Partner practices
Databricks partnershipPlatforms we build on
Microsoft AzureGoogle CloudAnthropic ClaudeOpenAISnowflakeHow we choose
Regulatory frameworksVendor neutrality
TriageIQ2026
The twelve-agent pipeline already running in production at a Lloyd’s-market MGA, productised. Reads each inbound submission, scores it against your appetite, and routes it with a context pack attached.
The reality
Senior underwriters were spending around 40% of their time on administration rather than decisions, and there was no audit trail on routing at all — which is an FCA exposure quite separate from the lost business.
Inbox rules and shared mailboxes route on sender and subject line, not on risk or appetite.
A single large model asked to “triage this submission” produces a plausible answer with no traceable reasoning and no consistent structure.
Triage needs a chain of single-purpose agents, each producing a structured, auditable output that the next one consumes.
How it delivers value
01
Ingestion, classification, extraction, sanctions screening, capacity check, risk profiling, appetite scoring, historical comparison, priority scoring, XAI summary, routing and context pack. No monolithic prompt anywhere in the chain.
02
Appetite scoring runs against your own criteria, not a market default, and the score is explained in terms a underwriter recognises.
03
The underwriter receives the submission alongside extracted entities, screening results, capacity position and comparable historical risks.
04
Any flagged decision routes to a person, and the override is logged. The pipeline moves work forward; it does not make the underwriting call.
Capabilities
The approach
Written for the technical evaluator rather than the buyer. If you are going to put this in front of an architecture review board, this is the section they will read.
Each agent is purpose-built for a single task and emits a structured output. There are no monolithic prompts and no hallucination chains, because every step is validated before the next one consumes it.
The output schema between agents is defined up front. That is what allows any single step to be replaced, re-run or audited without disturbing the rest of the pipeline.
Sanctions screening and capacity checks are gates rather than signals. A submission does not progress past a failed gate, regardless of how attractive the risk scores.
Appetite and priority scoring run against your written criteria and your historical decisions, so the ranking reflects how your firm actually selects risk.
The pipeline has been running in production at an MGA since deployment. TriageIQ is that system generalised, not a new build seeking its first customer.
The breakthrough
These are delivered results from the production deployment this product is built on.
