The Agentic Bank

Compensation Benchmarking Agent

⬡ Plumb Builds market comp benchmarks and flags pay-equity outliers.
◆ Assistive Specialist

Ingests survey data, internal pay distributions and offer history and models defensible comp ranges per role and market, continuously across the role-family sweep. Surfaces pay-equity outliers as advisory output; the comp-committee agent owns every pay decision under the fairness judge.

Memory

Working The role family + market under analysis.
Episodic Prior cycle benchmarks and approved exceptions.
Semantic Survey methodologies, geo differentials, leveling framework.
Store File-based memory tool + comp data warehouse

Orchestration

MCP

Harness · Managed Agents … sandboxed code execution for benchmark modelling.

Tools

{ } Comp survey feeds (e.g. market data) API { } Internal HRIS pay data API ›_ Benchmark modelling sandbox Code exec

Evals & guardrails

  • Pay-equity / fairness analysis across protected groups on every benchmark.
  • The comp-committee agent signs off all range changes … this agent is advisory only.
  • Source-citation discipline: every range traces to its survey inputs.

Frontier edge

  • World-model simulation: models the attrition and offer-acceptance impact of a proposed range before the comp-committee agent commits it, so the trade-off is quantified, not guessed.
  • Causal pay-gap reasoning: separates legitimate level/geo drivers from unexplained residual gaps instead of flagging raw correlations.
  • Confidential compute: individual salary records are reasoned over inside a secure enclave; the committee agent sees ranges and outliers, never named pay in clear.

In numbers

continuous
Benchmark refresh
tracked per cycle
Pay-equity outliers surfaced

Handoffs

Across ⇢ Finance / Treasury for comp-budget planning inputs

More on the Human Resources desk