Compliance teams currently assemble fair-lending evidence by hand — pulling exports from systems they do not own, arriving at exams with gaps they did not know existed. Avarent turns the decision records you already collect into a structured evidence packet, without requiring direct model access or storing raw applicant PII.
The problem
Lenders have deployed non-linear machine learning models faster than their internal teams can monitor them. That gap is where regulatory exposure lives.
Model risk management was built for logistic regression. Gradient-boosted trees making thousands of micro-decisions across protected-class cohorts require a different kind of monitoring — one most internal teams do not yet have.
Approval rates diverge across protected-class cohorts without triggering any internal flag. The disparity is in your data right now.
Behavioral specificity is required. "Insufficient income" is not compliant. The reason codes exist — the defensible language does not.
CSVs from the LOS, joined to bureau data, analyzed with pivot tables and undocumented methodology. Gaps surface during the exam, not before it.
Which statistical analyses are blocked by missing fields? Nobody knows until leadership asks a question the data cannot answer.
HMDA submissions and complaint databases give examiners a structured picture of your decisions before the exam starts. Avarent gives you the same picture first.
Avarent surfaces which analyses can run on your current data, calculates disparity metrics in real time, and compiles the evidence packet before the exam starts.
Analysis readiness scoring surfaces which statistical analyses can run on current data fields and which are blocked — before the team presents to leadership.
Calculates approval-rate disparity and the disparate impact ratio with plain-language labels alongside technical outputs.
Validates reason-code completeness and specificity across decline cohorts against CFPB Circular 2023-03 requirements.
Cohort context, open findings, methodology references, and limitations compiled into a format structured for MRM and fair-lending exam teams.
Evidence packet automation that meets regulatory standards
of adverse action notices in production fail the CFPB specificity standard
Circular 2023-03 · Behavioral specificity required
Readiness scoring surfaces which statistical analyses can run on your current fields — and which are blocked — before you present to leadership.
Track how approval-rate disparity shifts across origination periods before it becomes a finding.
No raw applicant PII stored. Decision-level and cohort-level records only. Built for governance, not surveillance.
| Prime cohort | DIR: 1.02 | Pass |
| Thin File cohort | DIR: 0.61 | Review recommended |
Approval-rate disparity and the disparate impact ratio calculated with plain-language labels alongside technical outputs.
The $89 million Apple/Goldman Sachs fair-lending penalty was not the result of an algorithm nobody could have caught. It was the result of a pattern that existed in the data and was not seen by the institution before a regulator saw it. The same dynamic runs in 71% of adverse action notices currently in production — the reason code exists, the required specificity does not, and the gap is visible in the decision record. Regulators have a structured view of your approval and decline data. Avarent gives you the same view, so the questions below have answers before anyone external asks them.
No. Avarent works from decision-level outputs and logs — it does not require API access to the model itself, model weights, or the underlying scoring engine.
No. Avarent produces statistical analysis and structured documentation — it is not legal review, does not make compliance determinations, and its outputs do not constitute legal advice.
Disparate impact analysis against the four-fifths rule. Proxy-risk variable flagging for compliance review. Adverse action reason-code validation against Reg B and CFPB Circular 2023-03. Drift monitoring across origination periods. Methodology documentation structured for MRM exam teams.
Request access to the evidence packet pilot. Get the structured analysis your team needs before anyone external asks for it.