and completes the purchase and payment once the traveler confirms.

That final step has proved hard across the industry. In March 2026 OpenAI stepped back from checkout inside ChatGPT and moved its focus to product discovery, about six months after launch, because the first version did not offer the flexibility OpenAI wanted to give merchants. Reports at the time pointed to single-item checkout with no basket, no loyalty or discount handling, and no real-time view of inventory.

The Swisper Mobility Agent books a multi-leg, multi-day journey as one transaction, applies the traveler's Halbtax and regional travelcards, and works directly against SBB's own booking and fare services. The result is a purchased SBB ticket. It is built in such a way that pricing, terms, discounts, and bundles can be configured per provider. Agentic checkout runs in Production since August 2026. It could even work across providers, with Swisper taking over dispatching of orders and payments in the background, a complex and promising project for the future.

Seven weeks from signature to accepted proof of concept

The contract was signed at the end of March 2026. The proof of concept was accepted in the middle of May. Two people built it.

The Swisper multi-agent orchestration platform already handles planning, fare comparison, purchase, and payment, so the two-person team spent its seven weeks on integration and domain logic.

Date
Milestone
January 2026
First conversations between Swisper and SBB
End of March 2026
Contract signed
April to mid-May 2026
Development by a team of two
Mid-May 2026
Proof of concept accepted, seven weeks after signature
Mid-May to mid-July 2026
Testing with 70 SBB employees
August 2026
Opened to selected end users from the SBB community
August 2026
Release into production

How the collaboration was set up

The project succeeded because each partner brought its strengths to the table. It also laid the foundation for a long-term strategic partnership.

Swisper

The agentic framework and the AI engineering, covering orchestration across planning, fare comparison, purchase, and payment, the conversational interface, and the production hardening.

SBB

The domain expertise in public transport, the booking and fare interfaces, and access to a large internal test group of everyday rail travelers.

That test group changed the product. Ticket requirements for dogs above a certain height, and train occupancy as a criterion the traveler can state in the prompt, both came directly out of the testing phase.

An external team does not find these requirements on its own. They come from how people travel day to day.

Results

29 min

Manual booking

6 min

With Swisper, from one prompt

Manual booking means copying destination information out of the calendar into the booking website, buying a separate ticket for each day, and comparing the price offers for each day by hand.

With Swisper, 90 percent of the six minutes is idle time while the agent works, which the traveler spends away from the screen. The remaining 10 percent goes to reviewing the travel suggestions and confirming booking and payment.

The comparison is a like-for-like test of the same multi-day itinerary. The time saved matters, and so does where the traveler's attention goes, from assembling the booking to approving it.

What comes next

Swisper and SBB are deepening the collaboration in three directions.

SwissPass integration

The traveler's existing SBB products and subscriptions are recognized automatically. This integration is in development and not yet live.

Accommodation

Next, we are integrating hotel bookings into the Mobility Agent.

More mobility providers

A seamless travel experience for Switzerland in which a journey is planned and paid across providers. The agent is the layer that makes that possible without every provider building its own.

About Swisper

Swisper is a Swiss AI platform that builds and runs production agents for regulated industries and large enterprises. Data stays in Switzerland. The Mobility Agent with SBB has been live in production since August 2026.

Talk to us about what an agent could do in your own domain.