Overview
Migration programs overrun more often than they finish on plan. McKinsey's survey of roughly 450 CIOs and IT decision makers found that 75 percent of cloud migrations ran over budget, the average company spent 14 percent more on migration than planned each year, and 38 percent of companies saw their migrations slip by more than a quarter. That research covers cloud migrations across industries, not insurance alone. In my experience, insurance programs carry extra weight on top of those numbers: state regulatory examinations, ACORD mapping, multi-year record retention, and open claims that have to survive the cutover.
The carriers that finish on plan are not the ones with the most expensive tooling. They are the ones whose CIO, COO, and Chief Underwriting Officer signed the same five principles before the migration team was on the clock.
In short: effective insurance data migration rests on five principles. Plan from regulatory requirements backward. Embed business Data Stewards instead of consulting them. Treat compliance as a data requirement from day one. Run a Phase Zero of profiling and cleansing before migration starts. Validate with full dress rehearsals, not record counts alone.
These principles are not original - any senior practitioner will recognize them. What I have tried to do is sharpen each one with the numbers and decisions that make it operational rather than a slide in a kickoff deck. For the conditions that make them necessary, start with our pillar guide to insurance data migration challenges. For the full seven-step checklist, from pre-migration planning to post-migration review, see our best practices for insurance data migration. This article is about how to apply the five principles that decide whether that checklist holds.
A note on fit. This playbook is written for mid-tier US P&C carriers in the $500M to $5B GWP range, which is the segment Decerto is built for. If you are a $5B+ enterprise carrier running a multi-country core replacement with a large systems integrator, Decerto is not the right partner for that program, and parts of this article will feel undersized for you.
The 5 principles at a glance
- Plan from the regulatory requirements backward, not from the technology forward.
- Embed business stakeholders, do not just consult them.
- Treat compliance as a first-class data requirement, not a cutover checkbox.
- Profile, dedupe, and standardize before the migration team starts.
- Validate end-to-end with dress rehearsals, not just record counts.
Below, what each one means in practice, with the numbers I anchor proposals against.
Principle 1: Plan from the regulatory requirements backward
Most generic migration advice is written for retail, manufacturing, or general enterprise IT. It misses what matters for an insurance carrier: ACORD mapping, state record-retention rules, state DOI examinations, claims-in-flight handling, and reinsurance contract data. A working insurance playbook starts from the regulatory requirements and works backward into the technical plan, not the other way around.
In practice, the planning team builds scope around three regulatory anchors before any technical decision is made. First, retention: on the programs I have run, 7 to 10 years for most P&C lines depending on state, longer for life and certain commercial books. Every historical record needs an explicit decision to preserve, transform, or archive. Second, examination continuity: regulators expect audit evidence to stay available through the migration window, not only after it.
Third, ACORD. ACORD's Property & Casualty standards come in two formats, AL3 (a one-way batch format for policy and commission data) and XML. Source and target implementations rarely match field for field, and every delta needs a documented mapping decision.
The numbers I anchor against. For a mid-tier US P&C carrier ($500M to $5B GWP), I scope 8 to 12 weeks of pre-migration regulatory mapping before the technical migration team is fully on the clock. Skipping this phase, in my experience, adds 3 to 6 months on the back end as compliance findings surface during UAT and force rework. I recommend treating regulatory mapping as a funded phase with its own deliverables, not as setup work bundled into the migration vendor's scope.
For the readiness side, our guide on how to prepare your insurance company for a data migration project covers the planning phase, the choice of migration approach, and a readiness checklist with typical timeframes.
Principle 2: Embed business stakeholders, do not just consult them
The single most expensive mistake I see in insurance data migration is delegating it fully to IT. Engineers move the data. Underwriters, claims adjusters, product managers, actuaries, and compliance officers know what the data means. They know which "Code 73" was used for two years to flag a specific reinsurance treaty before someone wrote it into a proper field. They know that the "Notes" column on a claims record has carried binding decisions since 2014.
