Specialist analytics for banking & insurance · London → Bengaluru
Technology · The Wersel Stack

A reference architecture engineered for regulated finance.

Four interlocking layers — Data, Model, AI and LLM — bound by a governance and lineage spine that runs end to end. Built ground-up for the way banks and insurers actually ship work the regulator can defend.

Reference architecture · v2026.2Audit-ready by design
Why a stack, not a tool list

Engineering principles, not a vendor bill of materials.

The Wersel Stack is the reference architecture our practitioners use to ship Data, Model, AI and LLM workloads inside a regulated bank or insurer. It is opinionated about layering, lineage and governance — and deliberately neutral about specific products, because the right vendor in your environment is rarely the same as the right vendor in someone else’s.

What it does fix is the shape of the work: which capabilities sit at which layer, how the governance spine cuts through every tier, and where the seams are that audit, model risk and the supervisor will inspect. Use it as a planning canvas for your next regulatory milestone, or as a target architecture for an estate that has grown faster than its operating model.

The Wersel Stack

Four layers. One governance spine. Built to be audited.

Each layer ships with its own engineering deliverables and its own model-risk and lineage obligations. The spine ensures every input, every model and every output traces back to a controlled source — at the cadence your regulator now expects.

Operating plane All controls operational
Operating Plane — engagement & ops console
A single pane of glass for every engagement: SLAs, model inventory, pipeline health, evaluation harness results and the auditable evidence trail behind each delivery.
Engagement healthModel inventoryLineage telemetryEvaluation harnessEvidence vault
Layer 01
Day 1
lineage on by default
CDE
tier as a live attribute

Data Layer — the substrate every model depends on

The ingestion, contract, warehouse and lineage primitives that turn source-system chaos into a versioned, governed, model-ready surface.

Source contracts & ingestion

Typed contracts at the boundary, replayable ingestion, schema-evolution policy. The source-system surprises absorbed once, not every quarter.

Contract-firstReplayableSchema-evolution
Lakehouse warehouse

A versioned analytical store with time-travel, branching and reproducible builds. The audit can re-run last quarter’s report against last quarter’s data.

Time-travelBranchingReproducible
Sustainable BCBS 239 lineage

Captured automatically at ingestion and transformation; CDE status carried as a live attribute, not a spreadsheet label.

Automatic captureCDE-awareAudit-ready
Privacy & sovereignty controls

Tokenisation, residency-aware routing and lawful-basis tags — engineered into the pipeline rather than bolted on at the report.

Cross-borderLawful basisTokenised
Layer 02
IFRS 9
ECL, IRB & Standardised
IFRS 17
CSM, RA & onerous-contract

Model Layer — quantitative craft, auditable on first review

Where the classical risk, actuarial and pricing models live — engineered with the validation harness, challenger review and documentation the regulator will ask for.

Credit risk & ECL models

PD / LGD / EAD, IFRS 9 stage transitions, macro-overlays and reverse-stress narratives — built with continuous lineage and explainable methodology bridges.

IFRS 9IRBMacro-overlays
Actuarial & capital models

IFRS 17 measurement, Solvency II / UK reform, BPA pricing under PRA cashflow tests, Matching Adjustment ALM monitors.

IFRS 17BPAMA monitor
Pricing & decisioning engines

Application scorecards, affordability matrices and personal-lines pricing with Consumer Duty fair-value evidence attached at the decision.

ScorecardsAffordabilityFair value
Validation harness & challenger review

Pre-release validation, ongoing monitoring, documented challenger review and methodology bridge — generated alongside the model, not added before the meeting.

Pre-releaseOngoingChallenger
Layer 03
SS1/23
tiered model-risk regime
EU AI Act
risk-tier aware

AI Layer — applied ML inside the model-risk perimeter

Production ML for fraud, claims, AML, marketing and operations — deployed under the same model-risk discipline as the classical credit and actuarial estate.

Feature store & training pipelines

Versioned features, reproducible training runs, point-in-time correctness — so the model that shipped is the model the validation report describes.

Point-in-timeVersionedReproducible
AML, fraud & claims models

Real-time scoring with calibrated thresholds, false-positive demographic monitoring and investigator-loop feedback wired into retraining.

Real-timeCalibratedLooped
Bias, drift & fairness monitoring

Continuous evaluation against documented thresholds — bias proxies, drift, refusal behaviour and Consumer Duty signals all under live control.

Bias proxiesDriftLive thresholds
Model-risk inventory & lifecycle

Every model — vendor or in-house — carried as an inventory record with owner, purpose, tier, validation cadence and retirement plan.

