We help insurers modernise models, reporting and production with Python and controlled AI. Every change is traceable, testable and subject to actuarial sign-off.
Breadth across a public market sample
Each verdict tied back to report evidence
A declared suite, not a handful of spot checks
Across 37 public stress scenarios at the declared tolerance. Equivalence is not a correctness claim.
Focused engagements spanning models, production, documentation and core actuarial delivery.
Extract spreadsheet and legacy-platform logic into traceable specifications and executable tests—strengthening governance now, with a proven baseline if migration comes later.
Turn repeatable reserving and model-review procedures into governed workflows with diagnostics, rerouting and review packs.
Produce and review SFCR, ORSA, methodology and model documentation against defined evidence, checks and sign-off gates.
Hands-on actuarial support across reporting, valuation, asset-liability management, financial modelling and model governance.
A prompt describes what you want. A harness defines what must be true: which evidence may be used, which procedure must run, which tests must pass and where human judgement remains decisive.
Analysis, code, drafts and proposed corrections accelerate the work.
Approved evidence, commands, tests, tolerances and stopping conditions govern acceptance.
Assumptions, judgement, exceptions and final acceptance stay with accountable professionals.
You do not need to change platforms to benefit. We extract the calculations, assumptions, branches and historical workarounds buried in an existing model, then test that understanding against its actual behaviour. The result improves documentation, review, change control and onboarding—and creates a proven baseline if migration follows.
Document the existing model — its calculations, assumptions, branches and historical workarounds.
Extract reproducible golden values mechanically from the model in its current environment.
Require a source-cited specification and blind prediction before implementation.
Turn formulas and dependencies into a traceable account that can be challenged independently.
Exercise boundaries, switches, negative rates and original error behaviour.
Strengthen the current model, control future changes or use the evidence as the baseline for migration.
Extraction does not certify the model. It establishes a traceable, tested account of how the source behaves within declared scenarios and tolerances. That evidence is useful whether the current platform stays or the model moves.
Each example exposes the method, the control layer and the boundary of the claim.
AI maps and specifies existing model logic; deterministic tests challenge that understanding; an actuary approves the documented basis.
22,570 comparisons · 1e-9 tolerance · 0 failures
Explore the solutionBounded evidence packs, defined procedures, claim-level support and stopping conditions turn the report into a review interface.
Illustrative comparison · simulated results
See the harness impactA model-agnostic agent layer routes plain-language requests through diagnostics, sensitivities and a reviewable reserving pack.
1:59 recorded workflow · Skill · Harness · Human review
Watch the production harnessQualified actuary · Founder, IP Consulting Group
Ivan works across life and non-life insurance, IFRS 17 implementation, ALM and actuarial modelling, combining actuarial judgement with hands-on Python, AI and software engineering.
Clients work directly with the person who frames the actuarial problem, builds the control approach and remains accountable for the result.
Choose the closest starting point. Your message will go directly to Ivan.
