Portfolio migration in the insurance sector: Why data is the new legacy problem and how AI is changing that

Proven migration methods meet AI-powered analysis – a practical look at the modernisation of portfolio management systems in insurance IT

Insurers are modernising their portfolio management systems, consolidating claims systems, realigning payment processing and commission processes, and modernising their data platforms for analytics and generative AI. The real bottleneck is rarely the new technology itself, but rather understanding and migrating the existing portfolio data. This is precisely where portfolio migration becomes the critical path of the transformation.

This blog post explains why portfolio migration in insurance is so challenging, which approaches have proven effective, and provides an example of how an AI-powered tool can specifically address key bottlenecks.

It is not just systems that are legacy – the portfolios are too

Anyone involved in insurance IT modernisation programmes will recognise the pattern: the new core system has been selected, the architecture and operating model are in place, and the rollout is planned. By this stage at the latest, portfolio migration takes centre stage.

For it is not only the legacy systems that have evolved over time. The portfolios themselves also contain tariff and product logic, implicit business rules and specialist exceptions that have developed over decades. Many of these rules are embedded in coding, flags, key lists or procedures for premium calculation, benefit verification or cancellation. Data structures often reflect compromises that have evolved over time, whilst documentation is incomplete or out of date. It is precisely at this interface between legacy portfolios and the new target system that portfolio migration almost always becomes the bottleneck of a transformation.

Portfolio migration is a multi-discipline challenge

Portfolio migration in insurance is not a single technical step, but an interplay of business expertise, technology, organisation and governance. Typically, there are multiple source systems with different data models for portfolios, benefits or claims, commissions, partners and payment transactions. Added to this are complex contract portfolios with historical tariff generations, individual agreements and special provisions, as well as high requirements for traceability, audit compliance and contract-specific accuracy. Particularly in regulated environments, tight timeframes for system switchover represent an additional risk factor.

The success of a portfolio migration therefore depends primarily on four factors: the business interpretation of legacy data, a robust technical migration design, a structured testing and cutover process, and governance that meets audit and regulatory requirements.

Which approaches have proven successful

In insurance projects, an approach has become established that is tailored to this business and regulatory complexity.

A key success factor is evolutionary modernisation rather than a ‘big bang’. Portfolios are not migrated all at once, but in stages, for example by line of business, product line or system domain.

This reduces risks, creates technical learning loops and allows insights from early migration waves to be applied to later ones.

Equally important is a phase-based approach with clear maturity levels. Successful projects make a clear distinction between analysis, design, migration and testing, as well as cutover and hypercare. Multi-stage test runs, in which the migration is refined step by step – from technical migration through test migrations to the final dress rehearsal – have proven their worth. In this way, the cutover is run through in full with production-ready data and processes before the actual switchover.

Audit compliance and clear roles from the outset

In regulated insurance companies, audit compliance, traceability and repeatability are not downstream documentation tasks. They must be incorporated into the migration design from the outset. This includes a clear separation of the business and technical migration concepts, end-to-end logging of migration runs, repeatable runs with clear rule sets, and business validation at the contract and transaction level.

Equally crucial is the clear division of responsibilities between the business department, IT and migration managers. The business department is responsible for the business rules of the legacy migration, the IT architecture, interfaces and operation of the migration environment, whilst a dedicated migration team implements the technical transfer. This separation is a key success factor, particularly for legacy systems that have evolved over time.

The blind spot: source analysis as a constant bottleneck

Even when the process model, governance and tools are in place, one bottleneck almost always remains: understanding the business-technical structure of legacy data. Data experts and data engineers invest a great deal of time in exploratory analysis before reliable mappings can be established. At the same time, knowledge of legacy portfolios is often tied to specific individuals.

This source system analysis is the real bottleneck in many projects. It is precisely here that AI-supported tools can deliver a noticeable productivity gain.

Where AI provides concrete support in legacy migration

Modern AI tools do not replace migration methodologies. They complement them where the greatest manual effort is currently required: in understanding historically evolved data structures and in establishing reliable field mappings between source and target systems.

Here, adesso has developed an AI tool that tackles precisely these bottlenecks. It analyses the database structures of legacy systems and automatically generates entity-relationship diagrams (ERDs), table and attribute descriptions, as well as business domain mappings, such as for contracts, claims, partners or payments. This enables data structures to be mapped more quickly and systematically prepared for the mapping process.

Furthermore, the tool makes existing documentation, such as business concepts or data dictionaries, centrally accessible via a RAG-based search. Business departments and migration managers can access this information using natural language without having to wade through scattered sources or database structures.

Another key benefit lies in the preparation of mapping workshops. Based on semantic similarities, the tool suggests field mappings between the source and target models. These candidates are validated by a language model that evaluates each mapping across several dimensions: semantic similarity, data type compatibility and consistency with the data model.

The result is a traceable confidence score that makes it clear why a mapping has been classified as suitable or unsuitable.

In the human-in-the-loop approach, subject matter experts evaluate the suggestions, correcting, supplementing or rejecting them; guided by the model’s reasoning, without having to delve deeply into the technical details of the source database themselves. The technical responsibility remains with humans; AI significantly accelerates the preparatory work.

Validated field mappings and transformation rules are stored as versioned artefacts, form the basis for technical implementation and can be further refined in subsequent workshops.

Migration methodology and AI go hand in hand

The added value of AI does not lie in replacing best practices. A phased approach, evolutionary modernisation, audit-proof governance and clear roles remain the backbone of successful legacy migrations. AI helps where a particularly large amount of manual analysis work is currently required: in source analysis, in the structuring of knowledge and in the preparation of robust field mappings.

For insurers, this shifts the focus: away from person-dependent structural knowledge and time-consuming preparatory work, towards technical assessment, validation and management of the migration. Field mappings, decisions and approvals are documented in a more structured manner, and projects become easier to manage overall.

Conclusion

Proven process models, clear responsibilities and governance designed for audit compliance remain the foundation of any legacy migration. AI tools can address the biggest bottleneck where a great deal of time and expert knowledge is currently tied up: in the analysis of established data structures and the preparation of robust field mappings.

Insurers that consistently integrate AI-supported analysis and mapping support into their migration programmes are not creating an alternative methodology, but rather strengthening the existing one. Structural knowledge becomes more readily available, preparatory work is accelerated, and specialist departments can focus more on decision-making. The tools required for this are already available today.