The difference between "embedded" and "consulted" matters. Consulted means the business stakeholder attends a 90-minute kickoff and reviews the data mapping in week 16. Embedded means a Data Steward with named accountability for data definitions, quality, and correctness in their domain, assigned to the program at 30 to 50 percent of their time for its full duration. I have not seen a successful migration without this pairing. I have seen several fail without it.
What this looks like operationally. Appoint Data Stewards from underwriting, claims, billing, finance, and compliance. Run mandatory joint workshops where IT and the business build the mapping together. Make stewards accountable for sign-off on their domain's mapping decisions, not just for advisory input. I have worked with carriers that ran weekly Data Steward standups with executive escalation paths for unresolved mapping decisions. In my experience, those programs finished on plan more often than not; the programs that treated the role as advisory finished late.
The gap between IT and the business is the first blind spot in our article on what insurance teams forget when planning a data migration.
Principle 3: Treat compliance as a first-class data requirement
Compliance during migration is not a checkbox at cutover. It is a data requirement that shapes every mapping decision, every staging environment, and every audit trail. Nonpublic personal information, claims medical records, payment instruments, and reinsurance contract data are all in motion during a migration, often handled by vendor staff with temporary access. I have reviewed plans with strong production security and weak migration-period security, and I have seen vendor laptops still holding copies of production data ninety days after an engagement ended.
The non-negotiables are short to list and longer to implement: encryption at rest (AES-256) and in transit (TLS 1.3), role-based access control on the migration tooling with named owners, record-level audit logging (every movement, by whom, when), a penetration-tested migration environment before any source data is connected, and a documented data-handling agreement with every vendor on the program.
The 2026 regulatory perimeter. The NAIC Insurance Data Security Model Law (Model #668) requires licensees to run a written, risk-based information security program and to oversee third-party service providers that touch nonpublic information. During a migration, that includes your migration vendor. The NAIC's adoption map for Model #668, updated as of August 31, 2026, shows where a version of the law is enacted or pending in your states.
New York goes further. The second amendment to 23 NYCRR Part 500, finalized on November 1, 2023, phased in new requirements through November 2025, including asset inventory and multi-factor authentication, and a temporary migration environment is an asset like any other. The NIST Cybersecurity Framework 2.0, published in February 2024, is the common control baseline; its new Govern function and its supply-chain emphasis map directly onto vendor oversight. California adds its own layer through the Insurance Information and Privacy Protection Act (Insurance Code §791 et seq.).
I recommend against starting cutover until your migration environment has passed an independent penetration test and the vendor's access controls have been verified against your production standards. Not lower than your production standards. The same. Our piece on data security during data migration in the insurance industry walks through the confidential-data assessment, the legal assessment, and the security plan in more detail.
Principle 4: Profile, dedupe, and standardize before the migration team starts
Teams are optimistic about their own data. The reality is harsher. Years of patches, ad-hoc fixes, and gaps in governance produce data erosion that is invisible from the outside. Gartner puts the average cost of poor data quality at $12.9 million per organization per year. The Deloitte 2026 global insurance outlook points to fragmented, messy data and outdated systems as the reason many insurers struggle to get value from AI, and tells carriers to prioritize data quality, integration, and master data management first.
For a carrier, this means customer duplicates ("John Smith" vs. "J. A. Smith"), incomplete addresses, inconsistent date formats, free-text fields used as structured fields, historical records that are simply wrong, and binding decisions stored in places that no longer have business owners. Move that into a new core and the new platform automates the errors. User trust collapses within weeks, and the investment looks broken before it has a chance to perform.
Phase Zero is the fix. Before the migration team is on the clock, run a dedicated profiling, auditing, and cleansing phase: anomaly detection, deduplication, format standardization, and gap-filling against source-of-truth records. In my experience, Phase Zero takes 8 to 12 weeks for a mid-tier carrier. ML-assisted data profiling compresses that to 6 to 8 weeks, which is one of the better-returning investments I have seen in 2026 programs. On the carriers I have worked with, measured data quality typically starts in the 60 to 80 percent range and needs to reach 95 percent or better before cutover.
