- Spotnana has deployed a multi-agent AI architecture to automate five routine travel servicing workflows.
- The architecture uses specialised agents for narrow tasks with a central orchestrator, limiting AI execution authority.
- The first release covers tasks like processing cancelled segments, refunds, and importing externally booked trips.
- Human agents remain in the loop for complex cases with co-pilot support for suggested responses.
- The platform’s ‘open by design’ approach allows customers and partners to build their own agents.
Spotnana has deployed a multi-agent AI architecture that automates five routine servicing workflows, allowing travel agents to focus on more complex cases. The travel-as-a-service provider added this architecture to its platform alongside a first release of AI capabilities for travel agents, with features for travellers and travel managers expected later this year. The system splits the work across specialised agents, each responsible for a narrow servicing task, with a central orchestration agent deciding which specialist should handle the request.
The architecture does not use one general-purpose assistant to answer every query but allows the language model no authority over execution. The agent can interpret the task and choose the next step, but the actual action runs through a mapped API and a governed workflow. The first release covers several servicing jobs, including processing cancelled segments, validating unticketed airline segments, issuing residual MCOs after eligible ticket exchanges, automating eligible refunds, and importing externally booked trips. Human agents remain in the loop for complex cases, supported by a co-pilot with suggested responses and conversation summaries.
Direct Travel is already using the new capabilities, with its Chief Product Officer framing the split as automating routine servicing so advisers can concentrate on complex situations and strategic guidance. The platform is “open by design”, allowing customers and partners to build their own agents. The proof will be in how much volume disappears from human queues over the coming months and whether the exception rate stays low enough during irregular operations.
The analysis shared that the architectural argument about determinism applied to a different problem, and the question is how to allow an autonomous system to act on a live booking without letting it invent the outcome.
