Just as hydraulic fracturing made previously unviable oil reserves economically extractable, AI-powered migration tooling is changing the unit economics of PAS migration – making books that were too small to justify migration suddenly viable. Here is what that shift looks like in practice.
Insurance PAS migrations have an economics problem. Not a technology problem – an economics problem.
The technology to migrate policy data from one system to another has existed for decades. ETL platforms, data migration consultancies, SI delivery teams – none of this is new. What is new is the possibility of doing it at a cost that makes sense for anyone other than the largest carriers.
I have been thinking about this through the lens of an analogy that keeps proving useful: shale oil.
Before hydraulic fracturing, the oil industry knew exactly where shale reserves were. The geology was well-understood. The oil was there. But the cost of extracting it exceeded the value of what you would get out. The reserves were real but economically unviable.
Insurance migration has the same structural problem. Mid-market insurers with 50,000 to 500,000 policies know they need a modern PAS. The business case for better underwriting tools, automated bordereaux, real-time analytics – it is clear. Their boards can see the strategic value.
But the migration cost kills the business case.
Traditional migration approaches rely on manual discovery, manual mapping, and manual rekeying. Each of these scales linearly with book complexity. Double the number of policy classes and you roughly double the discovery effort. Add specialty lines with custom endorsement logic and the mapping effort increases non-linearly. Every exception that the automated pipeline cannot handle gets rekeyed by hand.
We have seen this firsthand. The Major Insurer programme (2016-2019) consumed three years and eight figures migrating 1M+ personal lines policies from mainframe to Duck Creek. The per-policy migration cost at that scale was already eye-watering. For a mid-market insurer with a fraction of the policy count, the same methodology would produce a per-policy cost that no board would approve.
The reserves were there. The extraction cost was too high.
What hydraulic fracturing did for shale oil, AI-powered tooling does for insurance migration. It does not change what needs to happen – it changes the cost of making it happen.
Traditional discovery requires weeks or months of interviews, data dictionary analysis, and manual schema tracing. A team of consultants sits with business users, walks through screens, maps fields to database columns, and slowly builds a picture of what the source system actually does (as opposed to what the documentation says it does).
AI-powered schema inference compresses this dramatically. In our agricultural specialist engagement (H2 2025), we achieved 96% source-to-target data mapping from a completely undocumented black-box source system. No data dictionary. No vendor support. No schema diagrams. The previous vendor had walked away.
Automated discovery does not replace the need to understand the business. The underwriters still needed to confirm that Farm rating used acreage-based calculations and Equine policies used individual animal valuation schedules. But instead of spending months building the initial data model from scratch, we arrived at those conversations with a 96% complete model. The human expertise was applied to validation and refinement, not to basic cartography.
DataArc mapping uses AI to identify relationships, patterns, and business logic that manual analysis misses. Fields that a human analyst might classify as "unknown" – columns like FLD_17, CALC_AMT_3, RSV_CD – are cross-referenced against policy behaviour patterns to infer their purpose.
In the agricultural specialist engagement, we found 47 distinct tables with meaningful policy data, many with opaque column names. The mapping canvas identified relationships between tables that were not formally defined through foreign keys – relationships that existed only in application logic and batch job sequencing.
This is the kind of work that consumes months in a traditional engagement. A consultant maps one table, traces its relationships, documents the findings, moves to the next. AI-assisted mapping does not eliminate the work – it compresses the timeline and catches patterns that sequential human analysis misses.
100% programmatic migration eliminates manual rekeying entirely. Every record migrates through versioned, auditable rules. No exceptions handled by hand. No underwriter sitting at a terminal re-entering policies that the pipeline could not process.
Zero manual rekeying means the methodology scales without scaling the team. The cost of migrating 100,000 policies is not dramatically different from migrating 50,000 policies, because the effort is in building and validating the rules – not in executing them.
This is the fracturing technology equivalent. The per-unit extraction cost drops because the marginal cost of each additional record approaches zero once the rules are built and validated.
When discovery takes days instead of months, mapping is AI-assisted instead of purely manual, and execution is 100% programmatic instead of partially rekeyed – the per-policy migration cost drops dramatically.
Books that were previously "too small to justify migration" become viable. A specialty insurer with 75,000 policies and complex endorsement logic can now build a migration business case that the board will approve. The cost no longer exceeds the strategic value.
PAS vendors benefit too. Every modern PAS vendor – Guidewire, Duck Creek, Insurity, EIS, and others – has prospects who cannot buy because they cannot migrate. The prospect wants the platform. The vendor wants the sale. But the migration cost sits between them like a toll bridge that charges more than the destination is worth.
Reduce the toll, and the traffic flows.
The shale oil parallel is direct: the reserves (insurance books stuck on legacy systems) were always there. The demand (for modern PAS platforms) was always there. The technology to extract them economically is what changed.
This is the section where honesty matters more than marketing.
AI-powered tooling changes the economics of migration. It does not change the nature of what is being migrated. Insurance data carries regulatory obligations, policyholder contracts, and reserve calculations that require human judgement at specific points.
Underwriter validation. AI can map data structures and infer field purposes. But an underwriter must confirm that the mapped business logic actually reflects how the book operates. When we found custom rating algorithms in stored procedures during the agricultural specialist engagement, AI identified the patterns – but underwriters confirmed the intent.
Actuarial sign-off. Reserve calculations require actuarial review. An AI system can re-derive reserves from migrated data and flag variances. But the actuarial team must determine whether a variance represents a migration error or a correction of a source system error. In our engagement, we found cases where the migrated calculation was more accurate than the source – the actuaries confirmed this, not the algorithm.
Regulatory evidence. Regulators require human-accountable evidence. An automated reconciliation report is a tool for producing evidence, not evidence itself. The evidence is the actuarial sign-off, the compliance review, the board approval. AI produces the artefacts that humans review and attest to.
AI makes the migration affordable. Human expertise makes it trustworthy.
The shift is already happening. Our agricultural specialist engagement demonstrated the compressed timeline: 6 months from engagement to completion, against a typical 12-18 month timeline for comparable complexity. The per-policy cost was a fraction of what traditional approaches would have produced.
This was a two-person team delivering what traditionally requires a multi-team programme. Not because we worked harder – because the methodology eliminated the manual work that traditionally consumes 60-70% of migration effort.
KeystoneMigrate is the product version of this methodology. The AI-powered discovery, the DataArc mapping, the programmatic migration engine, the continuous evidence generation – packaged as software instead of consulting.
The goal is straightforward: make every insurance migration a business case the board can commit to. Not just the largest carriers. Every insurer with a book stuck on a legacy system that is holding back their business.
The reserves are there. The technology to extract them economically is here.
Want to understand what AI-powered migration would cost for your book? Book a discovery call and we will walk through the economics specific to your situation.
Get insurance migration insights delivered to your inbox. No spam — just methodology updates, industry analysis, and product news.
View our Privacy Policy