What every Phase Zero should produce. Four deliverables, each with a named owner:
- A profiling report by domain (party, policy, claims, billing, financial) with quantified completeness, validity, uniqueness, and consistency scores against the carrier's pre-defined targets.
- A deduplication output with merge candidates ranked by confidence and reviewed by domain stewards.
- A standardization plan covering dates, addresses, codes, and coded enums.
- A gap analysis that identifies critical fields where source data is missing or unreliable, and names the business owner for each gap.
Without those deliverables, Phase Zero is decorative.
To be transparent: Decerto's Data Migrator is built for this work on insurance-specific schemas (ACORD mapping, party model, policy lifecycle states, claims hierarchies), but if your data sits in a Talend or Informatica stack you already operate, Phase Zero can run there too. The tool matters less than running the phase rigorously.
Principle 5: Validate end-to-end with dress rehearsals, not just record counts
In many migration plans, the "Testing" chapter is dangerously thin. Teams focus on technical validation: does the source record count match the target, and were the fields transferred correctly? That is necessary, and it is not enough. The real question is whether the business can run on the migrated data.
Users do not work with individual tables; they work inside business processes. Can you process a multi-vehicle claim with several injured parties using only migrated data? Can the system generate a renewal offer for a commercial policy with ten endorsements in the past year? Can an adjuster pull the full handling history of a litigated file, including internal coverage decisions? Answering those questions requires end-to-end User Acceptance Testing, run by business users, on data that reflects the complexity of their day.
The dress rehearsal pattern. I have not run a successful migration with fewer than three full dress rehearsals into a test environment that mirrors production. Three is the floor. On one large multi-line program, we ran seven, and rehearsal five caught a regulatory reporting bug that would have caused six-figure fines. Without the rehearsals, we would have shipped it. In my experience, the rehearsal that finds the expensive bug is rarely the first one.
Each rehearsal should produce a reconciliation report scored against quantitative success criteria: record-level match rates by domain, transaction-level reconstruction tests for active policies and open claims, preservation of regulatory artifacts against the audit trail, and end-to-end business process tests run by business stewards. The COO and the CIO sign those criteria before the first rehearsal, not after results come in.
Rehearsals also set your cutover window and your rollback trigger. For that side of the plan, our guide on minimizing downtime during insurance data migration covers parallel system operation, real-time synchronization, and backup and recovery.
Decerto migration reference: Generali Group Poland
When Generali Group Poland acquired another insurance company, it had to merge not only operations but the entire data estate of the two entities. Every insurance product, business process, and customer record had to move from the acquired company's IT architecture onto Generali's infrastructure. The source data was incomplete and inconsistent, and the migration had to fit a short, predefined window with no room for major errors, data loss, or customer service downtime. Generali chose Decerto as its technology partner for the program.
One limitation I should be honest about: this was a life insurance migration in Poland, not a US P&C program. The regulators and the product structures differ. The mechanics that made it work do not.
- Analysis first: a field-by-field analysis of every relevant object and process in both companies, mapping the acquired company's products, workflows, and customer data against Generali's business processes before anything was built.
- A dedicated migration tool: built for this program to import legacy data, validate it for completeness and accuracy, correct and standardize it to Generali's quality standards, transform it into the target data model, export it, and produce migration reports for tracking and verification.
- Rehearsed, not hoped: dozens of trial migrations plus detailed technical and business testing before go-live.
- One weekend: the production migration ran over a single weekend to keep disruption to Generali's operations to a minimum.
- Outcome: no major errors, no data loss, financial transaction balances intact, and better data consistency and accuracy than the source system had.
The five principles in this article are the ones that program ran on: analysis before build, Generali and Decerto teams working as a single unit, data validated and corrected before it moved rather than after, and dozens of trial migrations that caught what record counts alone would have missed. The full write-up is in our Generali Group Poland data migration case study.