InventoryTieredLifecycle
Layer 04
GTG-1002
aware threat modelling
HITL
tier-based escalation

LLM Layer — generative AI under regulated guardrails

Retrieval pipelines, agentic workflows and copilot surfaces — engineered with prompt/response lineage, evaluation harnesses and human-in-the-loop patterns the supervisor can inspect.

Retrieval & grounding pipelines

Document, policy and customer-data retrieval with provenance preserved — grounding the model on a controlled corpus rather than its training set.

ProvenanceControlled corpusHybrid retrieval
Agentic workflows & tool use

Bounded agents that decompose work into auditable steps, with tool allowlists, blast-radius limits and step-level evaluations.

BoundedAllowlistsStep-evaluated
Prompt & response lineage

Context window, tool calls, model outputs and downstream actions captured per interaction — the audit trail generative AI was missing.

Per-interactionTool-callDownstream
Eval harness & HITL escalation

Hallucination, accuracy, refusal and Consumer-Duty evaluations on live traffic, with tier-based escalation to a human reviewer when confidence drops.

Live evalHallucinationTiered HITL

Governance & Lineage spine

A continuous control fabric that cuts vertically through Data, Model, AI and LLM — so every change has a controlled cause and every output a defensible source.

01
Lineage captured at ingestion, transformation, inference and consumptionAutomatic — never a hand-drawn diagram. Survives reorganisations and vendor swaps.
02
Model inventory across classical, ML and generative tiersOne register; tier-appropriate validation cadence; ownership baked into the record.
03
Evaluation harnesses on live trafficAccuracy, drift, bias proxies and refusal behaviour under continuous threshold monitoring.
04
Evidence pack generated, not assembledBCBS 239, SS1/23, Consumer Duty and EU AI Act artefacts emitted as a side effect of the build.
Color key:DataModelAILLMGovernance spineOperating plane
Each card is a deliverable, not a product. Engineered to be lifted into your stack and operated by your team after handover.
Domain overlays

How the stack lands in banking and insurance.

The same four layers, the same spine — instantiated against the regulatory reality of each vertical. These are the targets our embedded teams ship against.

Banking

For Tier 1, challenger and private-credit balance sheets.

Engineered for the credit lifecycle and the supervisory dialogue around it — from BNPL perimeter change to SWES 2 readiness.

  • DataSustainable BCBS 239 lineage, post-BNPL bureau attributes and CRR / Standardised data marts.
  • ModelIFRS 9 ECL with macro-overlay calibration; IRB scorecards; affordability matrices re-fit on the new signal.
  • AIReal-time fraud, agentic KYC / pKYC and AML triage — deployed under SS1/23 model-risk discipline.
  • LLMCopilots for credit memos, SAR drafting and complaint triage with full prompt/response lineage.
Insurance

For life, general and reinsurance capital strategies.

Engineered for the actuarial close, the BPA pipeline and the Consumer Duty evidence chain — with M&A integration patterns built in.

  • DataIFRS 17 cohort-aware data substrate, BPA inforce ingestion and MA eligible-asset registers.
  • ModelIFRS 17 measurement, Solvency II / UK reform, BPA pricing under PRA cashflow tests, MA ALM monitor.
  • AIUnderwriting scoring, fraud and leakage detection, parametric instant-payout decisioning.
  • LLMAdjuster co-pilots, Consumer Duty explanation layers and policy-comparison agents with HITL escalation.
Auto-generated stack spec

Every engagement ships its own machine-readable architecture.

Generated bywersel stack render --target audit --format mermaid
// wersel-stack.md · commit: a7c4f3b · regenerated 2026-05-22
graph TD
  SRC[Source systems · core / policy admin / cards / claims] --> DL[Data Layer]
  DL --> ML[Model Layer · IFRS 9 · IFRS 17 · pricing]
  DL --> AIL[AI Layer · fraud · claims · AML]
  DL --> LLM[LLM Layer · copilots · agents · retrieval]
  ML --> DEC[Decisions · capital · pricing · reserving]
  AIL --> DEC
  LLM --> DEC
  GOV[Governance & Lineage spine] -.-> DL
  GOV -.-> ML
  GOV -.-> AIL
  GOV -.-> LLM
  OP[Operating plane] -.governs.-> GOV
  DEC --> EV[Evidence vault · BCBS 239 · SS1/23 · Consumer Duty · EU AI Act]

Engineer your stack against the next supervisory cycle.

Talk to our practitioners about adopting the Wersel Stack as a target architecture — or commissioning a current-state assessment of your Data, Model, AI and LLM estate.

Or write to us directly: hello@wersel.io