brianletort.ai
← Library

Zurich Commercial Insurance (Zurich Insurance Group)

Company-reportedIntake redesignedStep before the decision

Cytora headless risk digitization feeding Z-Horizon+ underwriting workbench

AI can digitize and triage submissions before underwriters make the underwriting decision.

Insurance · Five countries in the initial 90-day wave (live by mid-October 2025); 11 countries live as of May 2026; expansion toward 22 countries by end of 2026

Collections: Queue eliminated · Regulated autonomy

An editorial scene for Zurich Commercial Insurance (Zurich Insurance Group) contrasts manually read, review, and interpret every incoming risk submission (unstructured, multilingual documents) with converts multilingual submissions into structured risks aligned with zurich's underwriting framework. in the cytora headless risk digitization feeding z-horizon+ underwriting workbench workflow.

Executive brief

The operating-model shift, in one view.

The breakthrough was not the model; it was standardising the intake schema and the execution workbench first, so a headless AI layer could be dropped into five countries in 90 days. Deployment velocity came from the target operating model, not from AI heroics.

AI value · Manual triage time per submission

Company-reported

15 minutes (80% reduction)

Zurich-attributed figures carried by independent trade press (Reinsurance News; The Insurer) and the Cytora vendor page; no independent measurement and no Zurich-owned disclosure located.

Before

Manually read, review, and interpret every incoming risk submission (unstructured, multilingual documents) → Re-key and re-format submission data across multiple disconnected systems → Assess, quote, and bind from manually prepared file

After

Converts multilingual submissions into structured risks aligned with Zurich's underwriting framework. → Search external data sources, including web research, to enrich the risk with information underwriters need → Routes a reported 95% of submissions to underwriters without manual triage, in the local language. →…

Human boundary

AI owns intake digitisation, triage, and enrichment; underwriters retain assess/quote/bind authority. No automated binding or auto-declination is disclosed in any retrieved source.

Why it matters

AI can digitize and triage submissions before underwriters make the underwriting decision.

This case is company-reported. Use it for the operating-model shift; do not treat the numbers as independently measured.

How the work changed

Before

How the work ran before the change.

  1. Step 1 of 3

    Underwriter

    Manually read, review, and interpret every incoming risk submission (unstructured, multilingual documents)

  2. Step 2 of 3

    Underwriter

    Re-key and re-format submission data across multiple disconnected systems

  3. Step 3 of 3

    Underwriter

    Assess, quote, and bind from manually prepared file

What changed

AI can digitize and triage submissions before underwriters make the underwriting decision.

Decision rightHuman decision retained

After

How the same work runs now.

  1. Step 1 of 4

    Cytora headless intelligence layer

    Converts multilingual submissions into structured risks aligned with Zurich's underwriting framework.

    ControlMaster schema validation; reported 98% digitisation accuracy

  2. Step 2 of 4

    AI agents

    Search external data sources, including web research, to enrich the risk with information underwriters need

    ControlZurich-defined global view of risk with local market variation

  3. Step 3 of 4

    Intake straight-through flow

    Routes a reported 95% of submissions to underwriters without manual triage, in the local language.

    ControlAutomated triage rules; exceptions surface to underwriters

  4. Step 4 of 4

    Underwriter

    Assess, quote, and bind the decision-ready risk (or pass to specialist underwriting platforms)

    ControlHuman underwriting decision retained

Process model built from the published workflow evidence for Zurich Commercial Insurance (Zurich Insurance Group). Every step, actor, and control appears in full below.
Every step, actor, and control

Exception path

Submissions that fail digitisation or triage rules surface to underwriters; complex risks pass from Z-Horizon+ to specialist underwriting platforms.

Work removed

  • Manual triage of every submission (reported 80% time reduction)
  • Re-keying and re-formatting submission data
  • Switching between multiple disconnected systems

Decision authority

AI owns intake digitisation, triage, and enrichment; underwriters retain assess/quote/bind authority. No automated binding or auto-declination is disclosed in any retrieved source.

Before

  1. 01

    Underwriter

    Manually read, review, and interpret every incoming risk submission (unstructured, multilingual documents)

    Control: Individual judgment; no consistent prioritisation; capacity constrained to a fraction of submissions received

  2. 02

    Underwriter

    Re-key and re-format submission data across multiple disconnected systems

    Control: Manual data entry

  3. 03

    Underwriter

    Assess, quote, and bind from manually prepared file

    Control: Human underwriting decision

After

  1. 01

    Cytora headless intelligence layer

    Converts multilingual submissions into structured risks aligned with Zurich's underwriting framework.