FAQ
Five principles separate insurancemigrations that finish on plan from those that overrun: plan from regulatoryrequirements backward, embed business Data Stewards rather than just consultingthem, treat compliance as a first-class data requirement, run a rigorous PhaseZero before the migration team starts, and validate end-to-end with dressrehearsals rather than record counts.
For a mid-tier US P&C carrier,the migration workstream itself typically runs 6 to 18 months, inside a fullcore modernization of 18 to 36 months. Big Bang approaches sit at the short endof that range and carry the most cutover risk; Strangler Fig approaches runlongest but keep the business running on both systems during transition. Add 8to 12 weeks of regulatory mapping and 8 to 12 weeks of Phase Zero on the frontend.
Phase Zero is a dedicated profiling,auditing, and cleansing stage that runs before the migration team is on theclock. It typically lasts 8 to 12 weeks for a mid-tier carrier, and 6 to 8weeks with ML-assisted data profiling. It produces a profiling report, adeduplication output, a standardization plan, and a gap analysis with namedbusiness owners.
The buying committee includes theCIO (final technical decision), the Enterprise Architect (patterns and tools),the COO (cutover and operational continuity), the CFO (budget and risk), theCompliance Director (NAIC and state DOI), and business Data Stewards fromunderwriting, claims, billing, and finance. Each role has named accountabilitybefore the program starts.
Three full rehearsals is the floor.Each one should mirror production, produce a quantitative reconciliationreport, and include end-to-end business process testing by business stewards.On one large multi-line program we ran seven, and rehearsal five caught aregulatory reporting bug that would have caused six-figure fines.
Through reconciliation, not hope.Define record-level and transaction-level match criteria per domain before thefirst rehearsal, run at least three full dress rehearsals, reconcile financialbalances as well as record counts, and keep a tested rollback path until thefinal reconciliation is signed. On the Generali Group Poland program, thatapproach ended with no data loss and financial transaction balances intact.
In my experience, migration accountsfor 20 to 30 percent of a full core modernization budget, which typically sitsin the $1M to $15M range for a mid-tier US P&C carrier ($500M to $5B GWP),depending on portfolio complexity, target architecture, and legacy depth.Budget contingency matters: McKinsey found the average company spending 14percent more on cloud migration than planned each year.
Talk to Decerto about your migration strategy
If you are scoping a migration program now, the five principles work as a self-assessment checklist. Where your program stands against each one is what separates programs that finish on plan from programs that do not.
I run a 30-minute Migration Architecture Review covering where your program sits against each principle, preliminary RTO and RPO targets, and a cost estimate range. Your data, your gaps, your priorities. No demo loop.
Sources
- McKinsey & Company (2021). Cloud-migration opportunity: Business value grows, but missteps abound. McKinsey ITaaS Survey 2020-21.
- National Association of Insurance Commissioners (2017). Insurance Data Security Model Law (Model #668).
- National Association of Insurance Commissioners. Implementation of Model Act #668, Insurance Data Security Model Law: state adoption map (status as of August 31, 2026).
- New York State Department of Financial Services (2023). 23 NYCRR Part 500, Cybersecurity Requirements for Financial Services Companies, Second Amendment (effective November 1, 2023).
- National Institute of Standards and Technology (2024). The NIST Cybersecurity Framework (CSF) 2.0, NIST CSWP 29, February 26, 2024.
- California Legislature. California Insurance Code, Division 1, Part 2, Chapter 1, Article 6.6: Insurance Information and Privacy Protection Act (§791 - §791.29).
- ACORD. Property & Casualty Data Standards (AL3 and XML).
- Gartner. Data Quality: data and analytics topic research (average annual cost of poor data quality).
- Deloitte Center for Financial Services (2025). 2026 global insurance outlook.
.avif)


.avif)
.avif)

