Back to Release Notes
2026.11.02November 2, 2026
A sharper, more dependable virtual assistant
This is a focused release centred on the conversational assistant that customers use to look up their cover, ask questions and book a service. The assistant now runs on a newer, faster language model, and the rare case where it returned an empty reply is now handled so a conversation is never left hanging. Everything else in the platform continues to behave exactly as before.
Products updated: Customer Web · Platform API
Virtual assistant — faster, more capable conversations (Customer Web, Platform API)
ImprovedThe assistant that handles customer conversations was moved onto a newer generation of language model.
- Newer language model behind the assistant. Customer conversations are now handled by a newer, faster model. Answers arrive more quickly and follow the thread of a conversation more closely, including when a customer changes their mind mid-way or asks several things at once.
- Steadier answers on familiar journeys. Identifying a customer, listing the services available to them, booking one of those services and adding a note to the request all read more naturally than before. The assistant keeps to the same wording and the same steps each time it is asked.
Assistant reliability — no conversation left unanswered (Customer Web, Platform API)
Improved- Empty replies no longer stall a conversation. In the uncommon case where the assistant produced nothing to say, it retried indefinitely and the customer was left waiting. That retry is now bounded, so the conversation either recovers straight away or moves on with a clear message instead of going quiet.
- A clean route to a person. When the assistant cannot help, or is unavailable, the conversation hands over to customer service with the right contact details rather than ending abruptly. Customers who cannot be identified are pointed to the correct form for their request.
Conversation quality — reviewed release by release (Customer Web)
Improved- Each release is checked against real conversation scenarios. Assistant conversations are now reviewed end to end against a standing set of everyday situations, covering customer identification, browsing available services, declining what is offered, booking, and refund questions. The result is that answer quality stays consistent from one release to the next rather than drifting as the assistant changes.
Ready to see Sodtrack in action?
Book a demo and our team will walk you through the latest features tailored to your operation.