Summary

A captive software house, judged by its shipped systems

Any large insurer can announce an AI strategy. The test of a captive IT organisation is narrower: which production systems did its engineers actually rebuild, migrate or automate, and when? For Allianz Technology SE — the entity through which the Allianz group centralises software development, infrastructure and operations — the public record allows a partial but genuinely evidence-based answer, because several of its largest projects left traces in sources that are not Allianz's own.

The most consequential of these is the fate of the Allianz Business System. ABS is the group's core policy administration platform, and by 2019 it was described as handling around 60 million insurance policies across 19 countries https://news.microsoft.com/source/2019/11/14/allianz-partners-with-microsoft-to-digitally-transform-the-insurance-industry/. In November 2019, Allianz SE and Microsoft announced a strategic partnership under which core pieces of ABS would move to Microsoft Azure, with parts of the platform's core open-sourced; the same announcement set up Syncier, an Allianz-founded insurtech, to offer a configurable "ABS Enterprise Edition" as a service alongside an Azure-based software marketplace https://news.microsoft.com/source/2019/11/14/allianz-partners-with-microsoft-to-digitally-transform-the-insurance-industry/. The newswire distribution of the same release on the same date corroborates the commitment and its terms https://www.prnewswire.com/news-releases/allianz-partners-with-microsoft-to-digitally-transform-the-insurance-industry-300958051.html.

That Azure commitment, however, presupposed a prior and largely unheralded piece of execution. Independent German and English trade press reported that in mid-2019 Allianz Technology decided to migrate ABS from the mainframe to standardised x86 servers running Linux — a move the company's CTO framed with the sentence "We haven't been on the mainframe since Easter 2022" https://www.cio.com/article/479654/examining-allianz-technologys-decision-to-replatform.html. The German-language trade press covered the same decision under the headline "Die Allianz wechselt auf Linux" https://www.cio.de/article/3695492/die-allianz-wechselt-auf-linux.html. The scale of the cutover is what makes this credible execution rather than positioning: the ABS database, described as one of the largest Linux-based DB2 databases in existence, could only be migrated as a whole, and the complete dataset transfer reportedly took 48 hours over the Easter weekend of 2022 https://www.cio.com/article/479654/examining-allianz-technologys-decision-to-replatform.html. The same account describes a budget in the low three-figure millions of euros with payback in three to four years, a core project team of roughly 500 with around 3,000 people involved at times and about 1,000 employees used for testing, and the German regulator BaFin being informed in advance https://www.cio.com/article/479654/examining-allianz-technologys-decision-to-replatform.html. The replatforming is explicitly framed in that reporting as "the preliminary stage for cloudification" — which is precisely the sequence the Azure partnership then assumed.

It is worth being precise about what this evidence does and does not establish. The mainframe exit is independently reported with named executives, quantified scope and a regulator angle https://www.cio.com/article/479654/examining-allianz-technologys-decision-to-replatform.html. The Azure migration, by contrast, is a 2019 commitment whose subsequent completion status is not documented in the sources assembled here; what the record shows is intent with named partners and a dated signature, not a verified Azure landing of the policy core.

Allianz Direct: the software layer with third-party documentation

The strongest independent documentation of Allianz Technology's software work concerns Allianz Direct, the group's direct digital insurer. According to a case study published by the Cloud Native Computing Foundation, Allianz Direct began in a private cloud and migrated to the AWS public cloud in 2020 for elasticity and scale; in mid-2022 it reworked its CI/CD pipeline from Jenkins to Tekton for build and integration and Argo CD for deployment, cutting the release pipeline from roughly 200 workflows to 10–15 workflows comprising 40–50 components, with nine of the eleven pipeline tools being CNCF projects https://www.cncf.io/case-studies/allianz/.

A second, more technical CNCF-hosted architecture description fills in the organisational detail: the platform-engineering function serves approximately 500 engineers and 1,500 employees, runs on Amazon EKS and Fargate ECS with multi-tenancy at the AWS account, namespace and Fargate levels, and uses GitOps tooling built on Argo CD, Terraform with Atlantis, and Crossplane https://architecture.cncf.io/architectures/allianz/. The same description states the setup took about four years to reach as of October 2024, became an Allianz Group standard used by both B2C and new B2B businesses, and that three platform architectures exist with a consolidation goal toward two https://architecture.cncf.io/architectures/allianz/. Both documents are foundation-hosted but company-submitted, so their metrics carry the usual caveat; what they add beyond an Allianz press release is named technology, versioned architecture and a dated organisational snapshot that outside observers can interrogate.

The commercial side of the same platform is documented in two McKinsey publications. A case study describes Allianz Direct's "60-second claim" service, enabled by AI-based loss assessment, allowing customers to process a claim in under a minute by uploading photos and documents; it also states that roughly one-third of Allianz Direct employees work in technology or data roles and that the operating model uses cross-functional agile squads https://www.mckinsey.com/capabilities/tech-and-ai/how-we-help-clients/rewired-in-action/allianz-direct-advancing-as-europes-leading-digital-insurer. An interview with Allianz Direct's then-CEO Bart Schlatmann describes the one-business, one-platform model across multiple regulated European markets, a microservice architecture with automated testing, a single harmonised data and process model spanning agent, website and contact-centre channels, and home insurance going from idea to first-market launch in four months https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/a-digital-business-builder-an-interview-with-the-ceo-of-allianz-direct. These are consultancy and executive accounts — promotional in register — but they describe mechanisms, not just aspirations, and they are consistent with the engineering detail in the CNCF material.

