AiNOS One
The insurance ontology
One evidence-linked model of customers, policies, rules and decisions, resolvable to any date. Build it once; every product stands on it.
One data infrastructure for the whole insurer.
Fragmented, legacy data, orchestrated into one scalable model of the business: the insurance ontology. Built once, and every product stands on it.
None of the seven domains becomes AI-native on top of a legacy core nobody can read. The highest-value work is turning decades of administration data, undocumented codes and institutional knowledge into a structured, evidence-linked picture of the business that both people and AI can query. Every workflow on the map is built on this layer.
AiNOS Decode
Legacy data meaning recovery
Thirty years of mainframe records, turned into data an AI can read, with every recovered meaning tied to its evidence. The way in to One.
AiNOS Recall
Institutional memory
Anyone asks in plain words. Recall queries every system through One and answers with the records and the query behind it. What is not on record is marked, never guessed.
Why this is where the value is
- Insurers run on administration systems that are older than most of the people operating them. Field meanings are lost, codes are undocumented, and the people who remember are retiring.
- A language model given raw tables guesses. Given an ontology with every entity traced to its source, it reasons on facts and can show its work.
- The same structured layer serves underwriting, compliance, persistency and finance. Build it once; every domain reuses it.
Seven layers, built once
Every product on the map reuses them. Building them for one job builds most of them for the next.
Legacy data meaning recovery
Mainframe and legacy fields are understood by a few people, documented nowhere, and migrated by guesswork.
Field meaning is recovered from code, data patterns and expert interviews, with each recovered meaning tied to its evidence and a confidence level.
- Copybooks, schemas, job logic and sample records are read together to reconstruct what each field actually holds.
- Undocumented codes are resolved against transaction history and product documents rather than assumed.
- Exception cases, such as a zero premium on an in-force policy, are surfaced as questions for an expert rather than silently transformed.
Insurance ontology
Each system has its own notion of policy, customer, agent, coverage and event, and reconciliation happens in spreadsheets.
One compiled model of the business: policies, parties, coverages, rules, events and decisions, with every instance resolved across systems and traced to its source row.
- Entities are resolved across administration, commission, claims and service systems into one identity with a full timeline.
- The model is versioned, so a question asked about last year is answered with last year's structure.
- Facts and inferences are stored separately, so an AI's conclusion never overwrites a source.
Versioned rules and law
Laws, regulations, internal rules and guidelines are documents in folders, and the version in force on a given date is reconstructed by hand.
Every source is captured from its original, compiled into controlling states and frozen per version, so any rule can be resolved as of any date.
- Statutes, regulations, supervisory guidance, internal rules and carrier guidelines are all held the same way.
- A contract from 2021 is judged by the 2021 rule. This is what survives an examination or a court.
- A rule change is mapped to the workflows and decisions it affects before it takes effect.
Evidence linking
A number in a report cannot be traced to the row and the document that produced it without an analyst's week.
Every fact, flag and recommendation carries a link to the data row, the document page or the clause it came from.
- Documents are preserved as originals and are read-only to any AI.
- Each derived fact records what it was derived from, so a reviewer can open the source in one step.
- Reproducibility is the standard: the same inputs produce the same explanation.
Migration validation
Old and new systems are reconciled on totals, and the exceptions are negotiated at go-live.
Old and new are reconciled record by record, with each difference explained and classified before anyone signs off.
- Validation runs continuously during the migration, not once at the end.
- Explained differences become test cases for the next product line.
- The reconciled result is itself evidence-linked, so the audit trail of the migration is complete.
Institutional knowledge capture
Manuals, contracts and the judgment of experienced staff are scattered, outdated or about to walk out the door.
Manuals and contracts are compiled into queryable knowledge, and expert judgment is captured as rules and worked cases that can be tested.
- An expert's decisions on real cases become the examples a system is evaluated against.
- Knowledge carries its source and its date, so an outdated instruction is visible as outdated.
- The same knowledge serves training, operations and AI alike.
AI governance
AI pilots run without a shared standard for what the model may see, do and claim, and nobody can audit a decision later.
Every AI workflow runs under the same controls: eligible evidence, applicable rules, allowed tools, validated outputs and human authorization, with a trace of each run.
- Readiness is graded, from research, to controlled pilot with human review, to production, and nothing moves up without evidence.
- Evaluation uses real work cases and holdout sets, and failures are refined before scale, not after.
- Quality, latency and cost are measured in operation, so the system is maintained as requirements change.
What stands on it first
The three products on the policy each draw on specific layers.
- AiNOS Bind
- Carrier guidelines compiled into versioned, testable rules.
- Policy, party and risk entities resolved across legacy administration systems.
- Document evidence linked to source rows, so a figure in the packet traces to its page.
- AiNOS Keep
- Policy, agent and customer histories resolved into one timeline from administration, commission and service systems.
- Commission scheme and product rules compiled so their effect on lapse can be tested.
- Outcome records kept so each intervention becomes evidence for the next.
- AiNOS Guard
- A versioned corpus of law and internal rules, resolvable to any past date.
- Contracts, agents and customers resolved across sales, administration and complaint systems.
- Every flag tied to the data row and the clause that produced it, so the explanation is reproducible.