AI in Evalio is a validation layer, not a decision-maker.
Evalio uses bounded AI agents as a support layer within the compensation governance workflow. Every agent is logged, bounded by policy, and overridable by the human reviewer. AI does not make pay decisions, grade decisions, or legal determinations.
Every AI interaction is logged, bounded, overridable, and human-governed.
Logged
Every bounded agent invocation is recorded with agent class, context hash, output summary, and timestamp.
Bounded
Agent scope is constrained by the Companion Policy. No agent operates outside its permitted context — enforced at the schema and policy layer, not by convention.
Overridable
Every AI-produced output can be amended or rejected by the human reviewer. Override events are also logged.
Human-governed
Every material decision — grade, pay, promotion, equity adjustment — requires a named human reviewer and produces a governed record.
Traceable
The output is traceable to what went in. Evidence basis, reviewer identity, and decision context are preserved in the Decision Room.
Four classes of bounded AI agent — each with explicit scope and explicit limits.
Evaluation Assistant
Supports structured role evidence review — surfaces contradictions and confidence signals for the human reviewer.
Boundary
Does not assign the grade. Does not expose scoring mechanics. Does not act without a human reviewer.
Calibration Insight
Surfaces pay equity patterns and range position context for annual cycle and mid-cycle review.
Boundary
Does not make the compensation decision. Does not approve adjustments. Does not compare named individuals without reviewer context.
Cycle Communication
Drafts bounded communication language for annual cycle outputs — manager talking points, employee-facing summaries.
Boundary
Does not send communications. Does not commit budget. Does not override the calibrated decision.
Equity Pattern
Identifies cohort equity patterns across grade bands for CHRO and Total Rewards review.
Boundary
Does not flag named individuals without reviewer context. Does not determine remediation. Does not produce a legal conclusion.
What Evalio AI will never do.
Hard exclusions — not policy, not convention, enforced at architecture level
- Assign a grade autonomously
- Approve a pay decision
- Issue a promotion decision without a human reviewer
- Produce a legal conclusion or compliance certification
- Expose raw evaluation scores, weights, thresholds, or protected methodology mechanics
- Operate outside the bounded policy context
- Claim AI-produced outputs are certified actuarial, legal, or regulatory outputs
- Take action on a decision without a recorded human reviewer
AI outputs use bounded, honest language — not confident assertions.
Grade outcome
"Evalio Grade [X] — indicated from the evidence provided"
Market position
"Indicative market positioning — not a certified survey output"
Equity pattern
"Equity pattern identified — reviewer applies judgment"
Confidence signal
"High / moderate / limited confidence in the evidence basis"
Contradiction
"Contradiction in evidence detected — reviewer attention required"
Bounded output
"Bounded interpretation — human decision required"
For DPA, security questionnaire, and AI governance review.
For the data protection officer
AI interactions in Evalio process role evidence and compensation context provided by the customer. No personal data is used to train models. Inputs are passed to bounded agent layers under the Companion Policy. Every invocation is logged and the log is retained in the governed records substrate under the customer's data residency posture.
For the security review team
The AI layer is an internal bounded layer — not a public LLM API exposed to end users. Outputs are schema-validated before they reach the UI. The policy contract prevents out-of-scope outputs from being surfaced. All agent invocations are correlated to a reviewer session and workspace context.
Certification posture
AI governance controls are in process for ISO 27001 and SOC 2 Type II scope. Current status: governed by architecture and policy, not yet externally certified. DPA and security questionnaire responses are available under NDA for enterprise procurement. Talk to Evalio to start this conversation.
Next steps