Airclerk builds governed Claude workflows that connect to real systems, documents and teams - designed for regulated financial and professional services in New Zealand and Australia, sized for production.
Not chatbots. Workflow software with Claude at the centre - designed so the model does the judgement work and the platform handles state, evidence and control.
Most insurance work clusters around document-heavy review and triage. These are the shapes we know.
Submission packs, prior policies, claims history, missing-information detection, evidence-based recommendations.
Chronology, key issues, outstanding actions, evidence and citations across large claims files.
Compare current and prior year, identify changes, draft broker follow-up, track renewal state.
Classify shared-mailbox traffic, detect urgency, flag missing documents, draft responses.
Compare versions, identify exclusions, changes and coverage differences with evidence.
Permission-aware search across policies, procedures and documentation with citations.
We don't start with an open whiteboard. The first workflow is chosen on evidence.
One workflow at production quality beats five demos that never ship.
Document repositories, CRMs and core platforms stay the systems of record. Claude becomes the interface.
Audit, approval and evaluation patterns are part of the architecture - not a later phase.
Where Claude itself can be the workbench, we let it. Custom UI only where it actually earns its place.
Every implementation engagement starts from a clear plan. Most clients begin with the fixed-price Readiness Sprint.
Airclerk builds production-grade Claude workflows - not chatbots, but workflow software with Claude at the centre, where the model does the judgement work and the platform handles state, evidence and control. A typical build includes scoping, prompt and tool design, MCP and connector integration, document ingestion and analysis, human approval flows, audit logs, an evaluation harness and an operating model handover.
Airclerk builds around your existing systems: document repositories, CRMs and core platforms stay the systems of record, and Claude becomes the interface to them. Integration is handled through MCP servers and custom Claude tools, and data access is permission-aware, so workflows respect who is allowed to see what.
Governance is designed from day one as part of the architecture, not bolted on later - audit, approval and evaluation patterns are built in rather than added as a later phase. In practice that means human approval flows and review queues, audit logs and evidence capture, and evaluation and monitoring frameworks, so a person stays in the loop on the decisions that matter.
The workflows Airclerk builds are instrumented to produce evidence - audit logs, approvals and capture of what runs through them - which is the foundation of its AI Process Assurance approach: chartering the path the work should follow, recording what actually happened, and retaining an AI Workpaper. Airclerk records the work that runs through the workflows and tools it builds; it does not claim to capture everything outside those instrumented rails.
Most clients begin with the fixed-price Readiness Sprint, which produces a clear plan so the first workflow is chosen on evidence rather than from an open whiteboard. Airclerk scopes tightly - one workflow at production quality beats five demos that never ship - and common starting patterns include underwriting document review, claims file summarisation, renewal pack comparison, broker inbox triage, policy wording comparison and a permission-aware internal knowledge assistant.