Claims automation in production, on Allianz's own testimony

The AI portion of the record is almost entirely first-party, and should be weighted accordingly. Three systems are nonetheless worth detailing because each has a dated launch, a defined scope and a stated human-in-the-loop boundary.

Insurance Copilot was developed by Allianz Technology and launched in 2024 for automotive claims in Austria, originating as a late-2023 experiment. It automates data gathering, document and image analysis, decision support and the drafting of communications, while humans retain coverage and payout decisions; Allianz reports expansion to property claims and scaling across operating entities including Asia https://www.allianz.com/en/mediacenter/news/articles/250205-smarter-claims-management-smoother-settlements.html.

Project Nemo went further. Launched in Australia in July 2025 as Allianz's first integrated agentic AI claims solution, it uses seven specialised AI agents — planner, cyber, coverage, weather, fraud, payout and audit — for food spoilage claims. Allianz cites an approximately 80% reduction in claim processing and settlement time, a build completed in under 100 days, and an end-to-end automated workflow that completes in under five minutes before human review, with a human retaining the final payout decision https://www.allianz.com/en/mediacenter/news/articles/251103-when-the-storm-clears-so-should-the-claim-queue.html.

The same programme family shows up in Allianz's governance-focused reporting. A March 2026 company article describes a standardised AI lifecycle with use cases registered and monitored end-to-end, and quantifies production automation: in German pet insurance, fully automated processing accounted for 49.7% of claims in 2025, with payouts in hours for simple claims; in Australia, agent-based automation handles food spoilage claims where 95% of cases carry a maximum payout of AUD 500, cutting processing time for sub-AUD-500 claims from about seven days to under one day; potential rejections escalate to human handlers and customers retain human contact https://www.allianz.com/en/mediacenter/news/articles/260318-responsible-use-of-ai-at-allianz.html.

A single job posting for a Claims Solution Designer in Milan confirms that the claims-automation talent agenda extends into Allianz's European operating entities https://internal-careers.allianz.com/job/Milan-Claims-Solution-Designer_2059-20145/1403534033/ — contextual evidence of intent, not an outcome metric.

Scale, geography and the corroboration gap

Allianz Technology describes itself as more than 10,000 professionals working as one global team across hubs and delivery centres https://tech.allianz.com/en/who-we-are/our-people.html — a rounded, undated first-party figure. A third-party workforce-analytics estimate from Revelio Labs puts headcount near 11,396 in 2026 and around 12,100 in 2023, with the workforce concentrated in South Asia (~43.5%), Western Europe (~24.4%) and Southern Europe (~10.7%), and a functional split of roughly 40% finance and operations, 39.5% engineering and 20.5% sales and marketing https://www.reveliolabs.com/companies/allianz-technology/employees. That source is a modelled estimate derived from public profiles, is internally inconsistent in the retrieved snapshot, and conflicts with the company's own statement; it is usable only as a directional signal with those defects disclosed.

Vendor case-study evidence fills a regional gap. A SoftwareOne account of Allianz Ayudhya in Thailand describes an AWS migration begun in 2023, moving more than 500 servers in structured phases; the P&C phase in 2023 caused three days of disruption and moved cloud adoption from 30% to 45%, while the 2025 life and health phase limited disruption to one day and pushed adoption to 76% against the group's 75%-by-2025 cloudification target. The same account reports four data centres consolidated to one primary and one disaster-recovery site, server and data-centre run costs falling from roughly US$8 million in 2024 to a projected US$6 million by 2026, and more than 300 containerised applications migrated from Azure to AWS over six months https://www.softwareone.com/en/case-studies/global/finance/allianz-aws-migration. Every number here is vendor-authored and self-reported; the case study's value is geographic and architectural specificity, not independent verification.

What the record establishes, and what it does not

Taken together, the sources support a three-tier reading. Tier one — independently corroborated execution: the ABS mainframe exit at Easter 2022 with quantified scope https://www.cio.com/article/479654/examining-allianz-technologys-decision-to-replatform.html https://www.cio.de/article/3695492/die-allianz-wechselt-auf-linux.html, and Allianz Direct's cloud-native platform engineering documented in foundation-hosted technical records https://www.cncf.io/case-studies/allianz/ https://architecture.cncf.io/architectures/allianz/. Tier two — dated first-party commitments and launches: the 2019 Azure partnership https://news.microsoft.com/source/2019/11/14/allianz-partners-with-microsoft-to-digitally-transform-the-insurance-industry/ https://www.prnewswire.com/news-releases/allianz-partners-with-microsoft-to-digitally-transform-the-insurance-industry-300958051.html, Insurance Copilot's 2024 Austrian launch https://www.allianz.com/en/mediacenter/news/articles/250205-smarter-claims-management-smoother-settlements.html, Project Nemo's July 2025 deployment https://www.allianz.com/en/mediacenter/news/articles/251103-when-the-storm-clears-so-should-the-claim-queue.html, and the 49.7% pet-insurance automation figure https://www.allianz.com/en/mediacenter/news/articles/260318-responsible-use-of-ai-at-allianz.html. Tier three — self-reported or modelled metrics without independent verification: the 80% Nemo time reduction, the "60-second claim", the Ayudhya cost projections, and every headcount figure.

The honest conclusion is that Allianz Technology has demonstrably executed the hard, unglamorous parts of its software programme — a mainframe exit, a CI/CD rebuild, a platform engineering function with named tooling — while its AI-in-production story rests on its own accounting. The corroboration gap is not evidence of failure; it is the normal condition of corporate AI claims in 2026. But it means the AI ambition should be tracked through the same instruments that proved informative for the infrastructure arc: named systems, dated milestones and third-party technical records, rather than headline percentages.