
OpenDialog
An auditable AI agent platform for insurance

Tool details
OpenDialog is an enterprise-grade AI agent built by a London team that targets regulated industries only — in practice, for now, almost exclusively UK insurance. It does not sell a sign-up-and-go customer service chatbot; it sells an auditable conversational system that plugs into quote, policy, mid-term adjustment and claims workflows.
Core capabilities
- Five ready-made agents: Selma for quote conversion, Jamie for policy Q&A, Rhea for winning back abandoned quotes, Michaela for MTA changes, and Clark still to come
- Deterministic orchestration plus point-in-time LLM calls: the LLM is invoked through LLM Actions only after intent recognition and before a message is sent — not an end-to-end autonomous loop
- Swappable models: OpenAI, Azure OpenAI, Gemini (Vertex) or Claude, or managed key custody
- Semantic classification plus RAG: the LLM judges intent with no training phrases required, paired with proprietary vector retrieval since 2024-09
- SAFER five-dimension governance: security, query-scope detection, retrieval fidelity, compliance evaluation and recognition of its own knowledge limits — auditable both before and after launch
- conversation designer: a no-code canvas for building your own agents, with a Webchat SDK and Webhook actions for connecting to your own systems
Where it fits UK insurers, MGAs, brokers and insurance software vendors: lift quote conversion, keep policy Q&A in front of human teams, win back abandoned quotes, and automate mid-term adjustments and FNOL. Not for buyers who want an out-of-the-box customer service bot billed per token.
What sets it apart Sierra and Decagon sell customer service agents for any brand; OpenDialog sells the specific breakpoints in the UK insurance journey, and also acts as the white-label engine for software vendors (ICE Acturis, Open GI, Durell). SAFER is a proprietary standard, yet OEM partners cite it directly.
Editorial take The toughest trade-off is architectural: keeping the LLM in an auditable cage suits regulated settings better than an end-to-end agent. Selma's +20%, Jamie's 70% and Rhea's 20% are all vendor-reported figures with no third-party audit, and in OpenDialog's own press release Jamie's 70% was still a cost reduction rather than a self-service rate. Implementation is heavy: Sterling's actual build took 14 weeks, and the pricing page returns 404.
Pricing
Custom pricingCustom quoteService as Software billed monthly with no long-term commitment, charged per work task the agent actually executes.
A monthly subscription against a minimum committed volume, with overage billed per work task; annual billing takes a further ~20% off.
A white-label AI engine for technology vendors, billed on usage aggregated across end clients.
A discontinued tier that survives only in third-party directories and search index snapshots; it is not reachable on the current website and does not reflect current pricing.
Additional notes
The official /pricing returns 404, and the Billing and plans page in the official docs points users to that same 404 address; the site has no self-serve sign-up or free trial entry point, and the only path is Book a demo. A free Experiment tier and $65/$325 self-serve subscriptions did exist historically, but both are discontinued and should not be quoted as current pricing.
FAQ
User reviews
User Reviews
No reviews yet
No one has reviewed this tool yet