    Control: Master schema validation; reported 98% digitisation accuracy

  2. 02

    AI agents

    Search external data sources, including web research, to enrich the risk with information underwriters need

    Control: Zurich-defined global view of risk with local market variation

  3. 03

    Intake straight-through flow

    Routes a reported 95% of submissions to underwriters without manual triage, in the local language.

    Control: Automated triage rules; exceptions surface to underwriters

  4. 04

    Underwriter

    Assess, quote, and bind the decision-ready risk (or pass to specialist underwriting platforms)

    Control: Human underwriting decision retained

Work that left the path

  • Manual triage of every submission (reported 80% time reduction)
  • Re-keying and re-formatting submission data
  • Switching between multiple disconnected systems
  • Per-country custom intake builds (replaced by a reusable Intake Master Schema)

Human role before

Underwriters manually read, interpreted, triaged, and re-keyed every submission across disconnected systems, leaving capacity for only a fraction of inbound volume and inconsistent prioritisation.

Human role after

Underwriters receive complete, decision-ready risks directly in the Z-Horizon+ workbench in local language and spend their time assessing, quoting, and binding; manual triage, re-keying, and system switching are eliminated.

AI roleCytora operates as the upstream headless intelligence layer: digitising unstructured multilingual submissions, triaging intake, and enriching risks via agentic search of external data sources, delivered into Z-Horizon+ (a configured Microsoft Dynamics 365) as the downstream system of execution.

Outcomes

Manual triage time per submission

Company-reported

75 minutes→15 minutes (80% reduction)

90-day initial rollout (first five countries live by mid-October 2025); metrics reported May 2026 · Five countries in the initial wave; 11 countries live as of May 2026

Zurich-attributed figures carried by independent trade press (Reinsurance News; The Insurer) and the Cytora vendor page; no independent measurement and no Zurich-owned disclosure located. The Cytora page mixes achieved language with forward-looking 'clear path towards 15-minute triage time' language for the self-serve phase.

Intake straight-through processing rate

Company-reported

10%→95%

90-day initial rollout; reported May 2026 · Five countries (initial wave)

Company-attributed via trade press; not independently measured.

Digitisation accuracy

Company-reported

70-80%→98%

90-day initial rollout; reported May 2026 · Five countries (initial wave)

Company-attributed via trade press; accuracy methodology not disclosed.

What leaders can reuse

Anti-pattern

Reading vendor roadmap language ('clear path', '20+ markets in 16 months') as achieved outcomes; only the five-country wave (11 by May 2026) is delivered.

Questions

  1. 01Do we have a standardised execution workbench that an intelligence layer could feed, or would each deployment be a custom build?
  2. 02Which intake metrics would we let a vendor publish about us, and would we issue our own disclosure?
  3. 03What is our equivalent of the 75-to-15-minute triage compression, and who measures it independently?

Portability conditions

  • A globally standardised downstream workbench already in place (Z-Horizon+) before the AI layer arrives
  • Harmonised core data fields across countries and lines of business (Intake Master Schema)
  • Unstructured, multilingual, high-volume inbound document flow
  • Local underwriter engagement to build trust in the outputs

Reputation risk

low

Evidence and authority

What the public record supports.

Current · updated

2 independent, 1 vendor; publication outcomes are reported.

Bundle 1.0.0 · reviewed 2026-08-23 · stable ID 24f64271ddc66933

Related transformations

More in Insurance →

Deutsche Bahn · Intake redesigned

E-Check camera-gate, AI diagnostics, and robotic inspection

AI can flag deviations from train images; technicians validate findings and trigger work before manual repair.

Justice Connect · Intake redesigned

Free-text legal-problem classification

People seeking legal help should describe their problem, not diagnose its legal category before they can ask for help.

John Hancock / Manulife · Step before the decision

GenAI Quick Quote preliminary assessments

A licensed underwriter need not assemble every preliminary before an agent gets a non-binding indication.

AIA Group · Straight-through

GenAI-enabled medical claims straight-through processing

Medical claims require a human to read every document before payment. AIA instead built a three-tier GenAI stack (OCR intake, report summarization, auto-adjudication plus FWA anomaly detection) so most claims complete without human touch.

Sources

Read the evidence, freshness, caveat, and version policy.