In most insurers and reinsurers, finance, actuarial, and accounting run on parallel systems and reconcile after the fact. I built one computational graph under all three — policy data, cash-flow engine, measurement bases, ledger, reporting — then the AI layer that can walk it. I shipped the first version, brought on a team to take it to production, then wired the insight layer across the modules. Private equity firms adopted it. More than $3M in engagements.
In most insurers and reinsurers, finance, actuarial, and accounting run on parallel systems and reconcile after the fact. Actuarial models project cash flows and reserves. Finance builds plans and capital views on a separate assumption set. Accounting posts the results on yet another basis — GAAP / LDTI, IFRS 17, statutory, tax. The cost of that split is enormous and mostly invisible, because it is baked into headcount and cycle time.
Reconciliation is the dominant workload. A huge share of actuarial and finance hours goes to explaining why three systems disagree: broken data lineage, assumption mismatches, timing differences. LDTI and IFRS 17 made this worse. They force actuarial granularity into the ledger, so the fight is no longer at the total. It is at the cohort.
There is no one version of the truth for assumptions. Mortality, lapse, rates, expenses should flow into reserves, planning, pricing, and capital at the same time. Today an assumption update can take a quarter to propagate. Disagreements between pricing and valuation are common and often undocumented.
Speed is strategy. Close cycles that take weeks cannot support a deal calendar. Reinsurers feel this first: deal pricing, capital deployment, and retro decisions depend on knowing the current position now. A Bermuda writer that can re-project the whole balance sheet overnight can bid on blocks a slower incumbent cannot even evaluate in time.
And there is no bidirectional traceability. A ledger entry should walk back to the policy-level cash flow that generated it. A model change should walk forward to its P&L. That walk is the foundation of attribution, audit defense, and whether management trusts the numbers. On siloed systems it does not exist.
The integrated system is a single computational graph: policy data → cash-flow engine → measurement bases → ledger → reporting and planning. Assumptions and scenarios are parameters on that graph, not copies sitting in three workbooks.
One assumption set feeds reserves, pricing, planning, and capital at once. An update lands everywhere overnight, not over a quarter. Statutory, GAAP / LDTI, and IFRS 17 are projections of the same cash flows, not three exports that you argue about in week three of the close.
A number on a dashboard has a parent: a journal, a run, a methodology version. You walk it back to the policy cash flow. You walk a model change forward to earnings and capital before you close. That is the substrate. Without it, everything downstream is a story about last month.
I shipped the first version alone — ingest, the warehouse, the first engines, a surface a CFO could sit in. Then I brought on a team and we took it to production: environments, an audit log, role-based views, an overnight book that actually closes.
Policy and in-force extracts, custodian and OMS files, and GL feeds land in a landing zone. dbt builds a canonical Delta / Parquet warehouse on Databricks. Python calculation services behind FastAPI are versioned the way you version a service. Temporal runs the overnight graph: close positions, project cash flows, measure each basis, post journals, refresh the plan. An actuary sees assumption sets and cash-flow testing. A CFO sees earnings, capital, and the close. Same graph. Different cut.
After that graph was live I built the insight layer and wired it through every module. The sequencing is the point. Integration first. AI second.
AI on fragmented systems mostly automates the reconciliation swamp. Useful, and marginal. Once the graph exists, the value changes in kind. No model can do the things below on silos, however good the model is.
Explanation and attribution on demand. “Why did the CSM move $40M this quarter?” is the biggest pain in LDTI / IFRS 17 reporting. An agent walks the lineage and decomposes the movement — assumption changes versus experience variance versus new business — in minutes, with citations to the actual model runs. Not a two-week pack.
Anomaly detection with context. A lapse spike in one cohort will hit reserves, earnings, and the plan at the same time. On an integrated pipeline you see that before close. On silos you flag an anomaly in one book and have no idea what it does downstream.
Scenarios in natural language. “Show me earnings and RBC if rates drop 100bp and we reinsure 30% of the VA block” is a query, not a project. The AI orchestrates the runs. The graph guarantees the answers are consistent across bases.
Continuous assumption monitoring. Agents compare emerging experience to the assumption set across the whole book and propose updates with the financial impact on every basis at once. The annual assumption review becomes a process that runs all year.
Judgment leverage, not replacement. The scarce resource is senior actuarial and finance judgment. Integration plus AI spends that judgment on pricing, capital, and deals — not on assembling and reconciling the inputs to those decisions.
Integration converts the enterprise from documents and spreadsheets into a queryable system. AI is only as good as what it can query and act on. Integration is the substrate. AI is the interface and the autonomy. The value is not additive. It is multiplicative, because an agent on an integrated graph can do cross-basis attribution, whole-balance-sheet scenario reasoning, and an agentic close — things that are structurally impossible on siloed systems.
The commercial implication: the moat is not the AI layer. Models commoditize. The moat is the integrated data and computation graph underneath, which is slow, domain-heavy, and painful to build. That is also why the incumbents — Moody’s / RiskAgility, FIS / Prophet, Aon / PathWise, the Workiva-style close tools — are racing to own more of the stack rather than just their silo.
PE-backed insurers, reinsurers, and asset managers run this. More than $3M in engagements. It is the system they steer the ALM conversation from, not a pack rebuilt every quarter.
Python, FastAPI, React, TypeScript, Databricks, Delta / Parquet, dbt, Temporal, OpenAI Agents SDK, LiteLLM, LangSmith, DuckDB, Postgres, Redis, SSE, Terraform, Kubernetes.